38results about How to "Speech recognition is accurate" patented technology

The speech recognition apparatus ( 1 ) is equipped with the garbage acoustic model storage unit ( 110 ) storing the garbage acoustic model which learned the collection of the unnecessary words; the feature value calculation unit ( 101 ) which calculates the feature parameter necessary for recognition by acoustically analyzing the unidentified input speech including the non-language speech per frame which is a unit for speech analysis; the garbage acoustic score calculation unit ( 111 ) which calculates the garbage acoustic score by comparing the feature parameter and the garbage acoustic model; the garbage acoustic score correction unit ( 113 ) which corrects the garbage acoustic score calculated by the garbage acoustic score calculation unit ( 111 ) so as to raise it in the frame where the non-language speech is inputted; and the recognition result output unit ( 105 ) which outputs, as the recognition result of the unidentified input speech, the word string with the highest cumulative score of the language score, the word acoustic score, and the garbage acoustic score which is corrected by the garbage acoustic score correcting means.

The invention provides a speech enhancement method, a speech enhancement device, a smart speaker box, and a smart television. The speech enhancement method comprises the following steps: a microphone array arranged on a smart speaker box picks up music original sound played by the speaker box and human sound produced by a person, and the music original sound and the human sound are converted by an ADC into multiple digital signals; then, an FPGA converts the multiple digital signals into one digital signal and sends the digital signal to a CPU, and the CPU acquires an echo-cancellation reference signal from the digital signal; and finally, based on the reference signal, a music original sound signal picked up by the microphone array is offset by the reference signal by use of an AEC algorithm, and human sound data is output. According to the invention, a signal is directly extracted from the microphone array as an echo-cancellation reference signal. There is no need to change the speaker circuit, and the integrity of the speaker is ensured. Speech recognition is accurate. The output audio signal is strong, and the power is high.

The invention provides a voice data processing method and apparatus of a mobile terminal. The method includes: acquiring voice data input by users; recognizing the voice data; obtaining a keyboard corresponding to the voice data and identification information of a controlled electronic device according to a recognition result, and performing classification according to goals of the users in the voice data; searching a first historical voice control event matched with the keyboard and a classification result from historical voice control events, wherein an execution result of the controlled electronic device corresponding to the first historical voice control event is execution success; and obtaining a corresponding historical control instruction from the first historical voice control event, and sending the historical control instruction to the controlled electronic device corresponding to the identification information to control the controlled electronic device to execute an operation corresponding to the historical control instruction. By employing the above scheme, the speech recognition performance can be enhanced.

The invention discloses an embedded portable voice controller which comprises a power source management module, a voice receiving module, a voice recognition module and a wireless communication module. The power source management module is respectively electrically connected with the voice receiving module, the voice recognition module and the wireless communication module; the voice receiving module comprises a microphone and a peripheral circuit; the voice recognition module comprises a voice recognition circuit, a microprocessor and a voice feedback circuit, and the voice recognition circuit is respectively electrically connected with the voice receiving module, the voice feedback circuit and the microprocessor; the wireless communication module comprises an RF signal antenna, a radio frequency chip and a peripheral circuit, the RF signal antenna is electrically connected with the radio frequency chip, and the radio frequency chip is electrically connected with the microprocessor through a serial port. The embedded portable voice controller can support various wireless communication protocols, be integrated in a whole intelligent housing system well, and control all indoor controllable household appliances through voices.

The invention provides a voice enhancement processing method and a device. The method comprises the steps that voice information from terminal equipment is acquired, wherein voice enhancement auxiliary information is carried in the voice information; if voice enhancement processing is required to be performed on the voice information by judging according to equipment identification of the terminal equipment, a corresponding voice enhancement algorithm is acquired from a plurality of local voice enhancement algorithms according to the voice enhancement auxiliary information; and the voice enhancement processing is performed on the voice information according to the acquired voice enhancement algorithm. With the adoption of the technical scheme, a voice enhancement processing procedure can be more pertinent, and an unnecessary computation burden of a server is reduced under the condition that the voice enhancement quality is ensured.

The invention discloses an echo cancellation method and device for the speech recognition process. The method comprises steps that a microphone is used for receiving the sound of a near-end user and the sound of a far-end loudspeaker and forming a first simulation signal, the first simulation signal is converted into a digital signal through an AD converter, and the digital signal is subjected toecho cancellation through a voice state detection module, a filtering control module, a self-adaptive filter, an echo cancellation module, an error calibration module, a residual echo intercept moduleand a second nonlinear processing module and then is sent to the loudspeaker. The method is advantaged in that the echo collected by the microphone can be effectively eliminated, accuracy of speech recognition can be guaranteed, speech identification precision is improved, the defects of digital domain noise cancellation in the prior art can be effectively solved, echo cancellation quality is improved, and accurate speech recognition is performed under the condition of having no effective isolation of a speech cavity and having no reduction of the volume of the prompt speech.

The invention discloses a complex scene voice recognition method based on multiple modes. The method comprises the following steps: synchronously collecting an audio signal, a lip image signal and a facial electromyogram signal corresponding to voice input if a collected lip image of a user is detected to change, determining multi-source data features of the signals in a space domain and a time domain, and coding and modeling the multi-source data features by using a speech recognition model to obtain common information of different modal expression contents and to obtain multi-modal speech information, and synthesizing a text by using a language model. The invention further discloses a complex scene voice recognition device based on multiple modes. The device comprises a data acquisitionmodule, a feature extraction module, a coding and decoding module, a text synthesis module and an interaction module. According to the invention, efficient, accurate and robust voice recognition in complex scene environments with vocal cord damage, high noise, high closure, high privacy requirements and the like is realized, and more reliable voice interaction technology and system are provided for complex man-machine interaction scenes.

The invention is applicable to the technical field of multimedia, and provides a subtitle processing method and device and a terminal, and the method comprises the steps: obtaining first subtitle datacorresponding to a multimedia file, and extracting audio data of the multimedia file; performing voice recognition processing on the audio data to generate second subtitle data; and correcting the first caption data based on the second caption data. According to the method, the caption to be corrected and the audio frequency are acquired firstly, then the audio frequency is subjected to voice recognition, the first caption data are corrected according to the second caption data obtained after recognition, manual participation is not needed in the whole correction process, and therefore automatic correction of the caption text is achieved, and the problem that manual correction is long in time consumption is solved.

The invention discloses a quadruped robot blind guiding system and method. The system comprises an environment information obtaining module which is used for obtaining the three-dimensional point cloud information of an environment; an environment perception and exploration module used for carrying out semantic modeling on the environment according to the three-dimensional point cloud information to form an incremental grid map, obtaining an area through which people and the blind guiding robot can pass in parallel in the environment according to the semantic information of the incremental grid map, and forming a passable map, and marking dynamic obstacles in the passable map to form an environment composite map; a path planning module used for analyzing the environment composite map by utilizing the probability road map to obtain all global paths suitable for the robot to move, and selecting an optimal global path from all the global paths; and a robot used for guiding the user to move along the optimal global path. A safe barrier-free path suitable for a robot and a blind person to pass through can be planned according to sensed environment information, and autonomous guidance for the blind person is achieved.

The invention provides a navigation type virtual microscope based on an intention understanding model. The navigation type virtual microscope comprises a multi-mode input and sensing module, a multi-mode information integration module and an interactive application module; the multi-mode input and sensing module is used for obtaining voice information of a user through a microphone and obtaining an operation behavior of the user; and the multi-mode information integration module is used for processing the voice information through visual channel information and processing the operation behavior through tactile channel information, and then integrating the processed voice information with the operation behavior through the multi-channel information to complete the interaction between the microscope and the user. According to the invention, multi-modal information is acquired and integrated; simple sensing elements are utilized, signal input and intelligent sensing technologies of multiple modes are added, on the basis that the advantages of the digital microscope are guaranteed, vast common and poor middle school students can learn the microscope conditionally, the cognitive feelingof the micro world is improved, and the intelligent microscope is experienced.

The invention provides a smart home voice control system based on Bluetooth transmission, which is used for realizing voice control between a mobile phone terminal and a smart home electrical appliance. Because a mobile terminal adopts a Bluetooth transmission technology, the mobile terminal is more suitable for a local area network (LAN) in the smart home, thus greatly prolonging the distance of the voice control; and noises in the voice transmitted by the Bluetooth can be extremely eliminated from a pronunciation end, thus a transmitted voice command is more easy to identify and analyze. The system comprises a smart mobile phone terminal and a central control device, wherein the smart mobile terminal comprises a Bluetooth device and a voice input device, and is used for voice collection and Bluetooth transmission of a front end; and the central control device comprises a Bluetooth device, and is used for receiving voice data sent by the smart mobile terminal via the Bluetooth, analyzing the voice data, and sending target control signals to a controller end. A user can control the home electrical appliance by inputting a voice command in the smart mobile terminal, thus realizing a function that the mobile terminal is used to control the home electrical appliance in a short distance, and reaching the purpose of home intelligentization.

The invention provides a display device for simulation interaction based on an augmented reality technology, and relates to the technical field of virtual devices. Comprising a movable main body, a content generation module, a control terminal, an LED display screen arranged on the inner wall of the main body, a voice recognition module, a sound module, a camera used for capturing position information of a user and a somatosensory tracker used for dynamic capturing. The LED display screen is connected with the control terminal through the content generation module; the sound module, the voice recognition module, the camera and the somatosensory tracker are electrically connected with the control terminal. According to the technical scheme, a user can obtain corresponding content in a naked eye 3D interaction mode, the display limitation of physical space is broken through, and the equipment is portable, movable and reusable, so that the operation cost is low.

The invention relates to an assistant management robot head device for community hospital department query and control, and solves the problem that a community hospital lacks professional leading examining personnel due to the large medical staff turnover. The head device comprises a facial device, a neck device and a control system; the facial device comprises a facial support, a speech module and a mouth device; the neck device comprises a neck support and a neck movement device. For a complex background noise environment of a hospital, the control system adopts a DSP chip for speech processing, wherein the preprocessing process adopts an improved double-threshold endpoint detection algorithm to increase the accuracy rate of speech recognition under a low signal-to-noise ratio environment, and while the speech problem of patients is answered, the movements of the facial device and neck device are controlled to complete the humanoid dialogue action with higher personification degree in cooperation with a steering engine control board.

The invention provides a voice recognition method and device, electronic equipment and a storage medium. The method comprises the steps of determining to-be-recognized voice; based on the first voice recognition model, performing acoustic state prediction on the spectrum features of the to-be-recognized voice to obtain a first acoustic state posterior probability of the to-be-recognized voice; based on a second voice recognition model, acoustic state prediction is carried out on the semantic features of the to-be-recognized voice, and a second acoustic state posterior probability of the to-be-recognized voice is obtained; and fusing the first acoustic state posterior probability and the second acoustic state posterior probability, and performing speech recognition decoding based on a fused posterior probability obtained by fusion to obtain a recognition text of the speech to be recognized. According to the speech recognition method and device, the electronic equipment and the storage medium provided by the invention, speech recognition can be accurately carried out in a domain scene.

The invention relates to a fuse box, in particular to a fuse box adopting intelligent voice control. The fuse box comprises a switch module, a voice recognition module, a micro processor module, a power supply module and a storage module. According to the fuse box disclosed by the invention, external voices are acquired and recognized by the voice recognition module, the recognized results are sent to a master control chip, and the master control chip controls the fuse box to open and close according to the results. The fuse box further has a self-learning function, and can add new voice dataaccording to the tone and accent of a regular user, so that voice recognition is more precise. The voice data and the instruction data are stored in a storage unit of the master control chip.

The invention relates to a robot and a voice recognition device and method thereof, and belongs to the field of voice processing. Voice recognition can be accurately performed in various scenes. The voice recognition device applied to the robot comprises distributed microphone arrays and a voice processor, wherein the distributed microphone arrays include a first microphone array located on the front surface of the robot and a second microphone array located on the back surface of the robot, and are used for acquiring a first voice signal and a second voice signal respectively; and the voice processor is used for fusing the first voice signal and the second voice signal to perform voice recognition.

The invention discloses a method and a system for automatically generating a voice file based on a preset recording title. The method comprises the following steps: presetting the recording title according to first audio information; performing voice recognition processing on the recording title to obtain an identification text; performing official recording through recording equipment to obtain second audio information; and after the formal recording is completed, inputting the identification text to the title of the second audio information by the recording equipment to obtain a voice file containing the identification book. According to the method, presetting of the recording title can be completed by applying on-off control of the intelligent recording device, and the recording occasion, the recording content and the recording object are marked.

The invention discloses a method for implementing a voice short message on the basis of internet telephony. The method comprises the steps of: receiving the voice short message by a mobile terminal; storing the voice short message; carrying out primary encryption processing; carrying out internet telephone processing; and receiving by a receiving terminal. According to the method disclosed by the invention, the voice short message is stored and primarily encrypted by the mobile terminal, and a sent storage address and an addressing process which are generated after the voice short message is stored in an internet telephony module are used for secondary encryption, so that security and reliability of the voice short message are greatly improved; and a receiver can send a related instruction and randomly call the voice short message from the internet telephony module, and a voice identification module automatically carries out identification and blurring operation on the voice short message to identify the voice short message as text information, calls original stored information in the internet telephony processing, carries out background comparison, automatically identifies contents of the voice short message, carries out text correction, and gives out information feedback to the text information, thereby benefiting for the receiver to accurately receive the meaning of a sender and meeting the use requirement of an operator.

The application provides a voice interaction system for electrical equipment. The approach of someone is detected through a human body sensor and a speech recognition unit, after a voice is acquired,denoising processing of the voice is performed through a denoising processing device in a speech recognition and processing module, a voice characteristic value is analyzed and extracted, and comparedwith an instruction characteristic in a database memory, an entry is broadcasted through a speech synthesis unit according to a result, and a control command is transmitted to a master control unit of the equipment through a communication circuit to realize control. According to the system, through the collaborative effect of the speech recognition and processing module and the speech recognitionunit, the recognition precision is high, energy saving and high efficiency are realized, accurate speech recognition can be realized when noises exist through the denoising processing device in the speech recognition and processing module, and problems that the conventional and traditional electrical equipment control system adopts an operation panel formed by a display and buttons, the total cost is high, and loss and damage to the buttons and the display are more easily caused in a severe outdoor condition are solved.

The present application discloses an intelligent speech recognition method, device, computer equipment and storage medium based on three-level feature collection. The method includes: performing sound collection and processing to obtain a first sound signal; performing image processing on the speaker's lips Acquisition and processing to obtain the second image signal; sending a signal acquisition request to the intraoral sensor cluster; acquiring the third sensing signal set sent by the intraoral sensor cluster; combining the first sound signal, the second sensing signal subset and the third sensing signal set Input the sensing signal subset into the first semantic recognition model to obtain the first recognized text; input the second image signal, the first sensing signal subset and the second sensing signal subset into the second semantic recognition model to obtain second recognition text; calculate the text similarity value between the first recognition text and the second recognition text; if the text similarity value is greater than the text similarity threshold, then use the first recognition text as the intelligent speech recognition result.

The present invention is applicable to the field of multimedia technology, and provides a method, device and terminal for processing subtitles, including: acquiring first subtitle data corresponding to a multimedia file, and extracting audio data of the multimedia file; The identification process generates second subtitle data; and the first subtitle data is corrected based on the second subtitle data. The method first obtains subtitles and audio to be corrected, and then performs speech recognition on the audio to correct the first subtitle data according to the second subtitle data obtained after the recognition. The automatic correction of the subtitle text solves the problem that manual correction takes a long time.

The present invention is a voice system suitable for portable smart devices and its use method, including a main control module, a wake-up module, a sound receiving module, a distance sensing module, a gesture trajectory recognition module, a knocking module and a key displayed on the screen The voice button, the sound receiving module, the distance sensing module, the gesture trajectory recognition module, the knocking module and the wake-up module are all connected to the main control module, and the distance sensing module is arranged near the sound receiving module. The using methods of the system include hardware sensing using methods, software one-button using methods, track recognition using methods and knocking using methods. The advantages of the present invention are: multiple sensors are provided, the voice system can be used in various ways, and it can be used in different occasions; different usage methods can be used together to make the voice system more simple and convenient to use; there is a chat voice system The recognition and semantic speech recognition make the speech recognition more accurate.

The invention provides a voice enhancement method, device, smart speaker, and smart TV. In the voice enhancement method, the microphone array arranged on the smart speaker first picks up the original sound of music played by the speaker and the human voice produced by people talking, and converts it into multiple voices through an ADC. Then the FPGA converts the multi-channel digital signal into a digital signal, and sends it to the CPU, and the CPU obtains a reference signal for echo cancellation from the digital signal, and finally uses the AEC algorithm based on the reference signal. The reference signal cancels the original music signal picked up by the microphone array, and outputs voice data. The invention directly extracts the signal from the microphone array as the reference signal for echo cancellation without modifying the circuit of the sound box, ensures the integrity of the sound box, accurate speech recognition, strong output audio signal and high power.

The invention provides a voice tracking method and device, a storage medium and electronic equipment. According to the method, a voice tracking mode of irrelevant text exclusion-fuzzy positioning-accurate positioning is specifically adopted, and related algorithms of error correction processing and similarity matching are further provided, so that the response speed of the system is improved while accurate voice recognition is realized. The effect of following the position of the to-be-tracked text read by the user in real time is achieved. The technical problem that the data processing speed and the voice recognition accuracy cannot be both considered during voice tracking data processing in the prior art is solved.

The invention provides a canteen card sweeping device with voice recognition and an application method thereof, and relates to the technical field of card sweeping devices. The device comprises a mobile equipment terminal, a central processing unit and an LED display card sweeper; the central processing unit comprises a wireless network system module, a voice signal receiver, a voice recognition system and an A / D converter; the mobile equipment terminal is connected with the voice signal receiver via the wireless network system module; the voice signal receiver is connected with the voice recognition system; the voice recognition system is used for performing denoising treatment on a voice command, extracting features of the voice signal, building a GMM-HMM model by using a method of training a Gaussian mixing model by using a hidden Markov model (HMM) to acquire the maximum likelihood probability of each feature, and outputting a word corresponding to the feature according to the maximum likelihood probability; the voice recognition system is connected with the A / D converter; the A / D converter is connected with the LED display card sweeper. According to the device, the operation response sensitivity can be improved, and the card sweeping and meal buying speed of the canteen can be improved.

The invention discloses an echo cancellation method used in a voice recognition process, comprising the following steps: a microphone is used to receive the sound of a near-end user and a far-end speaker, and form a first analog signal, and the first analog signal passes through an AD converter It will be converted into a digital signal, and the digital signal will be processed by multiple echo cancellation modules such as the voice state detection module, filter control module, adaptive filter, echo cancellation module, error calibration module, residual echo interception module, and second nonlinear processing module. After that, send to the speakers. The present invention can effectively eliminate the echo collected by the microphone, thereby ensuring the accuracy of speech during speech recognition, thereby improving the accuracy of speech recognition, effectively solving the defect of eliminating noise in the digital domain in the prior art, and improving echo The quality of the cancellation enables accurate speech recognition without sound cavity for efficient isolation and without reducing the volume of the prompt tone.

The invention relates to the speech recognition technology, and in particular to a speech digital recognition method based on MFCC. The speech digital recognition method based on the MFCC comprises the following steps: firstly, sampling an input speech signal, and preprocessing the sampled speech signal; performing endpoint detection on the sampled and preprocessed speech signal to extract singledigital speech signals; extracting MFCC features of each digital speech signal; and matching the MFCC features of each digital speech signal with a MFCC digital speech signal parameter template obtained through training by using a mean square error MSE method to recognize numbers in the speech signal. The speech digital recognition method based on the MFCC combines the MFCC features with the MSE to realize speech digital recognition, which not only has a high recognition rate but also avoids a large amount of data calculation; therefore, the recognition efficiency is high, and the speech digital recognition method based on the MFCC can be applied in a complex environment.