system

Abstract

This enables fair and effective speech recognition for speakers from various regions. [Solution] The system includes means for using a terminal having the function of collecting voice data from speakers with diverse utterances; means for securely transmitting and storing the collected voice data; means for training a speech recognition model using the stored voice data; means for providing and updating the trained speech recognition model to each terminal; means for collecting user feedback on the accuracy of speech recognition; and means for improving the speech recognition model based on the collected feedback.

Description

Technical Field

[0005]

[0001] The technology of the present disclosure relates to a system.

Background Art

[0002] Patent Document 1 discloses a persona chatbot control method performed by at least one processor, the method including: receiving a user utterance; adding the user utterance to a prompt including an instruction sentence related to an explanation of a character of the chatbot; encoding the prompt; and inputting the encoded prompt into a language model to generate a chatbot utterance as a response to the user utterance.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] There is a problem that the speech recognition accuracy of voices with different speech characteristics depending on regions, especially Japanese voices including dialects, is low. This makes it difficult to communicate with speakers using dialects and hinders the spread of speech recognition technology. There is a demand for a system that solves this problem and enables fair and effective speech recognition for speakers in various regions.

Means for Solving the Problems

[0005] This invention provides a system that uses terminals installed throughout Japan to collect voice data from speakers with diverse speech patterns, and securely transmits and stores this data. Using the stored voice data, it trains a speech recognition model capable of identifying region-specific speech features and provides this model to each terminal. Furthermore, it includes a means to improve dialect recognition accuracy as a cloud-based speech recognition system by collecting user feedback on speech recognition accuracy and improving the speech recognition model based on this information.

[0006] A "terminal" is a device that collects voice data from speakers with diverse speech patterns, securely transmits it, and receives updated speech recognition models.

[0007] "Audio data" refers to information recorded in digital format from the sounds a speaker makes, and is used for training and evaluating speech recognition models.

[0008] A "server" is a central management system that stores collected audio data and uses it to train and update speech recognition models.

[0009] A "speech recognition model" is an algorithm trained to analyze speech data and extract text data and speech features.

[0010] "Feedback" refers to information that users provide to evaluate the results of speech recognition and to help improve the system. [Brief explanation of the drawing]

[0011] [Figure 1] This is a conceptual diagram showing an example of the configuration of a data processing system according to the first embodiment. [Figure 2] This is a conceptual diagram showing an example of the essential functions of a data processing device and a smart device according to the first embodiment. [Figure 3] This is a conceptual diagram showing an example of the configuration of a data processing system according to the second embodiment. [Figure 4]This is a conceptual diagram showing an example of the main functions of a data processing device and smart glasses according to the second embodiment. [Figure 5] This is a conceptual diagram showing an example of the configuration of a data processing system according to the third embodiment. [Figure 6] This is a conceptual diagram showing an example of the main functions of a data processing device and a headset-type terminal according to the third embodiment. [Figure 7] This is a conceptual diagram showing an example of the configuration of a data processing system according to the fourth embodiment. [Figure 8] This is a conceptual diagram showing an example of the main functions of a data processing device and a robot according to the fourth embodiment. [Figure 9] This shows an emotion map where multiple emotions are mapped. [Figure 10] This shows an emotion map where multiple emotions are mapped. [Figure 11] This is a sequence diagram showing the processing flow of the data processing system in Example 1. [Figure 12] This is a sequence diagram showing the processing flow of the data processing system in Application Example 1. [Figure 13] This is a sequence diagram showing the processing flow of the data processing system in Example 2, which incorporates an emotion engine. [Figure 14] This is a sequence diagram showing the processing flow of the data processing system in Application Example 2, which combines an emotion engine. [Modes for carrying out the invention]

[0012] Hereinafter, an example of an embodiment of the system relating to the technology of this disclosure will be described with reference to the attached drawings.

[0013] First, let's explain the terminology used in the following explanation.

[0014] In the following embodiments, the numbered processor (hereinafter simply referred to as "processor") may be a single arithmetic unit or a combination of multiple arithmetic units. Also, the processor may be a single type of arithmetic unit or a combination of multiple types of arithmetic units. Examples of arithmetic units include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), an APU (Accelerated Processing Unit), and the like.

[0015] In the following embodiments, the numbered RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a work memory by the processor.

[0016] In the following embodiments, the numbered storage is one or more non-volatile storage devices that store various programs, various parameters, and the like. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), or magnetic tapes, and the like.

[0017] In the following embodiments, the numbered communication I / F (Interface) is an interface including a communication processor, an antenna, and the like. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0018] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." That is, "A and / or B" means that it may be A alone, or B alone, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" applies when expressing three or more things linked by "and / or."

[0019] [First Embodiment]

[0020] Figure 1 shows an example of the configuration of the data processing system 10 according to the first embodiment.

[0021] As shown in Figure 1, the data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0022] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0023] The smart device 14 comprises a computer 36, a reception device 38, an output device 40, a camera 42, and a communication interface 44. The computer 36 comprises a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The reception device 38, output device 40, and camera 42 are also connected to the bus 52.

[0024] The reception device 38 is equipped with a touch panel 38A and a microphone 38B, etc., and receives user input. The touch panel 38A receives user input by detecting contact with an object (e.g., a pen or finger). The microphone 38B receives user input by detecting the user's voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0025] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form perceptible to the user 20 (e.g., audio and / or text). The display 40A displays visible information such as text and images according to instructions from the processor 46. The speaker 40B outputs audio according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0026] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various types of information between processor 46 and processor 28 via network 54.

[0027] Figure 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0028] As shown in Figure 2, in the data processing device 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" related to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0029] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0030] In the smart device 14, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The reception output program 60 is used in conjunction with a specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0031] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the smart device 14 as the "terminal".

[0032] This invention provides a method for collecting and processing voice data using terminals installed throughout Japan, as a voice recognition system that includes regionally specific dialects. This system consists of three components: a user, a terminal, and a server.

[0033] First, the user engages in a casual conversation with a nearby device. After obtaining necessary consent, the audio data collected during the conversation is recorded and segmented in real time by the device. Audio features are extracted, and provisional labels are assigned according to the type of dialect. The audio data undergoes noise reduction processing to remove unwanted noise, and the processed data is then encrypted and stored.

[0034] Next, the server receives the audio data transmitted from the terminal. The server analyzes this data and uses it as training data for a speech recognition model. Using supervised learning techniques, the server generates and updates the speech recognition model based on the audio data for each region. This speech recognition model is adjusted to have the ability to identify region-specific speech features.

[0035] Trained speech recognition models are automatically distributed from the server to each terminal. This ensures that terminals always perform speech recognition based on the latest model, improving recognition accuracy in each region. Furthermore, through user interaction, terminals provide the server with speech recognition results and user feedback. The server then retrains the model based on the collected feedback to further improve accuracy.

[0036] As a concrete example, a user speaking the local dialect engages in everyday conversation using a terminal installed in a public facility in a certain area. This audio is collected in real time and sent to a server as data with unique regional vocal characteristics. The server learns from this data and, for example, updates its speech recognition algorithm to one that is specialized for a particular dialect. The latest model is deployed to the terminal, enabling even higher accuracy the next time speech recognition is performed on that terminal.

[0037] In this way, this system can support diverse regional dialects and eliminate regional disparities in speech recognition technology.

[0038] The following describes the processing flow.

[0039] Step 1:

[0040] Users consent to the collection of voice data and engage in everyday conversations through the device. They can engage in natural conversations, including those using dialects, in public facilities and home environments where the device is installed.

[0041] Step 2:

[0042] The device begins recording the user's conversation in real time. The audio data is segmented by detecting breaks in speech, and each segment is assigned a temporary label.

[0043] Step 3:

[0044] The device then applies preprocessing, such as noise reduction, to the segmented audio data. This preprocessing improves sound quality and prepares data suitable for training the speech recognition model.

[0045] Step 4:

[0046] The device encrypts the processed voice data and sends it to the server via a secure communication protocol. The data is protected with the utmost consideration for user privacy.

[0047] Step 5:

[0048] The server saves the received audio data to a database. During saving, dialect information and speech characteristic labels associated with the audio data are also recorded.

[0049] Step 6:

[0050] The server uses stored audio data to train a speech recognition model. Using supervised learning, the speech recognition model learns region-specific speech features and improves the accuracy of speech recognition.

[0051] Step 7:

[0052] The server deploys the newly trained speech recognition model to each terminal. The terminal downloads this new model and uses it for speech recognition processing.

[0053] Step 8:

[0054] Users can provide feedback on the speech recognition results from their devices. This user feedback is valuable information for improving the speech recognition results.

[0055] Step 9:

[0056] The device periodically sends the collected feedback to the server. This feedback is used to consider whether to retrain the model.

[0057] Step 10:

[0058] The server analyzes the feedback data and adjusts the parameters of the speech recognition model as needed. This further improves speech recognition accuracy in each region and enhances the overall system performance.

[0059] (Example 1)

[0060] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server," and the smart device 14 will be referred to as the "terminal."

[0061] In speech recognition technology, there is a need to achieve highly accurate speech recognition while being able to handle diverse, region-specific speech characteristics. However, conventional speech recognition systems have the problem of failing to adequately handle regional pronunciations and dialects, resulting in reduced recognition accuracy. Furthermore, there is a lack of mechanisms to effectively incorporate user feedback and update the models, making continuous model improvement difficult.

[0062] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[0063] In this invention, the server includes: an information device means having the function of acquiring voice information from users with diverse voices; means for securely transmitting and storing the acquired voice information; and means for collecting user feedback regarding the accuracy of speech recognition. This enables rapid and accurate learning of various region-specific voice characteristics, and improves and continuously refines the accuracy of the speech recognition model.

[0064] "Diverse speech" refers to the vocalizations of multiple speakers with different pronunciations, dialects, and accents.

[0065] A "user" refers to an individual who interacts with the system and provides voice information through this system.

[0066] "Voice information" refers to data that is acquired and stored in digital format from speech uttered by users.

[0067] An "information device" is a device used to acquire, process, and transmit voice information, and includes components such as microphones and processors.

[0068] A "speech recognition model" is an algorithm or system built to analyze speech information and convert it into text or specific commands.

[0069] "Feedback" refers to opinions from users regarding the results of speech recognition, their impressions, and suggestions for improvement, which are used to improve the system.

[0070] "Supervised learning" is a method of training machine learning models using labeled datasets, which are then used to enable the models to make accurate predictions and classifications.

[0071] This section describes an embodiment of this system. This system aims to address diverse, region-specific speech characteristics using speech recognition technology. The system primarily consists of three elements: the user, the terminal, and the server.

[0072] Users transmit audio through the installed terminals. For example, when a user talks about a local landmark, the audio is collected by the terminal. This audio information is naturally gathered from the user's everyday conversations.

[0073] The device records the user's voice in real time and segments it as digital audio data. This device is equipped with a microphone and signal processing unit for extracting audio features. Features are extracted from the recorded audio data using techniques such as Mel-frequency cepstrum coefficients (MFCC). Noise reduction is also performed using digital signal processing to obtain clear audio data. Initial labels are then assigned based on the acoustic characteristics. The audio data is then encrypted using the AES (Advanced Encryption Standard) protocol and stored securely.

[0074] The server receives data transmitted from terminals and uses it to train speech recognition models. It utilizes guided learning techniques such as deep learning to learn the speech characteristics of specific regions and improve recognition accuracy. The trained speech recognition models are pushed to each terminal, allowing them to efficiently perform recognition tasks in their most up-to-date state. The server also collects user feedback and continuously improves the models based on it.

[0075] As a concrete example, when a user uses a device to speak in the local dialect, the audio is collected and analyzed by a server. Through this process, the speech recognition system becomes familiar with the local dialect and can improve its recognition accuracy.

[0076] An example of a prompt in a generative AI model is, "Update the speech recognition model using speech data collected from a specific region for a speech recognition system that identifies regional dialects."

[0077] In this way, this system can recognize voices from diverse regions with high accuracy, contributing to the elimination of regional disparities in speech recognition.

[0078] The flow of the specific processing in Example 1 will be explained using Figure 11.

[0079] Step 1:

[0080] The user begins a casual conversation into a terminal installed in their local area. The input information is the user's speech, which is captured as audio data by the terminal's high-sensitivity microphone. A specific example of this operation would be the user uttering something like, "I'm going to talk about the local festival." The output is raw audio data.

[0081] Step 2:

[0082] The terminal segments the acquired audio data in real time. The input is audio data from the user, and the digital signal processing unit divides this data so that it can be easily managed. Specifically, it divides the audio into segments of regular time intervals and saves each segment as a separate file. The output is a collection of segmented audio files.

[0083] Step 3:

[0084] The device extracts features from audio data. The input is a segmented audio file, from which acoustic features are extracted using techniques such as Mel-frequency cepstrum coefficients (MFCCs). Specifically, it analyzes the frequency and temporal characteristics of each audio segment and converts them into numerical data. The output is a list of the extracted acoustic features.

[0085] Step 4:

[0086] The device applies noise reduction processing to the feature data. The input is feature data, and it performs specific operations to remove noise using digital signal processing. This improves the accuracy of speech recognition. The output is clean speech data with reduced noise.

[0087] Step 5:

[0088] The device assigns a temporary label to the noise-reduced audio data, encrypts it, and saves it. The input is noise-reduced audio data. Specifically, a temporary label appropriate to the audio content is assigned, the data is encrypted using the AES protocol, and then saved. The output is labeled and encrypted audio data.

[0089] Step 6:

[0090] The terminal sends encrypted audio data to the server. The input is labeled encrypted data. The terminal uses the TLS protocol to perform the specific action of securely communicating and sending the data to the server. The output is the audio data securely transferred to the server.

[0091] Step 7:

[0092] The server trains a speech recognition model using the received audio data. The input is audio data sent from the terminal, and the model is optimized using deep learning techniques. Specifically, it learns new acoustic features and updates parameters to improve accuracy. The output is the new model parameters.

[0093] Step 8:

[0094] The server provides each terminal with a trained speech recognition model. The input is the updated model parameters. The server performs a push notification to each terminal to keep the model up-to-date. The output is each terminal with the latest speech recognition model installed.

[0095] Step 9:

[0096] The device interacts with the user using a new speech recognition model. The input is new speech data from the user, which is then processed by the new model. The output is the improved speech recognition result.

[0097] Step 10:

[0098] The terminal sends user feedback data to the server along with the recognition results. The input consists of the speech recognition results and feedback information, which are sent to the server as a specific action. The output is feedback information sent to the server, which is used as material for further model improvement.

[0099] (Application Example 1)

[0100] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server," and the smart device 14 will be referred to as the "terminal."

[0101] Speech recognition systems that include regional dialects need to be able to recognize speech with high accuracy even in everyday conversations using those dialects, thereby promoting their use in physical stores and improving customer service. However, improving recognition accuracy to handle diverse dialects remains a challenge. In particular, there is a need to build an environment where speech data can be collected and models updated in real time.

[0102] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[0103] In this invention, the server includes means for using a device that has the function of collecting speech data from speakers with diverse speech patterns, means for providing and updating trained speech recognition algorithms to each device, and means for returning information to an output device using the speech recognition results. This makes it possible to efficiently recognize speech including regional dialects and to immediately provide the results back to the user.

[0104] A "speaker with diverse speech patterns" refers to a speaker who has unique dialects or ways of speaking that differ depending on the region and culture.

[0105] "Audio data" refers to data that records a speaker's utterances in digital format.

[0106] "Device" refers to a device that includes hardware or software used to collect and process audio data.

[0107] A "secure method" refers to a method that uses encryption and authentication techniques to protect audio data from unauthorized access and tampering.

[0108] A "speech recognition algorithm" is a computational method or process for generating text from speech data.

[0109] "Response" refers to the feedback information provided by the user based on the results of speech recognition.

[0110] "Information" refers to the content and data output based on the results obtained through speech recognition.

[0111] This invention aims to construct a speech recognition system that can handle a variety of dialects. The system is primarily implemented through three elements: a server, a terminal, and a user.

[0112] The server performs a series of processes, from collecting audio data to training and updating recognition algorithms. Specifically, it processes the collected audio data in a secure manner and trains speech recognition algorithms. The audio data is used to generate region-specific models.

[0113] The terminal collects voice data through interaction with the user and transmits it to the server. Appropriate hardware and software are used for voice data collection, such as smart devices with microphones or personal computers. The voice data transmitted from the terminal to the server is securely transmitted via the internet to enable real-time speech recognition.

[0114] The user communicates with the system by speaking, including in dialect, using a terminal. The terminal collects the user's speech, encrypts it, and sends it to the server. The server trains a speech recognition model, returns the trained model to the terminal, and improves the accuracy of speech recognition.

[0115] As a concrete example, imagine a scenario where a user asks for tourist information in their local dialect at a terminal installed in a commercial facility in a certain region. In this case, the system receives the user's speech, performs voice recognition, and immediately returns appropriate tourist information to the user's smartphone.

[0116] One example of how a generative AI model can be used is a prompt such as, "Please tell me tourist information in the Okinawan dialect." This input allows the server to generate an audio guide and provide the user with the most relevant information.

[0117] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[0118] Step 1:

[0119] The device acquires voice input from the user via a microphone. The acquired voice is recorded as digital audio data. This data undergoes noise reduction processing to reduce background noise. The noise-reduced audio data then becomes the input for the next step.

[0120] Step 2:

[0121] The terminal segments the denoised audio data and extracts the audio features contained in each segment. An acoustic analysis algorithm is used for this process. Based on the audio features, a provisional label is assigned to the audio data. The labeled data is then transmitted to the server via a secure communication protocol.

[0122] Step 3:

[0123] The server receives audio data transmitted from the terminal and uses it to train the speech recognition algorithm. Supervised learning is applied to the training to improve the ability to identify diverse regional speech features. The output of this step is the updated speech recognition algorithm.

[0124] Step 4:

[0125] The server distributes the trained speech recognition algorithm to the terminal. The terminal incorporates the distributed algorithm into its processing mechanism and prepares for the next speech recognition. This update enables more accurate speech recognition.

[0126] Step 5:

[0127] The user accesses the device again and performs voice input. The device utilizes a newly trained speech recognition algorithm to recognize the user's voice. Based on the recognition result, it returns appropriate information to the user. This output is provided to the user as either voice or text.

[0128] Step 6:

[0129] The device collects user feedback on recognition accuracy and sends it to the server. The server uses this feedback to further improve the model. It analyzes the collected feedback data and updates the algorithm as needed.

[0130] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[0131] This invention provides a system for collecting speech data from speakers with diverse speech patterns and performing speech recognition and emotion recognition. The system has the function of analyzing user speech and improving the user experience. It consists of three main components: the user, the terminal, and the server.

[0132] First, the user begins a conversation into a terminal installed in a public facility or home. Once the conversation begins, the terminal records the audio in real time and uses it for speech recognition and emotion recognition. Here, the terminal divides the audio data into segments, and each segment is labeled based on its acoustic characteristics.

[0133] The device further uses an emotion engine to identify emotions from the user's voice. This emotion recognition improves accuracy not only by using acoustic features but also by combining it with facial expression data from the camera as needed. The results of emotion recognition and voice recognition are integrated, making it possible to understand the user's state in detail.

[0134] Voice and emotion data are securely transmitted to the server via encrypted communication. The server uses this data to train speech recognition and emotion recognition models. In particular, improvements in recognition accuracy are made to accommodate regional dialects and speech patterns.

[0135] Trained speech recognition and emotion recognition models are deployed from the server to each terminal. This allows the terminal to interact with the user using the latest speech and emotion recognition models.

[0136] As a concrete example, a user speaking the Kyoto dialect converses using a terminal installed in a public facility. The terminal collects and segments voice data in real time, and in the background, an emotion engine analyzes the user's emotions from the tone and speed of their voice. For example, if an emotion expressing joy is detected, that information is sent to the server along with the utterance and used to improve the recognition model. Ultimately, the recognition rate improves, and the quality of the user experience is enhanced.

[0137] In this way, this system can bridge the gap in voice and emotion recognition between regions and improve various user experiences.

[0138] The following describes the processing flow.

[0139] Step 1:

[0140] The user initiates a conversation with the installed terminal. The terminal obtains permission to begin recording immediately upon the start of the conversation.

[0141] Step 2:

[0142] The device divides the recorded audio into segments in real time. The segmented audio is then tentatively labeled based on specific speech intervals.

[0143] Step 3:

[0144] The device extracts acoustic features from the audio and uses an emotion engine to identify the user's emotions. In this process, it captures not only audio but also facial expression data with a camera and includes it in the analysis.

[0145] Step 4:

[0146] The device encrypts both voice and emotional data and transmits them to the server via a secure network. This ensures that the data is managed in a privacy-protected manner.

[0147] Step 5:

[0148] The server stores the received audio data and emotion data in a database. During storage, it adds relevant information such as the speech label and emotion identification result to each data.

[0149] Step 6:

[0150] The server uses the collected data to train speech recognition and emotion recognition models. It aims to improve recognition accuracy by learning region-specific speech patterns and emotional expressions.

[0151] Step 7:

[0152] The server deploys newly trained speech recognition and emotion recognition models to each terminal. The terminal receives this update and makes the latest recognition algorithms available.

[0153] Step 8:

[0154] Users can provide feedback on the speech recognition and emotion recognition results provided by the device, including evaluations and suggestions for improvement.

[0155] Step 9:

[0156] The device collects user feedback and periodically sends it to the server. This feedback is used to further improve the model.

[0157] Step 10:

[0158] The server readjusts the speech recognition and emotion recognition models based on feedback and retrains them as needed. This improves the accuracy of the recognition system and user satisfaction.

[0159] (Example 2)

[0160] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server" and the smart device 14 as the "terminal".

[0161] Conventional speech recognition systems struggle to accurately recognize diverse speech characteristics and emotions, resulting in a reduced user experience. Furthermore, there is a need to improve recognition accuracy to accommodate regionally specific dialects and speech patterns. This invention aims to solve these problems and provide a richer user experience by improving the accuracy of speech and emotion recognition.

[0162] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[0163] In this invention, the server includes means for using a device that has the function of collecting speech information from speech sources with diverse utterances, means for securely transmitting and storing the collected speech information, and means for training a virtual model for speech recognition and emotion recognition using the stored speech information. This makes it possible to develop a model that can effectively recognize diverse speech characteristics and is compatible with regionally specific dialects and speech patterns.

[0164] "Diverse speech" refers to speech that exhibits a variety of different phonetic patterns and linguistic characteristics.

[0165] A "sound source" is a subject that generates sound, and the concept includes sounds emitted by humans and machines.

[0166] "Audio information" refers to audio data and its characteristics obtained from an audio source, and includes data such as the waveform and frequency components of the audio.

[0167] "Device" refers to an interface equipped with hardware and software that has the function of collecting, processing, and transmitting audio information.

[0168] "Secure transmission" refers to methods for safely transferring voice information without disclosing it to third parties, and includes communication using encryption technology.

[0169] "Storage" refers to the act of securely saving acquired audio information in a database or storage device so that it can be used later.

[0170] A "virtual model" is a set of algorithms and their parameters built on audio information, and is a computerized simulation that enables speech and emotion recognition.

[0171] "Supervised learning" is a technique for training a virtual model using known input-output pairs to improve the accuracy of predictions on new data.

[0172] "Evaluation" refers to the act of collecting user feedback and quality information regarding the results of speech recognition and emotion recognition.

[0173] "Improvement" is the process of enhancing the accuracy and functionality of a virtual model based on the collected evaluation data.

[0174] This invention provides a speech recognition and emotion recognition system for accurately recognizing diverse speech characteristics and emotions. The system mainly consists of three components: a user, a terminal, and a server.

[0175] Users make voice inputs to terminals installed in public facilities or homes. The user's speech is captured through a microphone built into the terminal. In this process, the terminal is required to collect voice information in real time. For example, using a voice processing chip with noise reduction capabilities allows for the acquisition of cleaner voice data.

[0176] The device first converts the collected audio information into a digital format and extracts acoustic features. Here, using audio signal processing libraries such as Librosa, features such as Mel-frequency cepstrum coefficients (MFCCs) can be obtained from the audio waveform. The device also segments the audio and assigns an identification label to each segment. Furthermore, if necessary, the camera is used to acquire user facial expression data, which is then used to improve the accuracy of emotion recognition.

[0177] The device then encrypts the voice and emotional data and securely transfers it to the server. Secure communication methods such as SSL / TLS protocols are used for communication.

[0178] The server uses the received data to train a virtual model. Specifically, machine learning frameworks such as TENSORFLOW® are used, and the model parameters are adjusted based on supervised learning. By making full use of datasets specific to each region and speech pattern, the accuracy of speech recognition and emotion recognition is continuously improved.

[0179] Once the training is complete, the model is deployed from the server to the terminal, enabling highly accurate recognition in subsequent interactions.

[0180] For example, if a user says to the device, "Hello, tell me today's news," the voice is analyzed in real time, and emotions such as excitement and interest are recognized from the tone and speed of the user's voice, and this information is sent to the server. As a result, the system can provide feedback tailored to the user's needs.

[0181] An example of a prompt to input into a generative AI model is, "Identify what emotions the user is showing during the conversation." Using this prompt allows the system to perform more precise emotion recognition.

[0182] The flow of the specific processing in Example 2 will be explained using Figure 13.

[0183] Step 1:

[0184] The user speaks into the device's microphone. The input is the user's voice, and the output is digital audio data. The microphone built into the device converts the voice into a digital signal. A noise reduction filter is applied during this process to obtain clear audio data.

[0185] Step 2:

[0186] The terminal segments the acquired audio data and extracts acoustic features. The input is digital audio data, and the output is a set of acoustic features. Using an audio signal processing library such as Librosa, the audio data is divided into short time segments, and features such as Mel-frequency cepstrum coefficients (MFCCs) are extracted from each segment.

[0187] Step 3:

[0188] The device assigns an identification label to each audio segment and analyzes emotions using facial expression data acquired by the camera as needed. The input is an acoustic feature set and facial expression data, and the output is the emotion recognition result. The emotion engine evaluates the tone and pitch of the audio and integrates it with the facial expression analysis to determine the emotion label.

[0189] Step 4:

[0190] The terminal encrypts voice and emotion data and sends it to the server. The input is the voice recognition result and emotion recognition result, and the output is an encrypted data stream. Secure communication is performed using the SSL / TLS protocol, and the data is transferred to the server.

[0191] Step 5:

[0192] The server trains a virtual model using the received speech and sentiment data. The input is speech and sentiment data, and the output is an improved virtual model. Supervised learning is performed using TensorFlow to update the model parameters and improve its adaptability to regional dialects and speech patterns.

[0193] Step 6:

[0194] The server deploys the latest trained virtual model to the terminals. The input is the updated virtual model, and the output is the recognition model deployed on each terminal. Through the management platform, the model is installed on each terminal and becomes available for the next recognition process.

[0195] Step 7:

[0196] We collect user feedback and regularly evaluate and improve the accuracy of the model. The input is user feedback, and the output is the improved recognition model. Based on the user experience, the model is tuned, and the overall system performance is optimized.

[0197] (Application Example 2)

[0198] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as a "server" and the smart device 14 as a "terminal".

[0199] In systems that recognize diverse voices and emotions, there is a need for methods to improve user interaction in real time and provide a better experience. This requires not only improved accuracy in voice and emotion recognition, but also a system capable of providing immediate feedback.

[0200] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means.

[0201] In this invention, the server includes means for securely transmitting and storing voice data, means for training voice recognition and emotion recognition models, and means for displaying real-time feedback based on emotion data collected during the conversation. This allows the system to reflect the user's emotional state in real time while they are speaking, enabling a quick and adaptive response.

[0202] A "device" is hardware used to collect diverse speech and securely transmit and store audio data.

[0203] "Audio data" refers to digital information of sounds collected from speakers, which is used to train speech recognition and emotion recognition models.

[0204] A "speech recognition model" is an algorithm that analyzes audio data and converts the speaker's words into text.

[0205] An "emotion recognition model" is an algorithm that analyzes the characteristics of voice and facial expressions to determine the emotional state of a speaker.

[0206] "Feedback" refers to user opinions and evaluations regarding the accuracy of speech recognition and emotion recognition, as well as the user experience of the system.

[0207] "Displaying feedback in real time" means instantly presenting the results of the user's speech and emotional analysis via a display or similar device.

[0208] The system that realizes this invention consists of three main components: a terminal, a server, and a user.

[0209] The server utilizes a high-speed database and encrypted communication protocols for the secure storage and processing of speech and emotion data. Cloud-based computing resources and machine learning algorithms (such as TensorFlow and PyTorch) are used to train speech recognition and emotion recognition models, and the models are improved to accommodate regional dialects and emotional expressions.

[0210] The device is equipped with a microphone and camera, which are used to collect the user's voice and facial expressions in real time. The voice data is converted into text data using the Google® Cloud Speech-to-Text API, and the Emotion Recognition API is used for emotion recognition. The device combines these APIs to perform voice and emotion recognition and has the ability to provide real-time feedback to the user. The feedback is displayed on the device's display or connected smart glasses, enabling the user to respond immediately.

[0211] For example, when a customer smiles and says, "I really like the design of this product," the device recognizes the voice and determines that the emotion is "joy." This information is displayed in real time, allowing staff to introduce similar products to the customer as the next topic.

[0212] An example of a prompt when using a generative AI model is: "The customer has recently given positive feedback on a product. Use smart glasses to recognize the voice and emotion and provide feedback that suggests the customer is satisfied."

[0213] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[0214] Step 1:

[0215] The device uses a microphone and camera to collect the user's voice and facial expressions in real time. The input consists of the user's speech data and associated image data, which are converted into a digital format and passed on to the next processing step. The device digitizes the audio signal from the microphone and captures still image data from the camera at regular intervals.

[0216] Step 2:

[0217] The device sends the collected audio data to the Google Cloud Speech-to-Text API, where it is converted into linguistic data. The input is digitized audio data, which is then parsed and output as text data. This API call yields a set of words extracted from the audio.

[0218] Step 3:

[0219] The device sends facial expression data to the Emotion Recognition API to determine the user's emotions. The input is captured image data of facial expressions, and the output is the identified emotion label. Emotions are identified through image analysis, recognizing emotions such as happiness, sadness, and anger.

[0220] Step 4:

[0221] The device provides real-time feedback to the user on analyzed text and sentiment data. The input is the result of steps 2 and 3, and the output is information processed into a user-friendly format. The device then displays feedback such as "The customer is happy" on smart glasses or a display.

[0222] Step 5:

[0223] The server collects and stores voice and emotion data via a secure communication protocol. Input is data transmitted from the terminal, and output is an encrypted dataset. It communicates with a database to store the dataset.

[0224] Step 6:

[0225] The server updates the generative AI model using the collected data to improve the accuracy of speech and emotion recognition. The input is the stored dataset, and the output is the updated recognition model. Machine learning algorithms are used to retrain the model with the data, thereby improving the model's accuracy.

[0226] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0227] Data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of data generation model 58 is ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> ), Gemini (registered trademark) (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0228] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and the specific processing may also be performed by the smart device 14.

[0229] [Second Embodiment]

[0230] Figure 3 shows an example of the configuration of the data processing system 210 according to the second embodiment.

[0231] As shown in Figure 3, the data processing system 210 includes a data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0232] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0233] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication interface 44. The computer 36 includes a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The microphone 238, speaker 240, and camera 42 are also connected to the bus 52.

[0234] The microphone 238 receives voice signals from the user 20 and receives instructions from the user 20. The microphone 238 captures the voice signals from the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to the instructions from the processor 46.

[0235] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the area around the user 20 (for example, an imaging range defined by a field of view equivalent to the width of a typical healthy person's field of vision).

[0236] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various information between processor 46 and processor 28 via network 54. The exchange of various information between processor 46 and processor 28 using communication interfaces 44 and 26 is performed in a secure manner.

[0237] Figure 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Figure 4, the data processing device 12 performs specific processing using the processor 28. The storage 32 stores the specific processing program 56.

[0238] The specific processing program 56 is an example of a "program" relating to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0239] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0240] In the smart glasses 214, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0241] Next, the identification processing performed by the identification processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0242] This invention provides a method for collecting and processing voice data using terminals installed throughout Japan, as a voice recognition system that includes regionally specific dialects. This system consists of three components: a user, a terminal, and a server.

[0243] First, the user engages in a casual conversation with a nearby device. After obtaining necessary consent, the audio data collected during the conversation is recorded and segmented in real time by the device. Audio features are extracted, and provisional labels are assigned according to the type of dialect. The audio data undergoes noise reduction processing to remove unwanted noise, and the processed data is then encrypted and stored.

[0244] Next, the server receives the audio data transmitted from the terminal. The server analyzes this data and uses it as training data for a speech recognition model. Using supervised learning techniques, the server generates and updates the speech recognition model based on the audio data for each region. This speech recognition model is adjusted to have the ability to identify region-specific speech features.

[0245] Trained speech recognition models are automatically distributed from the server to each terminal. This ensures that terminals always perform speech recognition based on the latest model, improving recognition accuracy in each region. Furthermore, through user interaction, terminals provide the server with speech recognition results and user feedback. The server then retrains the model based on the collected feedback to further improve accuracy.

[0246] As a concrete example, a user speaking the local dialect engages in everyday conversation using a terminal installed in a public facility in a certain area. This audio is collected in real time and sent to a server as data with unique regional vocal characteristics. The server learns from this data and, for example, updates its speech recognition algorithm to one that is specialized for a particular dialect. The latest model is deployed to the terminal, enabling even higher accuracy the next time speech recognition is performed on that terminal.

[0247] In this way, this system can support diverse regional dialects and eliminate regional disparities in speech recognition technology.

[0248] The following describes the processing flow.

[0249] Step 1:

[0250] Users consent to the collection of voice data and engage in everyday conversations through the device. They can engage in natural conversations, including those using dialects, in public facilities and home environments where the device is installed.

[0251] Step 2:

[0252] The device begins recording the user's conversation in real time. The audio data is segmented by detecting breaks in speech, and each segment is assigned a temporary label.

[0253] Step 3:

[0254] The device then applies preprocessing, such as noise reduction, to the segmented audio data. This preprocessing improves sound quality and prepares data suitable for training the speech recognition model.

[0255] Step 4:

[0256] The device encrypts the processed voice data and sends it to the server via a secure communication protocol. The data is protected with the utmost consideration for user privacy.

[0257] Step 5:

[0258] The server saves the received audio data to a database. During saving, dialect information and speech characteristic labels associated with the audio data are also recorded.

[0259] Step 6:

[0260] The server uses stored audio data to train a speech recognition model. Using supervised learning, the speech recognition model learns region-specific speech features and improves the accuracy of speech recognition.

[0261] Step 7:

[0262] The server deploys the newly trained speech recognition model to each terminal. The terminal downloads this new model and uses it for speech recognition processing.

[0263] Step 8:

[0264] Users can provide feedback on the speech recognition results from their devices. This user feedback is valuable information for improving the speech recognition results.

[0265] Step 9:

[0266] The device periodically sends the collected feedback to the server. This feedback is used to consider whether to retrain the model.

[0267] Step 10:

[0268] The server analyzes the feedback data and adjusts the parameters of the speech recognition model as needed. This further improves speech recognition accuracy in each region and enhances the overall system performance.

[0269] (Example 1)

[0270] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server," and the smart glasses 214 will be referred to as the "terminal."

[0271] In speech recognition technology, there is a need to achieve highly accurate speech recognition while being able to handle diverse, region-specific speech characteristics. However, conventional speech recognition systems have the problem of failing to adequately handle regional pronunciations and dialects, resulting in reduced recognition accuracy. Furthermore, there is a lack of mechanisms to effectively incorporate user feedback and update the models, making continuous model improvement difficult.

[0272] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[0273] In this invention, the server includes: an information device means having the function of acquiring voice information from users with diverse voices; means for securely transmitting and storing the acquired voice information; and means for collecting user feedback regarding the accuracy of speech recognition. This enables rapid and accurate learning of various region-specific voice characteristics, and improves and continuously refines the accuracy of the speech recognition model.

[0274] "Diverse speech" refers to the vocalizations of multiple speakers with different pronunciations, dialects, and accents.

[0275] A "user" refers to an individual who interacts with the system and provides voice information through this system.

[0276] "Voice information" refers to data that is acquired and stored in digital format from speech uttered by users.

[0277] An "information device" is a device used to acquire, process, and transmit voice information, and includes components such as microphones and processors.

[0278] A "speech recognition model" is an algorithm or system built to analyze speech information and convert it into text or specific commands.

[0279] "Feedback" refers to opinions from users regarding the results of speech recognition, their impressions, and suggestions for improvement, which are used to improve the system.

[0280] "Supervised learning" is a method of training machine learning models using labeled datasets, which are then used to enable the models to make accurate predictions and classifications.

[0281] This section describes an embodiment of this system. This system aims to address diverse, region-specific speech characteristics using speech recognition technology. The system primarily consists of three elements: the user, the terminal, and the server.

[0282] The user transmits voice through the installed terminal. For example, when the user talks about a famous place in a specific area, the voice is collected by the terminal. The voice information at this time is naturally collected from the user's daily conversations.

[0283] The terminal records the voice emitted by the user in real time and segments it into digital voice data. This terminal is equipped with a microphone and a signal processing unit for extracting voice features. The recorded voice data is extracted with feature quantities using technologies such as Mel Frequency Cepstral Coefficients (MFCC). Also, noise reduction is performed using digital signal processing to obtain clear voice data. Thereby, an initial label is assigned based on acoustic features. Thereafter, the voice data is encrypted using the AES (Advanced Encryption Standard) protocol and securely stored.

[0284] The server receives the data transmitted from the terminal and uses it for the training of the voice recognition model. Utilizing supervised learning techniques such as deep learning, it learns the voice characteristics of a specific area and improves the recognition accuracy. The trained voice recognition model is push-delivered to each terminal, enabling efficient performance of recognition tasks in the latest state. Also, the server collects feedback from the user and continuously improves the model based on it.

[0285] As a specific example, when the user uses the terminal and speaks the unique dialect of the area where they live, the voice is collected and analyzed by the server. Through this process, the voice recognition system can become proficient in the dialect of that area and improve the recognition accuracy.

[0286] As an example of a prompt sentence in the generative AI model, there is "Please update the voice recognition model using the voice data collected in a specific area for a voice recognition system that identifies the dialect unique to the area."

[0287] In this way, this system can recognize voices from diverse regions with high accuracy, contributing to the elimination of regional disparities in speech recognition.

[0288] The flow of the specific processing in Example 1 will be explained using Figure 11.

[0289] Step 1:

[0290] The user begins a casual conversation into a terminal installed in their local area. The input information is the user's speech, which is captured as audio data by the terminal's high-sensitivity microphone. A specific example of this operation would be the user uttering something like, "I'm going to talk about the local festival." The output is raw audio data.

[0291] Step 2:

[0292] The terminal segments the acquired audio data in real time. The input is audio data from the user, and the digital signal processing unit divides this data so that it can be easily managed. Specifically, it divides the audio into segments of regular time intervals and saves each segment as a separate file. The output is a collection of segmented audio files.

[0293] Step 3:

[0294] The device extracts features from audio data. The input is a segmented audio file, from which acoustic features are extracted using techniques such as Mel-frequency cepstrum coefficients (MFCCs). Specifically, it analyzes the frequency and temporal characteristics of each audio segment and converts them into numerical data. The output is a list of the extracted acoustic features.

[0295] Step 4:

[0296] The device applies noise reduction processing to the feature data. The input is feature data, and it performs specific operations to remove noise using digital signal processing. This improves the accuracy of speech recognition. The output is clean speech data with reduced noise.

[0297] Step 5:

[0298] The device assigns a temporary label to the noise-reduced audio data, encrypts it, and saves it. The input is noise-reduced audio data. Specifically, a temporary label appropriate to the audio content is assigned, the data is encrypted using the AES protocol, and then saved. The output is labeled and encrypted audio data.

[0299] Step 6:

[0300] The terminal sends encrypted audio data to the server. The input is labeled encrypted data. The terminal uses the TLS protocol to perform the specific action of securely communicating and sending the data to the server. The output is the audio data securely transferred to the server.

[0301] Step 7:

[0302] The server trains a speech recognition model using the received audio data. The input is audio data sent from the terminal, and the model is optimized using deep learning techniques. Specifically, it learns new acoustic features and updates parameters to improve accuracy. The output is the new model parameters.

[0303] Step 8:

[0304] The server provides each terminal with a trained speech recognition model. The input is the updated model parameters. The server performs a push notification to each terminal to keep the model up-to-date. The output is each terminal with the latest speech recognition model installed.

[0305] Step 9:

[0306] The terminal conducts a conversation with the user using a new speech recognition model. The input is new utterance data from the user, and it performs specific operations to process it using the new model. The output is a speech recognition result with improved accuracy.

[0307] Step 10:

[0308] The terminal sends the feedback data from the user to the server together with the recognition result. The input is the speech recognition result and the feedback information, and as a specific operation, it sends them to the server. The output is that the feedback information is sent to the server and used as material for further model improvement.

[0309] (Application Example 1)

[0310] Next, Application Example 1 will be described. In the following description, the data processing device 12 is referred to as the "server", and the smart glasses 214 are referred to as the "terminal".

[0311] A speech recognition system that includes region-specific dialects needs to perform speech recognition with high accuracy even in the user's daily conversations in the dialect, aiming to promote utilization in actual stores and improve customer service. However, improving the recognition accuracy for dealing with various dialects remains an issue. In particular, there is a demand for constructing an environment where speech data can be collected and models can be updated in real time.

[0312] The specific processing by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following respective means.

[0313] In this invention, the server includes [means for using a device having a function of collecting speech data from speakers with various utterances, [means for providing and updating a trained speech recognition algorithm to each device, [means for using the speech recognition result to reply information to the output device. Thereby, it becomes possible to efficiently recognize speech including region-specific dialects and immediately feedback the result to the user.

[0314] A "speaker with diverse speech patterns" refers to a speaker who has unique dialects or ways of speaking that differ depending on the region and culture.

[0315] "Audio data" refers to data that records a speaker's utterances in digital format.

[0316] "Device" refers to a device that includes hardware or software used to collect and process audio data.

[0317] A "secure method" refers to a method that uses encryption and authentication techniques to protect audio data from unauthorized access and tampering.

[0318] A "speech recognition algorithm" is a computational method or process for generating text from speech data.

[0319] "Response" refers to the feedback information provided by the user based on the results of speech recognition.

[0320] "Information" refers to the content and data output based on the results obtained through speech recognition.

[0321] This invention aims to construct a speech recognition system that can handle a variety of dialects. The system is primarily implemented through three elements: a server, a terminal, and a user.

[0322] The server performs a series of processes, from collecting audio data to training and updating recognition algorithms. Specifically, it processes the collected audio data in a secure manner and trains speech recognition algorithms. The audio data is used to generate region-specific models.

[0323] The terminal collects voice data through interaction with the user and transmits it to the server. Appropriate hardware and software are used for voice data collection, such as smart devices with microphones or personal computers. The voice data transmitted from the terminal to the server is securely transmitted via the internet to enable real-time speech recognition.

[0324] The user communicates with the system by speaking, including in dialect, using a terminal. The terminal collects the user's speech, encrypts it, and sends it to the server. The server trains a speech recognition model, returns the trained model to the terminal, and improves the accuracy of speech recognition.

[0325] As a concrete example, imagine a scenario where a user asks for tourist information in their local dialect at a terminal installed in a commercial facility in a certain region. In this case, the system receives the user's speech, performs voice recognition, and immediately returns appropriate tourist information to the user's smartphone.

[0326] One example of how a generative AI model can be used is a prompt such as, "Please tell me tourist information in the Okinawan dialect." This input allows the server to generate an audio guide and provide the user with the most relevant information.

[0327] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[0328] Step 1:

[0329] The device acquires voice input from the user via a microphone. The acquired voice is recorded as digital audio data. This data undergoes noise reduction processing to reduce background noise. The noise-reduced audio data then becomes the input for the next step.

[0330] Step 2:

[0331] The terminal segments the denoised audio data and extracts the audio features contained in each segment. An acoustic analysis algorithm is used for this process. Based on the audio features, a provisional label is assigned to the audio data. The labeled data is then transmitted to the server via a secure communication protocol.

[0332] Step 3:

[0333] The server receives audio data transmitted from the terminal and uses it to train the speech recognition algorithm. Supervised learning is applied to the training to improve the ability to identify diverse regional speech features. The output of this step is the updated speech recognition algorithm.

[0334] Step 4:

[0335] The server distributes the trained speech recognition algorithm to the terminal. The terminal incorporates the distributed algorithm into its processing mechanism and prepares for the next speech recognition. This update enables more accurate speech recognition.

[0336] Step 5:

[0337] The user accesses the device again and performs voice input. The device utilizes a newly trained speech recognition algorithm to recognize the user's voice. Based on the recognition result, it returns appropriate information to the user. This output is provided to the user as either voice or text.

[0338] Step 6:

[0339] The device collects user feedback on recognition accuracy and sends it to the server. The server uses this feedback to further improve the model. It analyzes the collected feedback data and updates the algorithm as needed.

[0340] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[0341] This invention provides a system for collecting speech data from speakers with diverse speech patterns and performing speech recognition and emotion recognition. The system has the function of analyzing user speech and improving the user experience. It consists of three main components: the user, the terminal, and the server.

[0342] First, the user begins a conversation into a terminal installed in a public facility or home. Once the conversation begins, the terminal records the audio in real time and uses it for speech recognition and emotion recognition. Here, the terminal divides the audio data into segments, and each segment is labeled based on its acoustic characteristics.

[0343] The device further uses an emotion engine to identify emotions from the user's voice. This emotion recognition improves accuracy not only by using acoustic features but also by combining it with facial expression data from the camera as needed. The results of emotion recognition and voice recognition are integrated, making it possible to understand the user's state in detail.

[0344] Voice and emotion data are securely transmitted to the server via encrypted communication. The server uses this data to train speech recognition and emotion recognition models. In particular, improvements in recognition accuracy are made to accommodate regional dialects and speech patterns.

[0345] Trained speech recognition and emotion recognition models are deployed from the server to each terminal. This allows the terminal to interact with the user using the latest speech and emotion recognition models.

[0346] As a concrete example, a user speaking the Kyoto dialect converses using a terminal installed in a public facility. The terminal collects and segments voice data in real time, and in the background, an emotion engine analyzes the user's emotions from the tone and speed of their voice. For example, if an emotion expressing joy is detected, that information is sent to the server along with the utterance and used to improve the recognition model. Ultimately, the recognition rate improves, and the quality of the user experience is enhanced.

[0347] In this way, this system can bridge the gap in voice and emotion recognition between regions and improve various user experiences.

[0348] The following describes the processing flow.

[0349] Step 1:

[0350] The user initiates a conversation with the installed terminal. The terminal obtains permission to begin recording immediately upon the start of the conversation.

[0351] Step 2:

[0352] The device divides the recorded audio into segments in real time. The segmented audio is then tentatively labeled based on specific speech intervals.

[0353] Step 3:

[0354] The device extracts acoustic features from the audio and uses an emotion engine to identify the user's emotions. In this process, it captures not only audio but also facial expression data with a camera and includes it in the analysis.

[0355] Step 4:

[0356] The device encrypts both voice and emotional data and transmits them to the server via a secure network. This ensures that the data is managed in a privacy-protected manner.

[0357] Step 5:

[0358] The server stores the received audio data and emotion data in a database. During storage, it adds relevant information such as the speech label and emotion identification result to each data.

[0359] Step 6:

[0360] The server uses the collected data to train speech recognition and emotion recognition models. It aims to improve recognition accuracy by learning region-specific speech patterns and emotional expressions.

[0361] Step 7:

[0362] The server deploys newly trained speech recognition and emotion recognition models to each terminal. The terminal receives this update and makes the latest recognition algorithms available.

[0363] Step 8:

[0364] Users can provide feedback on the speech recognition and emotion recognition results provided by the device, including evaluations and suggestions for improvement.

[0365] Step 9:

[0366] The device collects user feedback and periodically sends it to the server. This feedback is used to further improve the model.

[0367] Step 10:

[0368] The server readjusts the speech recognition and emotion recognition models based on feedback and retrains them as needed. This improves the accuracy of the recognition system and user satisfaction.

[0369] (Example 2)

[0370] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0371] Conventional speech recognition systems struggle to accurately recognize diverse speech characteristics and emotions, resulting in a reduced user experience. Furthermore, there is a need to improve recognition accuracy to accommodate regionally specific dialects and speech patterns. This invention aims to solve these problems and provide a richer user experience by improving the accuracy of speech and emotion recognition.

[0372] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[0373] In this invention, the server includes means for using a device that has the function of collecting speech information from speech sources with diverse utterances, means for securely transmitting and storing the collected speech information, and means for training a virtual model for speech recognition and emotion recognition using the stored speech information. This makes it possible to develop a model that can effectively recognize diverse speech characteristics and is compatible with regionally specific dialects and speech patterns.

[0374] "Diverse speech" refers to speech that exhibits a variety of different phonetic patterns and linguistic characteristics.

[0375] A "sound source" is a subject that generates sound, and the concept includes sounds emitted by humans and machines.

[0376] "Audio information" refers to audio data and its characteristics obtained from an audio source, and includes data such as the waveform and frequency components of the audio.

[0377] "Device" refers to an interface equipped with hardware and software that has the function of collecting, processing, and transmitting audio information.

[0378] "Secure transmission" refers to methods for safely transferring voice information without disclosing it to third parties, and includes communication using encryption technology.

[0379] "Storage" refers to the act of securely saving acquired audio information in a database or storage device so that it can be used later.

[0380] A "virtual model" is a set of algorithms and their parameters built on audio information, and is a computerized simulation that enables speech and emotion recognition.

[0381] "Supervised learning" is a technique for training a virtual model using known input-output pairs to improve the accuracy of predictions on new data.

[0382] "Evaluation" refers to the act of collecting user feedback and quality information regarding the results of speech recognition and emotion recognition.

[0383] "Improvement" is the process of enhancing the accuracy and functionality of a virtual model based on the collected evaluation data.

[0384] This invention provides a speech recognition and emotion recognition system for accurately recognizing diverse speech characteristics and emotions. The system mainly consists of three components: a user, a terminal, and a server.

[0385] Users make voice inputs to terminals installed in public facilities or homes. The user's speech is captured through a microphone built into the terminal. In this process, the terminal is required to collect voice information in real time. For example, using a voice processing chip with noise reduction capabilities allows for the acquisition of cleaner voice data.

[0386] The device first converts the collected audio information into a digital format and extracts acoustic features. Here, using audio signal processing libraries such as Librosa, features such as Mel-frequency cepstrum coefficients (MFCCs) can be obtained from the audio waveform. The device also segments the audio and assigns an identification label to each segment. Furthermore, if necessary, the camera is used to acquire user facial expression data, which is then used to improve the accuracy of emotion recognition.

[0387] The device then encrypts the voice and emotional data and securely transfers it to the server. Secure communication methods such as SSL / TLS protocols are used for communication.

[0388] The server uses the received data to train a virtual model. Specifically, machine learning frameworks such as TensorFlow are used, and the model parameters are adjusted based on supervised learning. By utilizing datasets specific to each region and speech pattern, the accuracy of speech recognition and emotion recognition is continuously improved.

[0389] Once the training is complete, the model is deployed from the server to the terminal, enabling highly accurate recognition in subsequent interactions.

[0390] For example, if a user says to the device, "Hello, tell me today's news," the voice is analyzed in real time, and emotions such as excitement and interest are recognized from the tone and speed of the user's voice, and this information is sent to the server. As a result, the system can provide feedback tailored to the user's needs.

[0391] An example of a prompt to input into a generative AI model is, "Identify what emotions the user is showing during the conversation." Using this prompt allows the system to perform more precise emotion recognition.

[0392] The flow of the specific processing in Example 2 will be explained using Figure 13.

[0393] Step 1:

[0394] The user speaks into the device's microphone. The input is the user's voice, and the output is digital audio data. The microphone built into the device converts the voice into a digital signal. A noise reduction filter is applied during this process to obtain clear audio data.

[0395] Step 2:

[0396] The terminal segments the acquired audio data and extracts acoustic features. The input is digital audio data, and the output is a set of acoustic features. Using an audio signal processing library such as Librosa, the audio data is divided into short time segments, and features such as Mel-frequency cepstrum coefficients (MFCCs) are extracted from each segment.

[0397] Step 3:

[0398] The device assigns an identification label to each audio segment and analyzes emotions using facial expression data acquired by the camera as needed. The input is an acoustic feature set and facial expression data, and the output is the emotion recognition result. The emotion engine evaluates the tone and pitch of the audio and integrates it with the facial expression analysis to determine the emotion label.

[0399] Step 4:

[0400] The terminal encrypts voice and emotion data and sends it to the server. The input is the voice recognition result and emotion recognition result, and the output is an encrypted data stream. Secure communication is performed using the SSL / TLS protocol, and the data is transferred to the server.

[0401] Step 5:

[0402] The server trains a virtual model using the received speech and sentiment data. The input is speech and sentiment data, and the output is an improved virtual model. Supervised learning is performed using TensorFlow to update the model parameters and improve its adaptability to regional dialects and speech patterns.

[0403] Step 6:

[0404] The server deploys the latest trained virtual model to the terminals. The input is the updated virtual model, and the output is the recognition model deployed on each terminal. Through the management platform, the model is installed on each terminal and becomes available for the next recognition process.

[0405] Step 7:

[0406] We collect user feedback and regularly evaluate and improve the accuracy of the model. The input is user feedback, and the output is the improved recognition model. Based on the user experience, the model is tuned, and the overall system performance is optimized.

[0407] (Application Example 2)

[0408] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as the "server," and the smart glasses 214 will be referred to as the "terminal."

[0409] In systems that recognize diverse voices and emotions, there is a need for methods to improve user interaction in real time and provide a better experience. This requires not only improved accuracy in voice and emotion recognition, but also a system capable of providing immediate feedback.

[0410] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means.

[0411] In this invention, the server includes means for securely transmitting and storing voice data, means for training voice recognition and emotion recognition models, and means for displaying real-time feedback based on emotion data collected during the conversation. This allows the system to reflect the user's emotional state in real time while they are speaking, enabling a quick and adaptive response.

[0412] A "device" is hardware used to collect diverse speech and securely transmit and store audio data.

[0413] "Audio data" refers to digital information of sounds collected from speakers, which is used to train speech recognition and emotion recognition models.

[0414] A "speech recognition model" is an algorithm that analyzes audio data and converts the speaker's words into text.

[0415] An "emotion recognition model" is an algorithm that analyzes the characteristics of voice and facial expressions to determine the emotional state of a speaker.

[0416] "Feedback" refers to user opinions and evaluations regarding the accuracy of speech recognition and emotion recognition, as well as the user experience of the system.

[0417] "Displaying feedback in real time" means instantly presenting the results of the user's speech and emotional analysis via a display or similar device.

[0418] The system that realizes this invention consists of three main components: a terminal, a server, and a user.

[0419] The server utilizes a high-speed database and encrypted communication protocols for the secure storage and processing of speech and emotion data. Cloud-based computing resources and machine learning algorithms (such as TensorFlow and PyTorch) are used to train speech recognition and emotion recognition models, and the models are improved to accommodate regional dialects and emotional expressions.

[0420] The device is equipped with a microphone and camera, which are used to collect the user's voice and facial expressions in real time. The voice data is converted into text data using the Google Cloud Speech-to-Text API, and the Emotion Recognition API is used for emotion recognition. The device combines these APIs to perform voice and emotion recognition and has the ability to provide real-time feedback to the user. The feedback is displayed on the device's display or connected smart glasses, enabling the user to respond immediately.

[0421] For example, when a customer smiles and says, "I really like the design of this product," the device recognizes the voice and determines that the emotion is "joy." This information is displayed in real time, allowing staff to introduce similar products to the customer as the next topic.

[0422] An example of a prompt when using a generative AI model is: "The customer has recently given positive feedback on a product. Use smart glasses to recognize the voice and emotion and provide feedback that suggests the customer is satisfied."

[0423] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[0424] Step 1:

[0425] The device uses a microphone and camera to collect the user's voice and facial expressions in real time. The input consists of the user's speech data and associated image data, which are converted into a digital format and passed on to the next processing step. The device digitizes the audio signal from the microphone and captures still image data from the camera at regular intervals.

[0426] Step 2:

[0427] The device sends the collected audio data to the Google Cloud Speech-to-Text API, where it is converted into linguistic data. The input is digitized audio data, which is then parsed and output as text data. This API call yields a set of words extracted from the audio.

[0428] Step 3:

[0429] The device sends facial expression data to the Emotion Recognition API to determine the user's emotions. The input is captured image data of facial expressions, and the output is the identified emotion label. Emotions are identified through image analysis, recognizing emotions such as happiness, sadness, and anger.

[0430] Step 4:

[0431] The device provides real-time feedback to the user on analyzed text and sentiment data. The input is the result of steps 2 and 3, and the output is information processed into a user-friendly format. The device then displays feedback such as "The customer is happy" on smart glasses or a display.

[0432] Step 5:

[0433] The server collects and stores voice and emotion data via a secure communication protocol. Input is data transmitted from the terminal, and output is an encrypted dataset. It communicates with a database to store the dataset.

[0434] Step 6:

[0435] The server updates the generative AI model using the collected data to improve the accuracy of speech and emotion recognition. The input is the stored dataset, and the output is the updated recognition model. Machine learning algorithms are used to retrain the model with the data, thereby improving the model's accuracy.

[0436] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 238 to the data processing unit 12. In the data processing unit 12, the specific processing unit 290 acquires the audio data.

[0437] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0438] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and the specific processing may also be performed by the smart glasses 214.

[0439] [Third Embodiment]

[0440] Figure 5 shows an example of the configuration of the data processing system 310 according to the third embodiment.

[0441] As shown in Figure 5, the data processing system 310 includes a data processing device 12 and a headset terminal 314. An example of the data processing device 12 is a server.

[0442] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0443] The headset terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication interface 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The microphone 238, speaker 240, camera 42, and display 343 are also connected to the bus 52.

[0444] The microphone 238 receives voice signals from the user 20 and receives instructions from the user 20. The microphone 238 captures the voice signals from the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to the instructions from the processor 46.

[0445] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the area around the user 20 (for example, an imaging range defined by a field of view equivalent to the width of a typical healthy person's field of vision).

[0446] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various information between processor 46 and processor 28 via network 54. The exchange of various information between processor 46 and processor 28 using communication interfaces 44 and 26 is performed in a secure manner.

[0447] Figure 6 shows an example of the main functions of the data processing device 12 and the headset terminal 314. As shown in Figure 6, the data processing device 12 performs specific processing using the processor 28. The storage 32 stores the specific processing program 56.

[0448] The specific processing program 56 is an example of a "program" relating to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0449] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0450] In the headset terminal 314, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0451] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the headset terminal 314 will be referred to as the "terminal".

[0452] This invention provides a method for collecting and processing voice data using terminals installed throughout Japan, as a voice recognition system that includes regionally specific dialects. This system consists of three components: a user, a terminal, and a server.

[0453] First, the user engages in a casual conversation with a nearby device. After obtaining necessary consent, the audio data collected during the conversation is recorded and segmented in real time by the device. Audio features are extracted, and provisional labels are assigned according to the type of dialect. The audio data undergoes noise reduction processing to remove unwanted noise, and the processed data is then encrypted and stored.

[0454] Next, the server receives the audio data transmitted from the terminal. The server analyzes this data and uses it as training data for a speech recognition model. Using supervised learning techniques, the server generates and updates the speech recognition model based on the audio data for each region. This speech recognition model is adjusted to have the ability to identify region-specific speech features.

[0455] Trained speech recognition models are automatically distributed from the server to each terminal. This ensures that terminals always perform speech recognition based on the latest model, improving recognition accuracy in each region. Furthermore, through user interaction, terminals provide the server with speech recognition results and user feedback. The server then retrains the model based on the collected feedback to further improve accuracy.

[0456] As a concrete example, a user speaking the local dialect engages in everyday conversation using a terminal installed in a public facility in a certain area. This audio is collected in real time and sent to a server as data with unique regional vocal characteristics. The server learns from this data and, for example, updates its speech recognition algorithm to one that is specialized for a particular dialect. The latest model is deployed to the terminal, enabling even higher accuracy the next time speech recognition is performed on that terminal.

[0457] In this way, this system can support diverse regional dialects and eliminate regional disparities in speech recognition technology.

[0458] The following describes the processing flow.

[0459] Step 1:

[0460] Users consent to the collection of voice data and engage in everyday conversations through the device. They can engage in natural conversations, including those using dialects, in public facilities and home environments where the device is installed.

[0461] Step 2:

[0462] The device begins recording the user's conversation in real time. The audio data is segmented by detecting breaks in speech, and each segment is assigned a temporary label.

[0463] Step 3:

[0464] The device then applies preprocessing, such as noise reduction, to the segmented audio data. This preprocessing improves sound quality and prepares data suitable for training the speech recognition model.

[0465] Step 4:

[0466] The device encrypts the processed voice data and sends it to the server via a secure communication protocol. The data is protected with the utmost consideration for user privacy.

[0467] Step 5:

[0468] The server saves the received audio data to a database. During saving, dialect information and speech characteristic labels associated with the audio data are also recorded.

[0469] Step 6:

[0470] The server uses stored audio data to train a speech recognition model. Using supervised learning, the speech recognition model learns region-specific speech features and improves the accuracy of speech recognition.

[0471] Step 7:

[0472] The server deploys the newly trained speech recognition model to each terminal. The terminal downloads this new model and uses it for speech recognition processing.

[0473] Step 8:

[0474] Users can provide feedback on the speech recognition results from their devices. This user feedback is valuable information for improving the speech recognition results.

[0475] Step 9:

[0476] The device periodically sends the collected feedback to the server. This feedback is used to consider whether to retrain the model.

[0477] Step 10:

[0478] The server analyzes the feedback data and adjusts the parameters of the speech recognition model as needed. This further improves speech recognition accuracy in each region and enhances the overall system performance.

[0479] (Example 1)

[0480] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[0481] In speech recognition technology, there is a need to achieve highly accurate speech recognition while being able to handle diverse, region-specific speech characteristics. However, conventional speech recognition systems have the problem of failing to adequately handle regional pronunciations and dialects, resulting in reduced recognition accuracy. Furthermore, there is a lack of mechanisms to effectively incorporate user feedback and update the models, making continuous model improvement difficult.

[0482] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[0483] In this invention, the server includes: an information device means having the function of acquiring voice information from users with diverse voices; means for securely transmitting and storing the acquired voice information; and means for collecting user feedback regarding the accuracy of speech recognition. This enables rapid and accurate learning of various region-specific voice characteristics, and improves and continuously refines the accuracy of the speech recognition model.

[0484] "Diverse speech" refers to the vocalizations of multiple speakers with different pronunciations, dialects, and accents.

[0485] A "user" refers to an individual who interacts with the system and provides voice information through this system.

[0486] "Voice information" refers to data that is acquired and stored in digital format from speech uttered by users.

[0487] An "information device" is a device used to acquire, process, and transmit voice information, and includes components such as microphones and processors.

[0488] A "speech recognition model" is an algorithm or system built to analyze speech information and convert it into text or specific commands.

[0489] "Feedback" refers to opinions from users regarding the results of speech recognition, their impressions, and suggestions for improvement, which are used to improve the system.

[0490] "Supervised learning" is a method of training machine learning models using labeled datasets, which are then used to enable the models to make accurate predictions and classifications.

[0491] This section describes an embodiment of this system. This system aims to address diverse, region-specific speech characteristics using speech recognition technology. The system primarily consists of three elements: the user, the terminal, and the server.

[0492] Users transmit audio through the installed terminals. For example, when a user talks about a local landmark, the audio is collected by the terminal. This audio information is naturally gathered from the user's everyday conversations.

[0493] The device records the user's voice in real time and segments it as digital audio data. This device is equipped with a microphone and signal processing unit for extracting audio features. Features are extracted from the recorded audio data using techniques such as Mel-frequency cepstrum coefficients (MFCC). Noise reduction is also performed using digital signal processing to obtain clear audio data. Initial labels are then assigned based on the acoustic characteristics. The audio data is then encrypted using the AES (Advanced Encryption Standard) protocol and stored securely.

[0494] The server receives data transmitted from terminals and uses it to train speech recognition models. It utilizes guided learning techniques such as deep learning to learn the speech characteristics of specific regions and improve recognition accuracy. The trained speech recognition models are pushed to each terminal, allowing them to efficiently perform recognition tasks in their most up-to-date state. The server also collects user feedback and continuously improves the models based on it.

[0495] As a concrete example, when a user uses a device to speak in the local dialect, the audio is collected and analyzed by a server. Through this process, the speech recognition system becomes familiar with the local dialect and can improve its recognition accuracy.

[0496] An example of a prompt in a generative AI model is, "Update the speech recognition model using speech data collected from a specific region for a speech recognition system that identifies regional dialects."

[0497] In this way, this system can recognize voices from diverse regions with high accuracy, contributing to the elimination of regional disparities in speech recognition.

[0498] The flow of the specific processing in Example 1 will be explained using Figure 11.

[0499] Step 1:

[0500] The user begins a casual conversation into a terminal installed in their local area. The input information is the user's speech, which is captured as audio data by the terminal's high-sensitivity microphone. A specific example of this operation would be the user uttering something like, "I'm going to talk about the local festival." The output is raw audio data.

[0501] Step 2:

[0502] The terminal segments the acquired audio data in real time. The input is audio data from the user, and the digital signal processing unit divides this data so that it can be easily managed. Specifically, it divides the audio into segments of regular time intervals and saves each segment as a separate file. The output is a collection of segmented audio files.

[0503] Step 3:

[0504] The device extracts features from audio data. The input is a segmented audio file, from which acoustic features are extracted using techniques such as Mel-frequency cepstrum coefficients (MFCCs). Specifically, it analyzes the frequency and temporal characteristics of each audio segment and converts them into numerical data. The output is a list of the extracted acoustic features.

[0505] Step 4:

[0506] The device applies noise reduction processing to the feature data. The input is feature data, and it performs specific operations to remove noise using digital signal processing. This improves the accuracy of speech recognition. The output is clean speech data with reduced noise.

[0507] Step 5:

[0508] The device assigns a temporary label to the noise-reduced audio data, encrypts it, and saves it. The input is noise-reduced audio data. Specifically, a temporary label appropriate to the audio content is assigned, the data is encrypted using the AES protocol, and then saved. The output is labeled and encrypted audio data.

[0509] Step 6:

[0510] The terminal sends encrypted audio data to the server. The input is labeled encrypted data. The terminal uses the TLS protocol to perform the specific action of securely communicating and sending the data to the server. The output is the audio data securely transferred to the server.

[0511] Step 7:

[0512] The server trains a speech recognition model using the received audio data. The input is audio data sent from the terminal, and the model is optimized using deep learning techniques. Specifically, it learns new acoustic features and updates parameters to improve accuracy. The output is the new model parameters.

[0513] Step 8:

[0514] The server provides each terminal with a trained speech recognition model. The input is the updated model parameters. The server performs a push notification to each terminal to keep the model up-to-date. The output is each terminal with the latest speech recognition model installed.

[0515] Step 9:

[0516] The device interacts with the user using a new speech recognition model. The input is new speech data from the user, which is then processed by the new model. The output is the improved speech recognition result.

[0517] Step 10:

[0518] The terminal sends user feedback data to the server along with the recognition results. The input consists of the speech recognition results and feedback information, which are sent to the server as a specific action. The output is feedback information sent to the server, which is used as material for further model improvement.

[0519] (Application Example 1)

[0520] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[0521] Speech recognition systems that include regional dialects need to be able to recognize speech with high accuracy even in everyday conversations using those dialects, thereby promoting their use in physical stores and improving customer service. However, improving recognition accuracy to handle diverse dialects remains a challenge. In particular, there is a need to build an environment where speech data can be collected and models updated in real time.

[0522] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[0523] In this invention, the server includes means for using a device that has the function of collecting speech data from speakers with diverse speech patterns, means for providing and updating trained speech recognition algorithms to each device, and means for returning information to an output device using the speech recognition results. This makes it possible to efficiently recognize speech including regional dialects and to immediately provide the results back to the user.

[0524] A "speaker with diverse speech patterns" refers to a speaker who has unique dialects or ways of speaking that differ depending on the region and culture.

[0525] "Audio data" refers to data that records a speaker's utterances in digital format.

[0526] "Device" refers to a device that includes hardware or software used to collect and process audio data.

[0527] A "secure method" refers to a method that uses encryption and authentication techniques to protect audio data from unauthorized access and tampering.

[0528] A "speech recognition algorithm" is a computational method or process for generating text from speech data.

[0529] "Response" refers to the feedback information provided by the user based on the results of speech recognition.

[0530] "Information" refers to the content and data output based on the results obtained through speech recognition.

[0531] This invention aims to construct a speech recognition system that can handle a variety of dialects. The system is primarily implemented through three elements: a server, a terminal, and a user.

[0532] The server performs a series of processes, from collecting audio data to training and updating recognition algorithms. Specifically, it processes the collected audio data in a secure manner and trains speech recognition algorithms. The audio data is used to generate region-specific models.

[0533] The terminal collects voice data through interaction with the user and transmits it to the server. Appropriate hardware and software are used for voice data collection, such as smart devices with microphones or personal computers. The voice data transmitted from the terminal to the server is securely transmitted via the internet to enable real-time speech recognition.

[0534] The user communicates with the system by speaking, including in dialect, using a terminal. The terminal collects the user's speech, encrypts it, and sends it to the server. The server trains a speech recognition model, returns the trained model to the terminal, and improves the accuracy of speech recognition.

[0535] As a concrete example, imagine a scenario where a user asks for tourist information in their local dialect at a terminal installed in a commercial facility in a certain region. In this case, the system receives the user's speech, performs voice recognition, and immediately returns appropriate tourist information to the user's smartphone.

[0536] One example of how a generative AI model can be used is a prompt such as, "Please tell me tourist information in the Okinawan dialect." This input allows the server to generate an audio guide and provide the user with the most relevant information.

[0537] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[0538] Step 1:

[0539] The device acquires voice input from the user via a microphone. The acquired voice is recorded as digital audio data. This data undergoes noise reduction processing to reduce background noise. The noise-reduced audio data then becomes the input for the next step.

[0540] Step 2:

[0541] The terminal segments the denoised audio data and extracts the audio features contained in each segment. An acoustic analysis algorithm is used for this process. Based on the audio features, a provisional label is assigned to the audio data. The labeled data is then transmitted to the server via a secure communication protocol.

[0542] Step 3:

[0543] The server receives audio data transmitted from the terminal and uses it to train the speech recognition algorithm. Supervised learning is applied to the training to improve the ability to identify diverse regional speech features. The output of this step is the updated speech recognition algorithm.

[0544] Step 4:

[0545] The server distributes the trained speech recognition algorithm to the terminal. The terminal incorporates the distributed algorithm into its processing mechanism and prepares for the next speech recognition. This update enables more accurate speech recognition.

[0546] Step 5:

[0547] The user accesses the device again and performs voice input. The device utilizes a newly trained speech recognition algorithm to recognize the user's voice. Based on the recognition result, it returns appropriate information to the user. This output is provided to the user as either voice or text.

[0548] Step 6:

[0549] The device collects user feedback on recognition accuracy and sends it to the server. The server uses this feedback to further improve the model. It analyzes the collected feedback data and updates the algorithm as needed.

[0550] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[0551] This invention provides a system for collecting speech data from speakers with diverse speech patterns and performing speech recognition and emotion recognition. The system has the function of analyzing user speech and improving the user experience. It consists of three main components: the user, the terminal, and the server.

[0552] First, the user begins a conversation into a terminal installed in a public facility or home. Once the conversation begins, the terminal records the audio in real time and uses it for speech recognition and emotion recognition. Here, the terminal divides the audio data into segments, and each segment is labeled based on its acoustic characteristics.

[0553] The device further uses an emotion engine to identify emotions from the user's voice. This emotion recognition improves accuracy not only by using acoustic features but also by combining it with facial expression data from the camera as needed. The results of emotion recognition and voice recognition are integrated, making it possible to understand the user's state in detail.

[0554] Voice and emotion data are securely transmitted to the server via encrypted communication. The server uses this data to train speech recognition and emotion recognition models. In particular, improvements in recognition accuracy are made to accommodate regional dialects and speech patterns.

[0555] Trained speech recognition and emotion recognition models are deployed from the server to each terminal. This allows the terminal to interact with the user using the latest speech and emotion recognition models.

[0556] As a concrete example, a user speaking the Kyoto dialect converses using a terminal installed in a public facility. The terminal collects and segments voice data in real time, and in the background, an emotion engine analyzes the user's emotions from the tone and speed of their voice. For example, if an emotion expressing joy is detected, that information is sent to the server along with the utterance and used to improve the recognition model. Ultimately, the recognition rate improves, and the quality of the user experience is enhanced.

[0557] In this way, this system can bridge the gap in voice and emotion recognition between regions and improve various user experiences.

[0558] The following describes the processing flow.

[0559] Step 1:

[0560] The user initiates a conversation with the installed terminal. The terminal obtains permission to begin recording immediately upon the start of the conversation.

[0561] Step 2:

[0562] The device divides the recorded audio into segments in real time. The segmented audio is then tentatively labeled based on specific speech intervals.

[0563] Step 3:

[0564] The device extracts acoustic features from the audio and uses an emotion engine to identify the user's emotions. In this process, it captures not only audio but also facial expression data with a camera and includes it in the analysis.

[0565] Step 4:

[0566] The device encrypts both voice and emotional data and transmits them to the server via a secure network. This ensures that the data is managed in a privacy-protected manner.

[0567] Step 5:

[0568] The server stores the received audio data and emotion data in a database. During storage, it adds relevant information such as the speech label and emotion identification result to each data.

[0569] Step 6:

[0570] The server uses the collected data to train speech recognition and emotion recognition models. It aims to improve recognition accuracy by learning region-specific speech patterns and emotional expressions.

[0571] Step 7:

[0572] The server deploys newly trained speech recognition and emotion recognition models to each terminal. The terminal receives this update and makes the latest recognition algorithms available.

[0573] Step 8:

[0574] Users can provide feedback on the speech recognition and emotion recognition results provided by the device, including evaluations and suggestions for improvement.

[0575] Step 9:

[0576] The device collects user feedback and periodically sends it to the server. This feedback is used to further improve the model.

[0577] Step 10:

[0578] The server readjusts the speech recognition and emotion recognition models based on feedback and retrains them as needed. This improves the accuracy of the recognition system and user satisfaction.

[0579] (Example 2)

[0580] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[0581] Conventional speech recognition systems struggle to accurately recognize diverse speech characteristics and emotions, resulting in a reduced user experience. Furthermore, there is a need to improve recognition accuracy to accommodate regionally specific dialects and speech patterns. This invention aims to solve these problems and provide a richer user experience by improving the accuracy of speech and emotion recognition.

[0582] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[0583] In this invention, the server includes means for using a device that has the function of collecting speech information from speech sources with diverse utterances, means for securely transmitting and storing the collected speech information, and means for training a virtual model for speech recognition and emotion recognition using the stored speech information. This makes it possible to develop a model that can effectively recognize diverse speech characteristics and is compatible with regionally specific dialects and speech patterns.

[0584] "Diverse speech" refers to speech that exhibits a variety of different phonetic patterns and linguistic characteristics.

[0585] A "sound source" is a subject that generates sound, and the concept includes sounds emitted by humans and machines.

[0586] "Audio information" refers to audio data and its characteristics obtained from an audio source, and includes data such as the waveform and frequency components of the audio.

[0587] "Device" refers to an interface equipped with hardware and software that has the function of collecting, processing, and transmitting audio information.

[0588] "Secure transmission" refers to methods for safely transferring voice information without disclosing it to third parties, and includes communication using encryption technology.

[0589] "Storage" refers to the act of securely saving acquired audio information in a database or storage device so that it can be used later.

[0590] A "virtual model" is a set of algorithms and their parameters built on audio information, and is a computerized simulation that enables speech and emotion recognition.

[0591] "Supervised learning" is a technique for training a virtual model using known input-output pairs to improve the accuracy of predictions on new data.

[0592] "Evaluation" refers to the act of collecting user feedback and quality information regarding the results of speech recognition and emotion recognition.

[0593] "Improvement" is the process of enhancing the accuracy and functionality of a virtual model based on the collected evaluation data.

[0594] This invention provides a speech recognition and emotion recognition system for accurately recognizing diverse speech characteristics and emotions. The system mainly consists of three components: a user, a terminal, and a server.

[0595] Users make voice inputs to terminals installed in public facilities or homes. The user's speech is captured through a microphone built into the terminal. In this process, the terminal is required to collect voice information in real time. For example, using a voice processing chip with noise reduction capabilities allows for the acquisition of cleaner voice data.

[0596] The device first converts the collected audio information into a digital format and extracts acoustic features. Here, using audio signal processing libraries such as Librosa, features such as Mel-frequency cepstrum coefficients (MFCCs) can be obtained from the audio waveform. The device also segments the audio and assigns an identification label to each segment. Furthermore, if necessary, the camera is used to acquire user facial expression data, which is then used to improve the accuracy of emotion recognition.

[0597] The device then encrypts the voice and emotional data and securely transfers it to the server. Secure communication methods such as SSL / TLS protocols are used for communication.

[0598] The server uses the received data to train a virtual model. Specifically, machine learning frameworks such as TensorFlow are used, and the model parameters are adjusted based on supervised learning. By utilizing datasets specific to each region and speech pattern, the accuracy of speech recognition and emotion recognition is continuously improved.

[0599] Once the training is complete, the model is deployed from the server to the terminal, enabling highly accurate recognition in subsequent interactions.

[0600] For example, if a user says to the device, "Hello, tell me today's news," the voice is analyzed in real time, and emotions such as excitement and interest are recognized from the tone and speed of the user's voice, and this information is sent to the server. As a result, the system can provide feedback tailored to the user's needs.

[0601] An example of a prompt to input into a generative AI model is, "Identify what emotions the user is showing during the conversation." Using this prompt allows the system to perform more precise emotion recognition.

[0602] The flow of the specific processing in Example 2 will be explained using Figure 13.

[0603] Step 1:

[0604] The user speaks into the device's microphone. The input is the user's voice, and the output is digital audio data. The microphone built into the device converts the voice into a digital signal. A noise reduction filter is applied during this process to obtain clear audio data.

[0605] Step 2:

[0606] The terminal segments the acquired audio data and extracts acoustic features. The input is digital audio data, and the output is a set of acoustic features. Using an audio signal processing library such as Librosa, the audio data is divided into short time segments, and features such as Mel-frequency cepstrum coefficients (MFCCs) are extracted from each segment.

[0607] Step 3:

[0608] The device assigns an identification label to each audio segment and analyzes emotions using facial expression data acquired by the camera as needed. The input is an acoustic feature set and facial expression data, and the output is the emotion recognition result. The emotion engine evaluates the tone and pitch of the audio and integrates it with the facial expression analysis to determine the emotion label.

[0609] Step 4:

[0610] The terminal encrypts voice and emotion data and sends it to the server. The input is the voice recognition result and emotion recognition result, and the output is an encrypted data stream. Secure communication is performed using the SSL / TLS protocol, and the data is transferred to the server.

[0611] Step 5:

[0612] The server trains a virtual model using the received speech and sentiment data. The input is speech and sentiment data, and the output is an improved virtual model. Supervised learning is performed using TensorFlow to update the model parameters and improve its adaptability to regional dialects and speech patterns.

[0613] Step 6:

[0614] The server deploys the latest trained virtual model to the terminals. The input is the updated virtual model, and the output is the recognition model deployed on each terminal. Through the management platform, the model is installed on each terminal and becomes available for the next recognition process.

[0615] Step 7:

[0616] We collect user feedback and regularly evaluate and improve the accuracy of the model. The input is user feedback, and the output is the improved recognition model. Based on the user experience, the model is tuned, and the overall system performance is optimized.

[0617] (Application Example 2)

[0618] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as the "server," and the headset-type terminal 314 will be referred to as the "terminal."

[0619] In systems that recognize diverse voices and emotions, there is a need for methods to improve user interaction in real time and provide a better experience. This requires not only improved accuracy in voice and emotion recognition, but also a system capable of providing immediate feedback.

[0620] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means.

[0621] In this invention, the server includes means for securely transmitting and storing voice data, means for training voice recognition and emotion recognition models, and means for displaying real-time feedback based on emotion data collected during the conversation. This allows the system to reflect the user's emotional state in real time while they are speaking, enabling a quick and adaptive response.

[0622] A "device" is hardware used to collect diverse speech and securely transmit and store audio data.

[0623] "Audio data" refers to digital information of sounds collected from speakers, which is used to train speech recognition and emotion recognition models.

[0624] A "speech recognition model" is an algorithm that analyzes audio data and converts the speaker's words into text.

[0625] An "emotion recognition model" is an algorithm that analyzes the characteristics of voice and facial expressions to determine the emotional state of a speaker.

[0626] "Feedback" refers to user opinions and evaluations regarding the accuracy of speech recognition and emotion recognition, as well as the user experience of the system.

[0627] "Displaying feedback in real time" means instantly presenting the results of the user's speech and emotional analysis via a display or similar device.

[0628] The system that realizes this invention consists of three main components: a terminal, a server, and a user.

[0629] The server utilizes a high-speed database and encrypted communication protocols for the secure storage and processing of speech and emotion data. Cloud-based computing resources and machine learning algorithms (such as TensorFlow and PyTorch) are used to train speech recognition and emotion recognition models, and the models are improved to accommodate regional dialects and emotional expressions.

[0630] The device is equipped with a microphone and camera, which are used to collect the user's voice and facial expressions in real time. The voice data is converted into text data using the Google Cloud Speech-to-Text API, and the Emotion Recognition API is used for emotion recognition. The device combines these APIs to perform voice and emotion recognition and has the ability to provide real-time feedback to the user. The feedback is displayed on the device's display or connected smart glasses, enabling the user to respond immediately.

[0631] For example, when a customer smiles and says, "I really like the design of this product," the device recognizes the voice and determines that the emotion is "joy." This information is displayed in real time, allowing staff to introduce similar products to the customer as the next topic.

[0632] An example of a prompt when using a generative AI model is: "The customer has recently given positive feedback on a product. Use smart glasses to recognize the voice and emotion and provide feedback that suggests the customer is satisfied."

[0633] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[0634] Step 1:

[0635] The device uses a microphone and camera to collect the user's voice and facial expressions in real time. The input consists of the user's speech data and associated image data, which are converted into a digital format and passed on to the next processing step. The device digitizes the audio signal from the microphone and captures still image data from the camera at regular intervals.

[0636] Step 2:

[0637] The device sends the collected audio data to the Google Cloud Speech-to-Text API, where it is converted into linguistic data. The input is digitized audio data, which is then parsed and output as text data. This API call yields a set of words extracted from the audio.

[0638] Step 3:

[0639] The device sends facial expression data to the Emotion Recognition API to determine the user's emotions. The input is captured image data of facial expressions, and the output is the identified emotion label. Emotions are identified through image analysis, recognizing emotions such as happiness, sadness, and anger.

[0640] Step 4:

[0641] The device provides real-time feedback to the user on analyzed text and sentiment data. The input is the result of steps 2 and 3, and the output is information processed into a user-friendly format. The device then displays feedback such as "The customer is happy" on smart glasses or a display.

[0642] Step 5:

[0643] The server collects and stores voice and emotion data via a secure communication protocol. Input is data transmitted from the terminal, and output is an encrypted dataset. It communicates with a database to store the dataset.

[0644] Step 6:

[0645] The server updates the generative AI model using the collected data to improve the accuracy of speech and emotion recognition. The input is the stored dataset, and the output is the updated recognition model. Machine learning algorithms are used to retrain the model with the data, thereby improving the model's accuracy.

[0646] The specific processing unit 290 transmits the result of the specific processing to the headset terminal 314. In the headset terminal 314, the control unit 46A causes the speaker 240 and display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 238 to the data processing unit 12. In the data processing unit 12, the specific processing unit 290 acquires the audio data.

[0647] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0648] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and specific processing may also be performed by the headset terminal 314.

[0649] [Fourth Embodiment]

[0650] Figure 7 shows an example of the configuration of the data processing system 410 according to the fourth embodiment.

[0651] As shown in Figure 7, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0652] The data processing device 12 comprises a computer 22, a database 24, and a communication interface 26. The computer 22 is an example of a "computer" related to the technology of this disclosure. The computer 22 comprises a processor 28, RAM 30, and storage 32. The processor 28, RAM 30, and storage 32 are connected to a bus 34. The database 24 and the communication interface 26 are also connected to the bus 34. The communication interface 26 is connected to a network 54. An example of the network 54 is a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0653] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication interface 44, and a controlled object 443. The computer 36 includes a processor 46, RAM 48, and storage 50. The processor 46, RAM 48, and storage 50 are connected to a bus 52. The microphone 238, speaker 240, camera 42, and controlled object 443 are also connected to the bus 52.

[0654] The microphone 238 receives voice signals from the user 20 and receives instructions from the user 20. The microphone 238 captures the voice signals from the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to the instructions from the processor 46.

[0655] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an image sensor such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the area around the user 20 (for example, an imaging range defined by a field of view equivalent to the width of a typical healthy person's field of vision).

[0656] Communication interface 44 is connected to network 54. Communication interfaces 44 and 26 are responsible for the exchange of various information between processor 46 and processor 28 via network 54. The exchange of various information between processor 46 and processor 28 using communication interfaces 44 and 26 is performed in a secure manner.

[0657] The controlled object 443 includes a display device, LEDs in the eyes, and motors that drive the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. Furthermore, the robot 414's facial expressions can also be expressed by controlling the illumination state of the LEDs in its eyes.

[0658] Figure 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Figure 8, the data processing device 12 performs specific processing using the processor 28. The storage 32 stores the specific processing program 56.

[0659] The specific processing program 56 is an example of a "program" relating to the technology of this disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0660] The storage 32 stores the data generation model 58 and the emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0661] In robot 414, the processor 46 performs the reception output processing. The storage 50 stores the reception output program 60. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output processing is realized by the processor 46 operating as a control unit 46A according to the reception output program 60 executed on the RAM 48.

[0662] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[0663] This invention provides a method for collecting and processing voice data using terminals installed throughout Japan, as a voice recognition system that includes regionally specific dialects. This system consists of three components: a user, a terminal, and a server.

[0664] First, the user engages in a casual conversation with a nearby device. After obtaining necessary consent, the audio data collected during the conversation is recorded and segmented in real time by the device. Audio features are extracted, and provisional labels are assigned according to the type of dialect. The audio data undergoes noise reduction processing to remove unwanted noise, and the processed data is then encrypted and stored.

[0665] Next, the server receives the audio data transmitted from the terminal. The server analyzes this data and uses it as training data for a speech recognition model. Using supervised learning techniques, the server generates and updates the speech recognition model based on the audio data for each region. This speech recognition model is adjusted to have the ability to identify region-specific speech features.

[0666] Trained speech recognition models are automatically distributed from the server to each terminal. This ensures that terminals always perform speech recognition based on the latest model, improving recognition accuracy in each region. Furthermore, through user interaction, terminals provide the server with speech recognition results and user feedback. The server then retrains the model based on the collected feedback to further improve accuracy.

[0667] As a concrete example, a user speaking the local dialect engages in everyday conversation using a terminal installed in a public facility in a certain area. This audio is collected in real time and sent to a server as data with unique regional vocal characteristics. The server learns from this data and, for example, updates its speech recognition algorithm to one that is specialized for a particular dialect. The latest model is deployed to the terminal, enabling even higher accuracy the next time speech recognition is performed on that terminal.

[0668] In this way, this system can support diverse regional dialects and eliminate regional disparities in speech recognition technology.

[0669] The following describes the processing flow.

[0670] Step 1:

[0671] Users consent to the collection of voice data and engage in everyday conversations through the device. They can engage in natural conversations, including those using dialects, in public facilities and home environments where the device is installed.

[0672] Step 2:

[0673] The device begins recording the user's conversation in real time. The audio data is segmented by detecting breaks in speech, and each segment is assigned a temporary label.

[0674] Step 3:

[0675] The device then applies preprocessing, such as noise reduction, to the segmented audio data. This preprocessing improves sound quality and prepares data suitable for training the speech recognition model.

[0676] Step 4:

[0677] The device encrypts the processed voice data and sends it to the server via a secure communication protocol. The data is protected with the utmost consideration for user privacy.

[0678] Step 5:

[0679] The server saves the received audio data to a database. During saving, dialect information and speech characteristic labels associated with the audio data are also recorded.

[0680] Step 6:

[0681] The server uses stored audio data to train a speech recognition model. Using supervised learning, the speech recognition model learns region-specific speech features and improves the accuracy of speech recognition.

[0682] Step 7:

[0683] The server deploys the newly trained speech recognition model to each terminal. The terminal downloads this new model and uses it for speech recognition processing.

[0684] Step 8:

[0685] Users can provide feedback on the speech recognition results from their devices. This user feedback is valuable information for improving the speech recognition results.

[0686] Step 9:

[0687] The device periodically sends the collected feedback to the server. This feedback is used to consider whether to retrain the model.

[0688] Step 10:

[0689] The server analyzes the feedback data and adjusts the parameters of the speech recognition model as needed. This further improves speech recognition accuracy in each region and enhances the overall system performance.

[0690] (Example 1)

[0691] Next, we will describe Example 1. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[0692] In speech recognition technology, there is a need to achieve highly accurate speech recognition while being able to handle diverse, region-specific speech characteristics. However, conventional speech recognition systems have the problem of failing to adequately handle regional pronunciations and dialects, resulting in reduced recognition accuracy. Furthermore, there is a lack of mechanisms to effectively incorporate user feedback and update the models, making continuous model improvement difficult.

[0693] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 1 is realized by the following means.

[0694] In this invention, the server includes: an information device means having the function of acquiring voice information from users with diverse voices; means for securely transmitting and storing the acquired voice information; and means for collecting user feedback regarding the accuracy of speech recognition. This enables rapid and accurate learning of various region-specific voice characteristics, and improves and continuously refines the accuracy of the speech recognition model.

[0695] "Diverse speech" refers to the vocalizations of multiple speakers with different pronunciations, dialects, and accents.

[0696] A "user" refers to an individual who interacts with the system and provides voice information through this system.

[0697] "Voice information" refers to data that is acquired and stored in digital format from speech uttered by users.

[0698] An "information device" is a device used to acquire, process, and transmit voice information, and includes components such as microphones and processors.

[0699] A "speech recognition model" is an algorithm or system built to analyze speech information and convert it into text or specific commands.

[0700] "Feedback" refers to opinions from users regarding the results of speech recognition, their impressions, and suggestions for improvement, which are used to improve the system.

[0701] "Supervised learning" is a method of training machine learning models using labeled datasets, which are then used to enable the models to make accurate predictions and classifications.

[0702] This section describes an embodiment of this system. This system aims to address diverse, region-specific speech characteristics using speech recognition technology. The system primarily consists of three elements: the user, the terminal, and the server.

[0703] Users transmit audio through the installed terminals. For example, when a user talks about a local landmark, the audio is collected by the terminal. This audio information is naturally gathered from the user's everyday conversations.

[0704] The device records the user's voice in real time and segments it as digital audio data. This device is equipped with a microphone and signal processing unit for extracting audio features. Features are extracted from the recorded audio data using techniques such as Mel-frequency cepstrum coefficients (MFCC). Noise reduction is also performed using digital signal processing to obtain clear audio data. Initial labels are then assigned based on the acoustic characteristics. The audio data is then encrypted using the AES (Advanced Encryption Standard) protocol and stored securely.

[0705] The server receives data transmitted from terminals and uses it to train speech recognition models. It utilizes guided learning techniques such as deep learning to learn the speech characteristics of specific regions and improve recognition accuracy. The trained speech recognition models are pushed to each terminal, allowing them to efficiently perform recognition tasks in their most up-to-date state. The server also collects user feedback and continuously improves the models based on it.

[0706] As a concrete example, when a user uses a device to speak in the local dialect, the audio is collected and analyzed by a server. Through this process, the speech recognition system becomes familiar with the local dialect and can improve its recognition accuracy.

[0707] An example of a prompt in a generative AI model is, "Update the speech recognition model using speech data collected from a specific region for a speech recognition system that identifies regional dialects."

[0708] In this way, this system can recognize voices from diverse regions with high accuracy, contributing to the elimination of regional disparities in speech recognition.

[0709] The flow of the specific processing in Example 1 will be explained using Figure 11.

[0710] Step 1:

[0711] The user begins a casual conversation into a terminal installed in their local area. The input information is the user's speech, which is captured as audio data by the terminal's high-sensitivity microphone. A specific example of this operation would be the user uttering something like, "I'm going to talk about the local festival." The output is raw audio data.

[0712] Step 2:

[0713] The terminal segments the acquired audio data in real time. The input is audio data from the user, and the digital signal processing unit divides this data so that it can be easily managed. Specifically, it divides the audio into segments of regular time intervals and saves each segment as a separate file. The output is a collection of segmented audio files.

[0714] Step 3:

[0715] The device extracts features from audio data. The input is a segmented audio file, from which acoustic features are extracted using techniques such as Mel-frequency cepstrum coefficients (MFCCs). Specifically, it analyzes the frequency and temporal characteristics of each audio segment and converts them into numerical data. The output is a list of the extracted acoustic features.

[0716] Step 4:

[0717] The device applies noise reduction processing to the feature data. The input is feature data, and it performs specific operations to remove noise using digital signal processing. This improves the accuracy of speech recognition. The output is clean speech data with reduced noise.

[0718] Step 5:

[0719] The device assigns a temporary label to the noise-reduced audio data, encrypts it, and saves it. The input is noise-reduced audio data. Specifically, a temporary label appropriate to the audio content is assigned, the data is encrypted using the AES protocol, and then saved. The output is labeled and encrypted audio data.

[0720] Step 6:

[0721] The terminal sends encrypted audio data to the server. The input is labeled encrypted data. The terminal uses the TLS protocol to perform the specific action of securely communicating and sending the data to the server. The output is the audio data securely transferred to the server.

[0722] Step 7:

[0723] The server trains a speech recognition model using the received audio data. The input is audio data sent from the terminal, and the model is optimized using deep learning techniques. Specifically, it learns new acoustic features and updates parameters to improve accuracy. The output is the new model parameters.

[0724] Step 8:

[0725] The server provides each terminal with a trained speech recognition model. The input is the updated model parameters. The server performs a push notification to each terminal to keep the model up-to-date. The output is each terminal with the latest speech recognition model installed.

[0726] Step 9:

[0727] The device interacts with the user using a new speech recognition model. The input is new speech data from the user, which is then processed by the new model. The output is the improved speech recognition result.

[0728] Step 10:

[0729] The terminal sends user feedback data to the server along with the recognition results. The input consists of the speech recognition results and feedback information, which are sent to the server as a specific action. The output is feedback information sent to the server, which is used as material for further model improvement.

[0730] (Application Example 1)

[0731] Next, we will explain Application Example 1. In the following explanation, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[0732] Speech recognition systems that include regional dialects need to be able to recognize speech with high accuracy even in everyday conversations using those dialects, thereby promoting their use in physical stores and improving customer service. However, improving recognition accuracy to handle diverse dialects remains a challenge. In particular, there is a need to build an environment where speech data can be collected and models updated in real time.

[0733] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 1 is realized by the following means.

[0734] In this invention, the server includes means for using a device that has the function of collecting speech data from speakers with diverse speech patterns, means for providing and updating trained speech recognition algorithms to each device, and means for returning information to an output device using the speech recognition results. This makes it possible to efficiently recognize speech including regional dialects and to immediately provide the results back to the user.

[0735] A "speaker with diverse speech patterns" refers to a speaker who has unique dialects or ways of speaking that differ depending on the region and culture.

[0736] "Audio data" refers to data that records a speaker's utterances in digital format.

[0737] "Device" refers to a device that includes hardware or software used to collect and process audio data.

[0738] A "secure method" refers to a method that uses encryption and authentication techniques to protect audio data from unauthorized access and tampering.

[0739] A "speech recognition algorithm" is a computational method or process for generating text from speech data.

[0740] "Response" refers to the feedback information provided by the user based on the results of speech recognition.

[0741] "Information" refers to the content and data output based on the results obtained through speech recognition.

[0742] This invention aims to construct a speech recognition system that can handle a variety of dialects. The system is primarily implemented through three elements: a server, a terminal, and a user.

[0743] The server performs a series of processes, from collecting audio data to training and updating recognition algorithms. Specifically, it processes the collected audio data in a secure manner and trains speech recognition algorithms. The audio data is used to generate region-specific models.

[0744] The terminal collects voice data through interaction with the user and transmits it to the server. Appropriate hardware and software are used for voice data collection, such as smart devices with microphones or personal computers. The voice data transmitted from the terminal to the server is securely transmitted via the internet to enable real-time speech recognition.

[0745] The user communicates with the system by speaking, including in dialect, using a terminal. The terminal collects the user's speech, encrypts it, and sends it to the server. The server trains a speech recognition model, returns the trained model to the terminal, and improves the accuracy of speech recognition.

[0746] As a concrete example, imagine a scenario where a user asks for tourist information in their local dialect at a terminal installed in a commercial facility in a certain region. In this case, the system receives the user's speech, performs voice recognition, and immediately returns appropriate tourist information to the user's smartphone.

[0747] One example of how a generative AI model can be used is a prompt such as, "Please tell me tourist information in the Okinawan dialect." This input allows the server to generate an audio guide and provide the user with the most relevant information.

[0748] The flow of a specific process in Application Example 1 will be explained using Figure 12.

[0749] Step 1:

[0750] The device acquires voice input from the user via a microphone. The acquired voice is recorded as digital audio data. This data undergoes noise reduction processing to reduce background noise. The noise-reduced audio data then becomes the input for the next step.

[0751] Step 2:

[0752] The terminal segments the denoised audio data and extracts the audio features contained in each segment. An acoustic analysis algorithm is used for this process. Based on the audio features, a provisional label is assigned to the audio data. The labeled data is then transmitted to the server via a secure communication protocol.

[0753] Step 3:

[0754] The server receives audio data transmitted from the terminal and uses it to train the speech recognition algorithm. Supervised learning is applied to the training to improve the ability to identify diverse regional speech features. The output of this step is the updated speech recognition algorithm.

[0755] Step 4:

[0756] The server distributes the trained speech recognition algorithm to the terminal. The terminal incorporates the distributed algorithm into its processing mechanism and prepares for the next speech recognition. This update enables more accurate speech recognition.

[0757] Step 5:

[0758] The user accesses the device again and performs voice input. The device utilizes a newly trained speech recognition algorithm to recognize the user's voice. Based on the recognition result, it returns appropriate information to the user. This output is provided to the user as either voice or text.

[0759] Step 6:

[0760] The device collects user feedback on recognition accuracy and sends it to the server. The server uses this feedback to further improve the model. It analyzes the collected feedback data and updates the algorithm as needed.

[0761] Furthermore, an emotion engine that estimates the user's emotions may be incorporated. That is, the identification processing unit 290 may use the emotion identification model 59 to estimate the user's emotions and perform identification processing using the user's emotions.

[0762] This invention provides a system for collecting speech data from speakers with diverse speech patterns and performing speech recognition and emotion recognition. The system has the function of analyzing user speech and improving the user experience. It consists of three main components: the user, the terminal, and the server.

[0763] First, the user begins a conversation into a terminal installed in a public facility or home. Once the conversation begins, the terminal records the audio in real time and uses it for speech recognition and emotion recognition. Here, the terminal divides the audio data into segments, and each segment is labeled based on its acoustic characteristics.

[0764] The device further uses an emotion engine to identify emotions from the user's voice. This emotion recognition improves accuracy not only by using acoustic features but also by combining it with facial expression data from the camera as needed. The results of emotion recognition and voice recognition are integrated, making it possible to understand the user's state in detail.

[0765] Voice and emotion data are securely transmitted to the server via encrypted communication. The server uses this data to train speech recognition and emotion recognition models. In particular, improvements in recognition accuracy are made to accommodate regional dialects and speech patterns.

[0766] Trained speech recognition and emotion recognition models are deployed from the server to each terminal. This allows the terminal to interact with the user using the latest speech and emotion recognition models.

[0767] As a concrete example, a user speaking the Kyoto dialect converses using a terminal installed in a public facility. The terminal collects and segments voice data in real time, and in the background, an emotion engine analyzes the user's emotions from the tone and speed of their voice. For example, if an emotion expressing joy is detected, that information is sent to the server along with the utterance and used to improve the recognition model. Ultimately, the recognition rate improves, and the quality of the user experience is enhanced.

[0768] In this way, this system can bridge the gap in voice and emotion recognition between regions and improve various user experiences.

[0769] The following describes the processing flow.

[0770] Step 1:

[0771] The user initiates a conversation with the installed terminal. The terminal obtains permission to begin recording immediately upon the start of the conversation.

[0772] Step 2:

[0773] The device divides the recorded audio into segments in real time. The segmented audio is then tentatively labeled based on specific speech intervals.

[0774] Step 3:

[0775] The device extracts acoustic features from the audio and uses an emotion engine to identify the user's emotions. In this process, it captures not only audio but also facial expression data with a camera and includes it in the analysis.

[0776] Step 4:

[0777] The device encrypts both voice and emotional data and transmits them to the server via a secure network. This ensures that the data is managed in a privacy-protected manner.

[0778] Step 5:

[0779] The server stores the received audio data and emotion data in a database. During storage, it adds relevant information such as the speech label and emotion identification result to each data.

[0780] Step 6:

[0781] The server uses the collected data to train speech recognition and emotion recognition models. It aims to improve recognition accuracy by learning region-specific speech patterns and emotional expressions.

[0782] Step 7:

[0783] The server deploys newly trained speech recognition and emotion recognition models to each terminal. The terminal receives this update and makes the latest recognition algorithms available.

[0784] Step 8:

[0785] Users can provide feedback on the speech recognition and emotion recognition results provided by the device, including evaluations and suggestions for improvement.

[0786] Step 9:

[0787] The device collects user feedback and periodically sends it to the server. This feedback is used to further improve the model.

[0788] Step 10:

[0789] The server readjusts the speech recognition and emotion recognition models based on feedback and retrains them as needed. This improves the accuracy of the recognition system and user satisfaction.

[0790] (Example 2)

[0791] Next, we will describe Example 2. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[0792] Conventional speech recognition systems struggle to accurately recognize diverse speech characteristics and emotions, resulting in a reduced user experience. Furthermore, there is a need to improve recognition accuracy to accommodate regionally specific dialects and speech patterns. This invention aims to solve these problems and provide a richer user experience by improving the accuracy of speech and emotion recognition.

[0793] The identification process performed by the identification processing unit 290 of the data processing device 12 in Example 2 is realized by the following means.

[0794] In this invention, the server includes means for using a device that has the function of collecting speech information from speech sources with diverse utterances, means for securely transmitting and storing the collected speech information, and means for training a virtual model for speech recognition and emotion recognition using the stored speech information. This makes it possible to develop a model that can effectively recognize diverse speech characteristics and is compatible with regionally specific dialects and speech patterns.

[0795] "Diverse speech" refers to speech that exhibits a variety of different phonetic patterns and linguistic characteristics.

[0796] A "sound source" is a subject that generates sound, and the concept includes sounds emitted by humans and machines.

[0797] "Audio information" refers to audio data and its characteristics obtained from an audio source, and includes data such as the waveform and frequency components of the audio.

[0798] "Device" refers to an interface equipped with hardware and software that has the function of collecting, processing, and transmitting audio information.

[0799] "Secure transmission" refers to methods for safely transferring voice information without disclosing it to third parties, and includes communication using encryption technology.

[0800] "Storage" refers to the act of securely saving acquired audio information in a database or storage device so that it can be used later.

[0801] A "virtual model" is a set of algorithms and their parameters built on audio information, and is a computerized simulation that enables speech and emotion recognition.

[0802] "Supervised learning" is a technique for training a virtual model using known input-output pairs to improve the accuracy of predictions on new data.

[0803] "Evaluation" refers to the act of collecting user feedback and quality information regarding the results of speech recognition and emotion recognition.

[0804] "Improvement" is the process of enhancing the accuracy and functionality of a virtual model based on the collected evaluation data.

[0805] This invention provides a speech recognition and emotion recognition system for accurately recognizing diverse speech characteristics and emotions. The system mainly consists of three components: a user, a terminal, and a server.

[0806] Users make voice inputs to terminals installed in public facilities or homes. The user's speech is captured through a microphone built into the terminal. In this process, the terminal is required to collect voice information in real time. For example, using a voice processing chip with noise reduction capabilities allows for the acquisition of cleaner voice data.

[0807] The device first converts the collected audio information into a digital format and extracts acoustic features. Here, using audio signal processing libraries such as Librosa, features such as Mel-frequency cepstrum coefficients (MFCCs) can be obtained from the audio waveform. The device also segments the audio and assigns an identification label to each segment. Furthermore, if necessary, the camera is used to acquire user facial expression data, which is then used to improve the accuracy of emotion recognition.

[0808] The device then encrypts the voice and emotional data and securely transfers it to the server. Secure communication methods such as SSL / TLS protocols are used for communication.

[0809] The server uses the received data to train a virtual model. Specifically, machine learning frameworks such as TensorFlow are used, and the model parameters are adjusted based on supervised learning. By utilizing datasets specific to each region and speech pattern, the accuracy of speech recognition and emotion recognition is continuously improved.

[0810] Once the training is complete, the model is deployed from the server to the terminal, enabling highly accurate recognition in subsequent interactions.

[0811] For example, if a user says to the device, "Hello, tell me today's news," the voice is analyzed in real time, and emotions such as excitement and interest are recognized from the tone and speed of the user's voice, and this information is sent to the server. As a result, the system can provide feedback tailored to the user's needs.

[0812] An example of a prompt to input into a generative AI model is, "Identify what emotions the user is showing during the conversation." Using this prompt allows the system to perform more precise emotion recognition.

[0813] The flow of the specific processing in Example 2 will be explained using Figure 13.

[0814] Step 1:

[0815] The user speaks into the device's microphone. The input is the user's voice, and the output is digital audio data. The microphone built into the device converts the voice into a digital signal. A noise reduction filter is applied during this process to obtain clear audio data.

[0816] Step 2:

[0817] The terminal segments the acquired audio data and extracts acoustic features. The input is digital audio data, and the output is a set of acoustic features. Using an audio signal processing library such as Librosa, the audio data is divided into short time segments, and features such as Mel-frequency cepstrum coefficients (MFCCs) are extracted from each segment.

[0818] Step 3:

[0819] The device assigns an identification label to each audio segment and analyzes emotions using facial expression data acquired by the camera as needed. The input is an acoustic feature set and facial expression data, and the output is the emotion recognition result. The emotion engine evaluates the tone and pitch of the audio and integrates it with the facial expression analysis to determine the emotion label.

[0820] Step 4:

[0821] The terminal encrypts voice and emotion data and sends it to the server. The input is the voice recognition result and emotion recognition result, and the output is an encrypted data stream. Secure communication is performed using the SSL / TLS protocol, and the data is transferred to the server.

[0822] Step 5:

[0823] The server trains a virtual model using the received speech and sentiment data. The input is speech and sentiment data, and the output is an improved virtual model. Supervised learning is performed using TensorFlow to update the model parameters and improve its adaptability to regional dialects and speech patterns.

[0824] Step 6:

[0825] The server deploys the latest trained virtual model to the terminals. The input is the updated virtual model, and the output is the recognition model deployed on each terminal. Through the management platform, the model is installed on each terminal and becomes available for the next recognition process.

[0826] Step 7:

[0827] We collect user feedback and regularly evaluate and improve the accuracy of the model. The input is user feedback, and the output is the improved recognition model. Based on the user experience, the model is tuned, and the overall system performance is optimized.

[0828] (Application Example 2)

[0829] Next, we will explain application example 2. In the following explanation, the data processing device 12 will be referred to as the "server" and the robot 414 as the "terminal".

[0830] In systems that recognize diverse voices and emotions, there is a need for methods to improve user interaction in real time and provide a better experience. This requires not only improved accuracy in voice and emotion recognition, but also a system capable of providing immediate feedback.

[0831] The specific processing performed by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means.

[0832] In this invention, the server includes means for securely transmitting and storing voice data, means for training voice recognition and emotion recognition models, and means for displaying real-time feedback based on emotion data collected during the conversation. This allows the system to reflect the user's emotional state in real time while they are speaking, enabling a quick and adaptive response.

[0833] A "device" is hardware used to collect diverse speech and securely transmit and store audio data.

[0834] "Audio data" refers to digital information of sounds collected from speakers, which is used to train speech recognition and emotion recognition models.

[0835] A "speech recognition model" is an algorithm that analyzes audio data and converts the speaker's words into text.

[0836] An "emotion recognition model" is an algorithm that analyzes the characteristics of voice and facial expressions to determine the emotional state of a speaker.

[0837] "Feedback" refers to user opinions and evaluations regarding the accuracy of speech recognition and emotion recognition, as well as the user experience of the system.

[0838] "Displaying feedback in real time" means instantly presenting the results of the user's speech and emotional analysis via a display or similar device.

[0839] The system that realizes this invention consists of three main components: a terminal, a server, and a user.

[0840] The server utilizes a high-speed database and encrypted communication protocols for the secure storage and processing of speech and emotion data. Cloud-based computing resources and machine learning algorithms (such as TensorFlow and PyTorch) are used to train speech recognition and emotion recognition models, and the models are improved to accommodate regional dialects and emotional expressions.

[0841] The device is equipped with a microphone and camera, which are used to collect the user's voice and facial expressions in real time. The voice data is converted into text data using the Google Cloud Speech-to-Text API, and the Emotion Recognition API is used for emotion recognition. The device combines these APIs to perform voice and emotion recognition and has the ability to provide real-time feedback to the user. The feedback is displayed on the device's display or connected smart glasses, enabling the user to respond immediately.

[0842] For example, when a customer smiles and says, "I really like the design of this product," the device recognizes the voice and determines that the emotion is "joy." This information is displayed in real time, allowing staff to introduce similar products to the customer as the next topic.

[0843] An example of a prompt when using a generative AI model is: "The customer has recently given positive feedback on a product. Use smart glasses to recognize the voice and emotion and provide feedback that suggests the customer is satisfied."

[0844] The flow of a specific process in Application Example 2 will be explained using Figure 14.

[0845] Step 1:

[0846] The device uses a microphone and camera to collect the user's voice and facial expressions in real time. The input consists of the user's speech data and associated image data, which are converted into a digital format and passed on to the next processing step. The device digitizes the audio signal from the microphone and captures still image data from the camera at regular intervals.

[0847] Step 2:

[0848] The device sends the collected audio data to the Google Cloud Speech-to-Text API, where it is converted into linguistic data. The input is digitized audio data, which is then parsed and output as text data. This API call yields a set of words extracted from the audio.

[0849] Step 3:

[0850] The device sends facial expression data to the Emotion Recognition API to determine the user's emotions. The input is captured image data of facial expressions, and the output is the identified emotion label. Emotions are identified through image analysis, recognizing emotions such as happiness, sadness, and anger.

[0851] Step 4:

[0852] The device provides real-time feedback to the user on analyzed text and sentiment data. The input is the result of steps 2 and 3, and the output is information processed into a user-friendly format. The device then displays feedback such as "The customer is happy" on smart glasses or a display.

[0853] Step 5:

[0854] The server collects and stores voice and emotion data via a secure communication protocol. Input is data transmitted from the terminal, and output is an encrypted dataset. It communicates with a database to store the dataset.

[0855] Step 6:

[0856] The server updates the generative AI model using the collected data to improve the accuracy of speech and emotion recognition. The input is the stored dataset, and the output is the updated recognition model. Machine learning algorithms are used to retrain the model with the data, thereby improving the model's accuracy.

[0857] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires audio indicating user input for the result of the specific processing. The control unit 46A transmits the audio data indicating user input acquired by the microphone 238 to the data processing unit 12. In the data processing unit 12, the specific processing unit 290 acquires the audio data.

[0858] Data generation model 58 is a type of so-called generative AI (Artificial Intelligence). One example of data generation model 58 is ChatGPT (Internet search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search) <url: https: gemini.google.com ?hl="ja">Examples of generative AI include the following. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 is input with prompts containing instructions, and with inference data such as audio data representing speech, text data representing text, and image data representing images. The data generation model 58 infers from the input inference data according to the instructions indicated by the prompts, and outputs the inference results in data formats such as audio data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0859] In the above embodiment, an example was given in which specific processing is performed by the data processing device 12, but the technology of this disclosure is not limited thereto, and the specific processing may also be performed by the robot 414.

[0860] Furthermore, the emotion identification model 59, acting as an emotion engine, may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to a specific mapping, which is an emotion map (see Figure 9). Similarly, the emotion identification model 59 may also determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[0861] Figure 9 shows an emotion map 400 in which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. The closer to the center of the concentric circles, the more primitive the emotions are located. Further out of the concentric circles, emotions representing states and actions arising from mental states are located. Emotion is a concept that includes feelings and mental states. On the left side of the concentric circles, emotions that are generally generated from reactions occurring in the brain are located. On the right side of the concentric circles, emotions that are generally induced by situational judgment are located. Above and below the concentric circles, emotions that are generally generated from reactions occurring in the brain and induced by situational judgment are located. In addition, the emotion of "pleasure" is located on the upper side of the concentric circles, and the emotion of "displeasure" is located on the lower side. Thus, in the emotion map 400, multiple emotions are mapped based on the structure in which emotions arise, and emotions that are likely to occur simultaneously are mapped close together.

[0862] These emotions are distributed at the 3 o'clock position on the Emotion Map 400, and usually fluctuate between feelings of security and anxiety. In the right half of the Emotion Map 400, situational awareness takes precedence over internal feelings, resulting in a calm impression.

[0863] The inside of the Emotion Map 400 represents inner thoughts, while the outside represents actions. Therefore, the further you go from the outside of the Emotion Map 400, the more visible (expressed in actions) your emotions become.

[0864] Here, human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, it results in discomfort, and when they approach the ideal, it results in pleasure. Similarly, in robots, cars, motorcycles, etc., emotions can be created based on various balances, such as posture and battery level. When these balances deviate from the ideal, it results in discomfort, and when they approach the ideal, it results in pleasure. The emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on a system for analyzing brain physiological signals of speech emotion recognition and emotion, Tokushima University, doctoral dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map contains emotions belonging to a region called "response," where sensation is dominant. The right half of the emotion map contains emotions belonging to a region called "situation," where situational awareness is dominant.

[0865] The emotion map defines two emotions that promote learning. One is the emotion around the middle of the negative "repentance" and "reflection" on the situation side. In other words, it is when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is the emotion around the positive "desire" on the reaction side. In other words, it is when the robot has positive feelings such as "I want more" or "I want to know more."

[0866] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​representing each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple training data sets, which are combinations of user input and emotion values ​​representing each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions located close together have similar values, as shown in the emotion map 900 in Figure 10. Figure 10 shows an example where multiple emotions such as "reassured," "calm," and "confident" have similar emotion values.

[0867] The above description primarily focuses on the functions of the data processing device 12 in relation to this disclosure. However, the system related to this disclosure is not necessarily implemented on a server. The system related to this disclosure may be implemented as a general information processing system. This disclosure may be implemented, for example, as a software program that runs on a personal computer or as an application that runs on a smartphone. The method related to this disclosure may be provided to users in SaaS (Software as a Service) format.

[0868] In the above embodiment, an example was given in which a specific process is performed by a single computer 22. However, the technology of this disclosure is not limited thereto, and a distributed processing of the specific process may be performed by multiple computers, including computer 22. For example, a data generation model 58 may be provided in an external device of the data processing device 12, and the external device may generate data according to the input data.

[0869] In the above embodiment, an example was given in which the specific processing program 56 is stored in the storage 32, but the technology of this disclosure is not limited thereto. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-temporary storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-temporary storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes specific processing according to the specific processing program 56.

[0870] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[0871] Furthermore, it is not necessary to store the entirety of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store the entirety of the specific processing program 56 in the storage 32; it is acceptable to store only a portion of the specific processing program 56.

[0872] The following types of processors can be used as hardware resources to perform specific processing. Examples of processors include a CPU, a general-purpose processor that functions as a hardware resource to perform specific processing by executing software, i.e., a program. Other examples of processors include dedicated electrical circuits, such as FPGAs (Field-Programmable Gate Arrays), PLDs (Programmable Logic Devices), or ASICs (Application Specific Integrated Circuits), which have circuit configurations specifically designed to perform specific processing. All of these processors have built-in or connected memory, and all of them perform specific processing by using memory.

[0873] The hardware resource that performs a specific process may consist of one of these various processors, or it may consist of a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Alternatively, the hardware resource that performs a specific process may consist of a single processor.

[0874] Examples of configurations using a single processor include, firstly, a configuration in which one or more CPUs and software are combined to form a single processor, and this processor functions as a hardware resource that performs a specific process. Secondly, there is a configuration using a processor that realizes the functions of the entire system, including multiple hardware resources that perform a specific process, on a single IC chip, as exemplified by SoCs (System-on-a-chip). In this way, a specific process is realized using one or more of the above types of processors as hardware resources.

[0875] Furthermore, the hardware structure of these various processors can more specifically utilize electrical circuits that combine circuit elements such as semiconductor devices. Also, the specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps can be deleted, new steps added, or the processing order rearranged, as long as it does not deviate from the main purpose.

[0876] The descriptions and illustrations presented above are detailed explanations of the technical aspects of this disclosure and are merely examples of the technical aspects. For example, the above descriptions of the structure, function, operation, and effect are examples of the structure, function, operation, and effect of the technical aspects of this disclosure. Therefore, it goes without saying that you may delete unnecessary parts, add new elements, or replace elements in the descriptions and illustrations presented above, as long as you do not deviate from the essence of the technical aspects of this disclosure. Furthermore, in order to avoid confusion and facilitate understanding of the technical aspects of this disclosure, explanations of common technical knowledge and the like that do not require special explanation to enable the implementation of the technical aspects of this disclosure have been omitted from the descriptions and illustrations presented above.

[0877] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted as being incorporated by reference.

[0878] The following is further disclosed regarding the embodiments described above.

[0879] (Claim 1)

[0880] [Methods using a terminal that has the function of collecting voice data from speakers with diverse speech patterns,

[0881] [Means for securely transmitting and storing collected audio data,

[0882] [Methods for training a speech recognition model using stored audio data,

[0883] [Means for providing and updating trained speech recognition models to each terminal,

[0884] [Means for collecting user feedback on the accuracy of speech recognition,

[0885] A system including means for improving a speech recognition model based on collected feedback.

[0886] (Claim 2)

[0887] [The system according to claim 1, wherein, when collecting the above-mentioned audio data, the speech is segmented and labels are assigned based on acoustic characteristics.

[0888] (Claim 3)

[0889] [The system according to claim 1, which applies supervised learning to the training of a speech recognition model so that it can identify diverse speech features.

[0890] "Example 1"

[0891] (Claim 1)

[0892] [An information device having the function of acquiring voice information from users with diverse voices,

[0893] [Means for securely transmitting and storing acquired audio information,

[0894] [Means for training a speech recognition model using stored audio information,

[0895] [Means for providing and updating trained speech recognition models to each information device,

[0896] [Means for collecting user feedback on the accuracy of speech recognition,

[0897] A system including means for improving a speech recognition model based on collected feedback.

[0898] (Claim 2)

[0899] [The system according to claim 1, wherein when acquiring the above audio information, the audio is divided and labels are assigned based on the acoustic characteristics.

[0900] (Claim 3)

[0901] [The system according to claim 1, which applies guided learning to train a speech recognition model so that it can identify various speech characteristics.

[0902] "Application Example 1"

[0903] (Claim 1)

[0904] [Methods using a device that has the function of collecting speech data from speakers with diverse speech patterns,

[0905] [Means for transmitting and storing collected audio data in a secure manner,

[0906] [Methods for training a speech recognition algorithm using stored audio data,

[0907] [Means for providing and updating trained speech recognition algorithms to each device,

[0908] [Means for collecting user feedback regarding the accuracy of speech recognition,

[0909] [Means for improving the speech recognition algorithm based on collected responses,

[0910] A system including means for returning information to an output device using speech recognition results.

[0911] (Claim 2)

[0912] [The system according to claim 1, wherein, when collecting the above-mentioned audio data, the speech is segmented and labels are assigned based on acoustic characteristics.

[0913] (Claim 3)

[0914] [The system according to claim 1, wherein supervised learning is applied to the training of a speech recognition algorithm so that it can identify diverse speech features.

[0915] "Example 2 of combining an emotion engine"

[0916] (Claim 1)

[0917] [Means using a device that has the function of collecting speech information from speech sources with diverse speech,

[0918] [Means for securely transmitting and storing collected audio information,

[0919] [Means for training virtual models for speech recognition and emotion recognition using stored speech information,

[0920] [Means for supplying and updating trained virtual models to each device,

[0921] [Means for collecting user evaluations regarding the accuracy of speech recognition and emotion recognition,

[0922] A system including means for improving a virtual model based on collected evaluations.

[0923] (Claim 2)

[0924] [The system according to claim 1, wherein when collecting the above audio information, the audio is categorized and identification information is assigned based on the audio characteristics.

[0925] (Claim 3)

[0926] [The system according to claim 1, which applies supervised learning to enable the identification of diverse speech characteristics in the training of a virtual model.

[0927] "Application example 2 of combining emotional engines"

[0928] (Claim 1)

[0929] [Methods using a device that has the function of collecting speech data from speakers with diverse speech patterns,

[0930] [Means for securely transmitting and storing collected audio data,

[0931] [Means for training speech recognition and emotion recognition models using stored audio data,

[0932] [Means for providing and updating trained speech recognition and emotion recognition models to each device,

[0933] [Means for collecting user feedback regarding the accuracy of speech recognition and emotion recognition,

[0934] [Means for improving speech recognition and emotion recognition models based on collected feedback,

[0935] [A means of displaying real-time feedback based on emotional data collected during the conversation,

[0936] ...

[0937] A system that includes this.

[0938] (Claim 2)

[0939] [The system according to claim 1, wherein, when collecting the above-mentioned audio data, the speech is segmented and labels are assigned based on acoustic characteristics.

[0940] (Claim 3)

[0941] [The system according to claim 1, which applies supervised learning to enable the identification of various speech features in the training of speech recognition and emotion recognition models. [Explanation of Symbols]

[0942] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Devices 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robots< / url:> < / url:> < / url:> < / url:>

Claims

1. A means of using a terminal that has the function of collecting voice data from speakers with diverse speech patterns, Means for securely transmitting and storing collected audio data, A means for training a speech recognition model using stored audio data, A means of providing and updating trained speech recognition models to each terminal, A means of collecting user feedback on the accuracy of speech recognition, A system including means for improving a speech recognition model based on collected feedback.

2. The system according to claim 1, wherein, when collecting the above-mentioned audio data, the utterances are segmented and labels are assigned based on acoustic characteristics.

3. The system according to claim 1, wherein supervised learning is applied to the training of a speech recognition model so that it can identify diverse speech features.

Images