Speech recognition method and device
By obtaining the input audio information and identification of the object, the decoder is trained based on historical audio data updated based on preset time, which solves the problems of incomplete coverage and misrecognition in speech recognition technology, and improves the accuracy of speech recognition and personalized service capabilities.
Patent Information
- Application Number
- CN202310609853.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-26
- Publication Date
- 2024-11-26
AI Technical Summary
Existing speech recognition technologies are difficult to completely cover all instructions that a specific object may issue, and the frequency of entries in the training data cannot represent the preferences of different objects when using the product, resulting in misrecognition of instructions with similar pronunciation.
By obtaining the input audio information and identification of the object, the corresponding decoder is determined, and the historical audio data updated based on the preset time is trained to enhance the coverage ability of the training data and avoid misidentification.
It improves the accuracy of speech recognition, ensures that the training data can cover the common instructions of the object, adapts to the object's functional preferences, and reduces misrecognition of instructions with similar pronunciation.
Smart Images

Figure CN119028336A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of speech recognition technology, and in particular to a speech recognition method and device. Background Art
[0002] Speech recognition has always been a key technology for human-computer interaction and the connection between artificial intelligence and traditional industries, and it directly affects the intelligent development process of future society. Because in various industrial fields closely related to users' daily lives, such as smart home, smart education, and smart medical care, allowing smart devices to recognize user commands is a prerequisite for achieving natural interaction.
[0003] However, under the existing speech recognition technology, for a specific object, the data used to train the general model cannot completely cover all the commands that the object may issue, and the frequency of a certain term in the training data cannot represent the preference of different objects for the term when using the product, so that when encountering commands with similar pronunciations, misrecognition may occur. Summary of the invention
[0004] The present disclosure provides a speech recognition method, device, electronic device and storage medium.
[0005] According to a first aspect of the present disclosure, a speech recognition method is provided, the method comprising: obtaining input audio information of a first object and an identifier of the first object, the identifier of the first object being used to determine a first decoder; determining a first decoder corresponding to the identifier of the first object, the first decoder being trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object; decoding the input audio information based on the first decoder to obtain a recognition result of the input audio information.
[0006] In some embodiments, obtaining input audio information of the first object and an identification of the first object includes: extracting visual features of the first object and / or acoustic features of the input audio information; and determining an identification of the first object that matches the visual features and / or acoustic features.
[0007] In some embodiments, the method further includes: interpolating the first data set and the second data set based on preset weight coefficients to obtain a training data set, wherein the second data set is used to train a second decoder, the second data set includes the first data set, and the second decoder is used to decode the audio information of any object; and using the training data set to train the first decoder.
[0008] In some embodiments, the method further comprises: compressing the first decoder based on a preset compression coefficient,
[0009] Specifically, the first decoder is pruned based on a maximum entropy model.
[0010] In some embodiments, the input audio information is decoded based on a first decoder to obtain a recognition result of the input audio information, and then the method includes: storing the input audio information to a first data set; performing enhancement processing on the first data set at a preset time, and the enhancement processing includes at least one of utterance generalization, scrambling the order, multiple repetitions, and speed disturbance.
[0011] According to an embodiment of the present disclosure, by obtaining input audio information of a first object and an identifier of the first object, the identifier of the first object is used to determine a first decoder; determining a first decoder corresponding to the identifier of the first object, the first decoder is trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object; decoding the input audio information based on the first decoder to obtain a recognition result of the input audio information, thereby updating the training data at a preset time with the help of the historical data of the first object, and thus approaching the functional preference of the first object, ensuring that the historical data of the first object is complete, so that the training data can cover the commonly used instructions of the first object, avoiding misrecognition when encountering instructions with similar pronunciations, and improving the accuracy of speech recognition.
[0012] According to a second aspect of the present disclosure, a speech recognition device is provided, which includes: an acquisition unit, used to acquire input audio information of a first object and an identifier of the first object, the identifier of the first object being used to determine a first decoder; a determination unit, used to determine a first decoder corresponding to the identifier of the first object, the first decoder being trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object; a decoding unit, used to decode the input audio information based on the first decoder to obtain a recognition result of the input audio information.
[0013] According to an embodiment of the present disclosure, by obtaining input audio information of a first object and an identifier of the first object, the identifier of the first object is used to determine a first decoder; determining a first decoder corresponding to the identifier of the first object, the first decoder is trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object; decoding the input audio information based on the first decoder to obtain a recognition result of the input audio information, thereby updating the training data at a preset time with the help of the historical data of the first object, and thus approaching the functional preference of the first object, ensuring that the historical data of the first object is complete, so that the training data can cover the commonly used instructions of the first object, avoiding misrecognition when encountering instructions with similar pronunciations, and improving the accuracy of speech recognition.
[0014] According to a third aspect of the present disclosure, there is provided an electronic device, including:
[0015] at least one processor; and
[0016] a memory communicatively connected to at least one processor; wherein,
[0017] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable the at least one processor to execute the method of the first aspect described above.
[0018] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method of the aforementioned first aspect.
[0019] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, and when the computer program is executed by a processor, the computer program implements the method of the aforementioned first aspect.
[0020] According to a sixth aspect of the present disclosure, a vehicle is provided, comprising the speech recognition device as described in the second aspect or the electronic device as described in the third aspect.
[0021] According to an embodiment of the present disclosure, by obtaining input audio information of a first object and an identifier of the first object, the identifier of the first object is used to determine a first decoder; determining a first decoder corresponding to the identifier of the first object, the first decoder is trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object; decoding the input audio information based on the first decoder to obtain a recognition result of the input audio information, thereby updating the training data at a preset time with the help of the historical data of the first object, and thus approaching the functional preference of the first object, ensuring that the historical data of the first object is complete, so that the training data can cover the commonly used instructions of the first object, avoiding misrecognition when encountering instructions with similar pronunciations, and improving the accuracy of speech recognition.
[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.
[0024] Figure 1 A flowchart of a speech recognition method provided by an embodiment of the present disclosure;
[0025] Figure 2 A flowchart of a speech recognition method provided by an embodiment of the present disclosure;
[0026] Figure 3 A schematic diagram of a process of obtaining a first decoder provided in an embodiment of the present disclosure;
[0027] Figure 4 A schematic diagram of the structure of a speech recognition device provided by an embodiment of the present disclosure;
[0028] Figure 5 A schematic block diagram of an exemplary electronic device 500 provided for an embodiment of the present disclosure. DETAILED DESCRIPTION
[0029] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.
[0030] A speech recognition method, device, electronic device, and storage medium according to an embodiment of the present disclosure will be described below with reference to the accompanying drawings.
[0031] Speech recognition has always been a key technology for human-computer interaction and the connection between artificial intelligence and traditional industries, and it directly affects the intelligent development process of future society. In recent years, with the rapid development of speech recognition technology, speech recognition technology has gradually entered our lives and work. In various industrial fields closely related to users' daily lives, such as smart home, smart education, and smart medical care, allowing smart devices to recognize user commands is a prerequisite for achieving natural interaction.
[0032] However, under the existing speech recognition technology, the language model is usually a statistical model. For a certain term, the general model needs enough training data to estimate its probability more accurately. However, due to the unreliable or unavailable network connection, speech recognition requests often have high latency. Therefore, the ASR system (automotive anti-skid control system) running on mobile devices has the limitation of not consuming a large amount of memory or computing resources. Therefore, the general model cannot completely rely on the growth of training data to improve the recognition effect.
[0033] For a specific subject, the corpus used to train the general model cannot completely cover all the instructions that the subject may issue. Moreover, the frequency of a certain term in the training data cannot represent the preference of different subjects for the term when using the product, so that when encountering instructions with similar pronunciations, misidentification may occur. For example, in the vehicle control scenario, "turn on camping mode for me" and "turn on recording mode for me", for subjects who have completely different functional preferences when using the car, the general model cannot make some specific adaptations based on the preferences of the subject. In addition, the preferences of the subject will change or shift at irregular intervals over time; for example, a few weeks ago, subject A liked to listen to a rap song "Little Gorilla" on the Chinese Hot List, while subject B liked to listen to a song "Little Star" in the nostalgic golden song. Two weeks later, subject B also started listening to the rap song "Little Gorilla" on the Chinese Hot List on the recommendation of a friend.
[0034] Therefore, in view of the above-mentioned problems of the prior art, how to quickly iterate the model and make correct identification and recommendations for different users at different time points has become an important function to be improved. This disclosure proposes to use the user's personal data to complete the automatic iteration of the model and make predictions close to the user's commands in the smart cockpit, so as to provide more accurate personalized voice recognition services and accurately match the user's actual needs.
[0035] The speech recognition method proposed in the present disclosure can be applied to devices with speech recognition functions. The application scenarios are not limited in the embodiments of the present disclosure. The present disclosure mainly takes the in-vehicle speech recognition scenario as an example.
[0036] The following describes in detail a speech recognition method, device, electronic device, storage medium, and program product proposed in the present disclosure in conjunction with the accompanying drawings.
[0037] Figure 1 The following is a flow chart of a speech recognition method provided by an embodiment of the present disclosure. Figure 1 As shown, the method comprises the following steps:
[0038] Step 101: obtain input audio information of a first object and an identifier of the first object, where the identifier of the first object is used to determine a first decoder.
[0039] In one implementation of the present disclosure, the first object may be a first terminal or at least one object associated with the first terminal, wherein the first terminal may be a vehicle or other electronic device with a voice recognition function, and the at least one object associated with the first terminal may be the current user of the device or a user who has been connected to the electronic device. The user may be multiple or one, and the number is not limited in the present disclosure.
[0040] The input audio information of the first object is collected by a microphone and / or a camera. The installation position of the microphone and / or the camera can be installed according to the actual size or shape of the vehicle so that the microphone and / or the camera can collect the characteristics of the first object, which is not limited in the present disclosure.
[0041] The identifier of the first object is a personalized feature of the first object extracted through biometric recognition technology. It can be understood that there is only one identifier of the first object, which can be an acoustic feature or a visual feature.
[0042] Step 102: determining a first decoder corresponding to the identifier of the first object, wherein the first decoder is trained based on a first data set updated at a preset time, wherein the first data set includes historical audio information corresponding to the first object.
[0043] In one embodiment of the present disclosure, the first decoder refers to a speech recognition model for the first object, and the speech recognition model is obtained by fusing a personalized language model for the first object with a general model. The first object and the first decoder are in one-to-one correspondence.
[0044] It should be noted that the first data set and the second data set are weightedly fused to obtain the training data set, and the first decoder is trained based on the training data set. The second data set refers to the full amount of historical audio information, that is, the historical audio information of all objects received by the first terminal.
[0045] The historical audio information refers to the historical instructions issued by the first object, such as turning on the air conditioner, playing music, etc.
[0046] Step 103: decode the input audio information based on the first decoder to obtain a recognition result of the input audio information.
[0047] In one embodiment of the present disclosure, the audio information emitted by the first object is recognized by the first decoder (the speech recognition model for the first object) to obtain a corresponding recognition result.
[0048] It can be understood that since the first decoder is a decoder in the speech recognition model for the first object, when performing speech recognition, it is possible to obtain a recognition result that is adapted to the first object's historical preferences based on the first object's historical preferences. For example, when the first object issues an instruction to play "xiaoxingxing", the first decoder can automatically identify and execute the playback of "little gorilla" based on the first object's historical preference "little gorilla".
[0049] Therefore, according to an embodiment of the present disclosure, by obtaining the input audio information of the first object and the identifier of the first object, the identifier of the first object is used to determine a first decoder; the first decoder corresponding to the identifier of the first object is determined, and the first decoder is trained based on a first data set updated at a preset time, and the first data set includes historical audio information corresponding to the first object; the input audio information is decoded based on the first decoder to obtain a recognition result of the input audio information, and the training data is updated at a preset time with the help of the historical data of the first object, so as to be closer to the functional preference of the first object, and ensure that the historical data of the first object is complete, so that the training data can cover the commonly used instructions of the first object, avoid misrecognition when encountering instructions with similar pronunciations, and improve the accuracy of speech recognition.
[0050] Figure 2 A flowchart of a speech recognition method provided in an embodiment of the present disclosure. Figure 2 based on Figure 1 In the embodiment shown, step 101 and step 103 are further defined. Figure 2 In the embodiment shown, step 101 includes step 201, and step 103 includes step 203 and step 204. Figure 2 As shown, the method includes the following steps.
[0051] Step 201: extract visual features of a first object and / or acoustic features of input audio information.
[0052] In some embodiments of the present disclosure, when the first subject uses this function, he or she can register his or her personal face and voiceprint in advance, that is, associate them with the vehicle; when the first subject issues a command in the vehicle, the vehicle computer can use the microphone and camera as the medium, and utilize technologies such as visual retrieval and voiceprint recognition to extract the visual and / or audio features of the first subject.
[0053] Voiceprint recognition is a type of biometric technology, also known as speaker recognition, which is a technology that identifies the speaker's identity through sound. The so-called acoustic feature (voiceprint) is the sound wave spectrum that carries speech information displayed by an electroacoustic instrument. The vocal organs (tongue, teeth, larynx, lungs, nasal cavity) used by people when speaking vary greatly in size and shape, so the voiceprints of any two people are different. Therefore, the current object can be distinguished by the voiceprint features. Similarly, since everyone's facial features are different, visual retrieval can distinguish and judge by identifying the visual features of the current object (such as facial feature points).
[0054] Among them, the visual features and / or acoustic features of the input audio information corresponding to the first object can be a person, such as the driver or passenger of the current vehicle, or a group, such as all drivers or passengers who have used the vehicle in units of vehicles, which is not limited in the present disclosed embodiment.
[0055] Step 202: Determine an identification of a first object that matches the visual features and / or acoustic features.
[0056] In some embodiments of the present disclosure, according to the visual features and / or acoustic features extracted in step 201, a corresponding relationship table between the visual features and / or acoustic features and the first object is searched, and the identifier of the first object that issues the command in the smart cockpit is locked through the relationship table. For the identifier of the first object, each time the first object issues a command, the command is stored in the background, that is, in a log.
[0057] The identifier of the first object may be the ID of the current user, and each first object corresponds to one and only one identifier.
[0058] Step 203: determining a first decoder corresponding to the identifier, wherein the first decoder is trained based on a first data set updated at a preset time, wherein the first data set includes historical audio information corresponding to the first object.
[0059] In some embodiments of the present disclosure, there is only one first decoder corresponding to each identifier, and the speech recognition model performs enhancement processing on the first data set at a preset time, and the enhancement processing includes at least one of generalization of utterances, disordering of order, multiple repetitions, and speed disturbance. In other words, for a single user, the preset time can be in days, and the speech recognition model can perform data enhancement on the historical audio information corresponding to the first object every two weeks.
[0060] Specifically, data enhancement refers to generalizing the entity words in all historical instructions corresponding to the first object, disrupting the order, repeating multiple times, speed disturbance and other common methods to make the historical instructions richer. For example, when the historical instruction of the first object is "play the little gorilla", "play the little gorilla" can be generalized to "please play the little gorilla", or the order of inputting the instruction can be disrupted, etc. In this way, the timeliness can be maintained without losing the popularity of the previous entries.
[0061] It should be noted that the preset time can be in days or months. The specific period depends on the actual usage and can be two weeks or one month, which is not limited in the present embodiment.
[0062] The training of the first decoder includes: interpolating the first data set and the second data set based on a preset weight coefficient to obtain a training data set, wherein the second data set is used to train the second decoder, and the second data set includes the first data set; and using the training data set to train the first decoder. Compressing the first decoder based on a preset compression coefficient specifically includes pruning the first decoder based on a maximum entropy model.
[0063] The second data set refers to the full data, that is, it includes the historical audio information corresponding to all objects.
[0064] Step 204: decode the input audio information based on the first decoder to obtain a recognition result of the input audio information.
[0065] In some embodiments of the present disclosure, decoding of input audio information by the first decoder is a pattern recognition process, specifically referring to recognizing the instructions issued by the first object by the speech recognition model of the first object, comparing the unknown speech pattern with the known speech pattern, and the best matching reference pattern is regarded as the recognition result.
[0066] In some optional embodiments of the present disclosure, when the user considers personal privacy and does not want to register personal information or does not want the background to retain personal information every time a command is issued, the first object can be changed to complete the above-mentioned voice recognition process in units of vehicles, that is, all audio information of the vehicle in the past two weeks is collected every day and the voice recognition model for the vehicle is trained. In this way, for a family, no distinction is made between specific family members, but there is still a clear ability to distinguish between groups of people.
[0067] To summarize, according to an embodiment of the present disclosure, by extracting the visual features of the first object and / or the acoustic features of the input audio information, the identifier of the first object matching the visual features and / or acoustic features is determined, and the first decoder corresponding to the identifier is determined. The first decoder is trained based on a first data set updated at a preset time, and the first data set includes historical audio information corresponding to the first object. The input audio information is decoded based on the first decoder to obtain a recognition result of the input audio information. The training data is updated at a preset time with the help of the historical data of the first object, and the functional preference of the first object can be approached to ensure the integrity of the historical data of the first object, so that the training data can cover the commonly used instructions of the first object, avoid misrecognition when encountering instructions with similar pronunciations, and improve the accuracy of speech recognition.
[0068] based on Figure 1 and Figure 2 A speech recognition method is proposed, and the present disclosure provides a flowchart of obtaining a first decoder, such as Figure 3 shown.
[0069] Step 301: Based on a preset weight coefficient, interpolation processing is performed on the first data set and the second data set to obtain a training data set.
[0070] In the present disclosure, a second data set is used to train a second decoder, and the second data set includes the first data set.
[0071] In some embodiments of the present disclosure, the first data set refers to all historical instructions issued to the current first object after being updated at a preset time. The second data set refers to the full amount of historical instructions, that is, the historical instructions issued by all objects currently received. Since all objects include the first object, the second data set includes the first data set. Combining the first data set with the second data set can make the output recognition result more adapted to the historical preference of the first object.
[0072] The second decoder is used to decode the audio information of any object. The second decoder may specifically refer to a universal language model, that is, a universal language model is constructed using the full amount of data.
[0073] The first data set is fused with the second data set, and interpolation processing can be performed using preset weight coefficients to obtain a training data set. For example, the first data set is multiplied by a corresponding weight coefficient of 2 or 3, and the second data set is multiplied by a corresponding weight coefficient of 8 or 7 to perform weighted fusion to obtain a weighted training data set.
[0074] It can be understood that the preset weight coefficient can be set according to actual usage requirements and is a fixed constant to realize the process of automated training. The specific value of the preset weight coefficient is not limited in the embodiment of the present disclosure.
[0075] Step 302: Train the first decoder using the training data set.
[0076] In some implementations of the present disclosure, the first decoder is trained according to the training data set obtained in step 301. The first decoder is a language model for the first object.
[0077] Both the first decoder and the second decoder can be n-gram language models. The language model integrates the knowledge of grammar and words to calculate the probability of a word appearing in the sentence.
[0078] Step 303, compressing the trained first decoder based on a preset compression coefficient, specifically includes:
[0079] The first decoder is pruned based on a maximum entropy model to obtain a first decoder for input audio information of a first object.
[0080] In some embodiments of the present disclosure, since the first decoder may be an n-gram language model, the number of n-grams is too large. In certain specific cases, such as when the storage space on a handheld device is relatively small, considering the memory requirements, the first decoder needs to be compressed to a reasonable range and the performance difference between the compressed language model and the uncompressed language model is small.
[0081] Compression refers to the cutting of the n-gram language model based on the size of a preset compression factor. The preset compression factor can specifically be the size of the entropy, that is, by judging whether the change in entropy is large when the redundant items to be deleted by the n-gram language model are large. If the entropy does not change significantly, the redundant items can be deleted at this time. In the present disclosure, the value of entropy can be 5e-8, and the specific value can be set according to the actual memory requirements, which is not limited in the present disclosure.
[0082] According to the embodiments of the present disclosure, by using the user's personal historical data for data enhancement, the model can be closer to the user's personal preferences. At the same time, model automation can provide users with higher quality recognition services, and automatic updates at preset time (for example, in days) can ensure the timeliness of the model learning user preferences and ensure that the model does not lose the user's recent historical preferences.
[0083] Corresponding to the above-mentioned speech recognition method, the present disclosure also proposes a speech recognition device. Figure 4 FIG. 4 is a schematic diagram of a structure of a speech recognition device 400 provided in an embodiment of the present disclosure. Figure 4 As shown, including:
[0084] An acquiring unit 410 is configured to acquire input audio information of a first object and an identifier of the first object, where the identifier of the first object is used to determine a first decoder;
[0085] A determination unit 420, configured to determine a first decoder corresponding to the identifier of the first object, the first decoder being trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object;
[0086] The decoding unit 430 is used to decode the input audio information based on the first decoder to obtain a recognition result of the input audio information.
[0087] In some embodiments, the acquisition unit 410 is used to: extract visual features of the first object and / or acoustic features of the input audio information; and determine an identifier of the first object that matches the visual features and / or acoustic features.
[0088] In some embodiments, the device 400 also includes: a training unit, which is used to interpolate the first data set and the second data set based on a preset weight coefficient to obtain a training data set, wherein the second data set is used to train a second decoder, the second data set includes the first data set, and the second decoder is used to decode the audio information of any object; and the first decoder is trained using the training data set.
[0089] In some embodiments, the apparatus 400 further includes: a compression unit, configured to compress the first decoder based on a preset compression coefficient.
[0090] In some embodiments, the input audio information is decoded based on the first decoder to obtain a recognition result of the input audio information. After that, the device 400 also includes: a processing unit, which is used to store the input audio information into a first data set; and the first data set is enhanced for a preset time, and the enhanced processing includes at least one of utterance generalization, scrambling the order, multiple repetitions, and speed disturbances.
[0091] In some embodiments, the first object is the first terminal or at least one object associated with the first terminal.
[0092] In summary, according to the embodiments of the present disclosure, a speech recognition device is used to obtain input audio information of a first object and an identifier of the first object, and the identifier of the first object is used to determine a first decoder; a first decoder corresponding to the identifier of the first object is determined, and the first decoder is trained based on a first data set updated at a preset time, and the first data set includes historical audio information corresponding to the first object; the input audio information is decoded based on the first decoder to obtain a recognition result of the input audio information, and the training data is updated at a preset time with the help of the historical data of the first object, so as to be closer to the functional preference of the first object, ensure the integrity of the historical data of the first object, so that the training data can cover the commonly used instructions of the first object, avoid misrecognition when encountering instructions with similar pronunciations, and improve the accuracy of speech recognition.
[0093] It should be noted that since the device embodiment of the present disclosure corresponds to the above-mentioned method embodiment, the above-mentioned explanation of the method embodiment is also applicable to the device of the present embodiment, and the principle is the same. For details not disclosed in the device embodiment, reference can be made to the above-mentioned method embodiment, and they will not be repeated in this disclosure.
[0094] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.
[0095] Figure 5A schematic block diagram of an example electronic device 500 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.
[0096] like Figure 5 As shown, the device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 502 or a computer program loaded from a storage unit 508 to a RAM (Random Access Memory) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An I / O (Input / Output) interface 505 is also connected to the bus 504.
[0097] A number of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0098] The computing unit 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as a method for speech recognition. For example, in some embodiments, the method for speech recognition may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the aforementioned speech recognition method in any other appropriate manner (for example, by means of firmware).
[0099] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0100] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
[0101] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0102] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).
[0103] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.
[0104] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.
[0105] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.
[0106] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.
[0107] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A speech recognition method, characterized in that: The method comprises: Acquire input audio information of a first object and an identifier of the first object, where the identifier of the first object is used to determine a first decoder; Determine the first decoder corresponding to the identifier of the first object, the first decoder is trained based on a first data set updated at a preset time, the first data set includes historical audio information corresponding to the first object; The input audio information is decoded based on the first decoder to obtain a recognition result of the input audio information.
2. The method according to claim 1, characterized in that: The acquiring the input audio information of the first object and the identifier of the first object comprises: Extracting visual features of the first object and / or acoustic features of the input audio information; An identity of the first object to which the visual feature and / or acoustic feature matches is determined.
3. The method according to claim 1, characterized in that The method further comprises: Based on a preset weight coefficient, interpolating the first data set and the second data set to obtain a training data set, wherein the second data set is used to train a second decoder, the second data set includes the first data set, and the second decoder is used to decode audio information of any object; The first decoder is trained using the training data set.
4. The method according to any one of claims 1 to 3, characterized in that The method further comprises: Compressing the first decoder based on a preset compression coefficient specifically includes: The first decoder is pruned based on a maximum entropy model.
5. The method according to any one of claims 1 to 3, characterized in that The input audio information is decoded based on the first decoder to obtain a recognition result of the input audio information, and then the method includes: storing the input audio information into the first data set; The first data set is enhanced within the preset time, wherein the enhancement includes at least one of generalization of the statement, disruption of the order, multiple repetition, and speed disturbance.
6. A speech recognition device, characterized in that: include: an acquisition unit, configured to acquire input audio information of a first object and an identifier of the first object, wherein the identifier of the first object is used to determine a first decoder; A determining unit, configured to determine the first decoder corresponding to the identifier of the first object, the first decoder being trained based on a first data set updated at a preset time, the first data set including historical audio information corresponding to the first object; A decoding unit is used to decode the input audio information based on the first decoder to obtain a recognition result of the input audio information.
7. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.
8. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-5.
9. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 5.
10. A vehicle, characterized in that: Includes the speech recognition device as claimed in claim 6 or the electronic device as claimed in claim 7.
