100results about How to "Improve Speech Recognition Efficiency" patented technology

A method and apparatus for speech processing in a distributed speech recognition system having a front-end and a back-end. The speech processing steps in the front-end are as follows: extracting speech features from a speech signal and normalizing the speech features in order to alter the power of the noise component in the modulation spectrum in relation to the power of the signal component, especially with frequencies above 10 Hz. A low-pass filter is then used to filter the normalized modulation spectrum in order to improve the signal-to-noise ratio (SNR) in the speech signal. The combination of feature vector normalization and low-pass filtering is effective in noise removal, especially in a low SNR environment.

The embodiment of the invention provides a speech information identifying method and a piece of equipment, relates to the field of communication, and aims to improve the efficiency of speech recognition database determining by electronic equipment and improve the user experience. The method comprises the following steps: the electronic equipment acquires speech information input by a user and current geographic location information of the user; and the electronic equipment determines a speech recognition database based on the geographical location information and identifies the speech information according to the speech recognition database. The speech information identifying method and the equipment provided by the embodiment of the invention are used for identifying speech information.

The invention discloses a method for voice recognition and electronic equipment. The method for voice recognition is applied to the electronic equipment, and the method includes: receiving first voice data input by a user; based on identity information of the user, obtaining first high-frequency vocabulary database corresponding to the identity information of the user; and utilizing the first high-frequency vocabulary database to recognize the first voice data. Corresponding voice recognition is performed on the voice data input by the user through the high-frequency vocabulary database corresponding the user, thereby solving the technical problem in the prior art that since all voice data need to be compared and recognized in a standard vocabulary database, the efficiency of voice recognition is low, and achieving the technical effect of improving the efficiency of voice recognition.

The invention discloses a speech recognition method. The method comprises that a speech recognition request which contains speech information sent from a client is received; an optimal speech recognition engine is selected from multiple speech recognition engines, and converts the speech information into character information; and the character information is transmitted to the client. The invention discloses a speech recognition server correspondingly. Thus, preferred speech recognition engines can be selected to provide service according to different users, and the service efficiency of speech recognition is improved.

A method of speech recognition and language-understanding analysis is provided. According to a segmental word-concept-tag compound N-gram model, an input speech is divided into a plurality of segmental phrases. Each segmental phrase is attached a tag to indicate whether said segmental phrase is a meaningful segmental phrase or a meaningless segmental phrase. The meaningless segmental phrases are deleted, and only the meaningful segmental phrases are reserved. The language-understanding analysis is carried out to the meaningful segmental phrases according to segmental sub-grammars.

The invention discloses a voice shopping method and device and relates to the technical field of voice recognition. The method comprises a step of receiving a voice input of a user and performing semantic recognition on the voice input to determine a keyword of content quired by the user, a step of determining a search range of the keyword according to a shopping behavior record of the user, a step of searching the keyword to obtain a commodity information related to the quired content and carrying out voice output, and a step of receiving an order instruction inputted by the user through voice. According to the method and the device, the voice input can be processed according to the shopping habits of the user, thereby the voice recognition efficiency is improved, and a user experience isimproved.

The embodiments of the present invention disclose a speech recognition method and device. According to one specific embodiment of the invention, the method includes the following steps that: the identity information of a target user inputting speech input signals is determined in response to the received speech input signals; the commonly-used expression set of the target user is extracted from astored commonly-used expression database on the basis of the identity information of the target user, wherein the commonly-used expression set contains a plurality of commonly-used expressions; acoustic feature extraction is performed on the speech input signals, and the acoustic features of the speech input signals are inputted into an acoustic model, so that the acoustic model score of the speech input signals can be obtained; whether the content of the speech input signals is a commonly-used expression of the target user is determined on the basis of the acoustic model score of the speech input signals and the acoustic model scores of the commonly-used expressions in the stored commonly-used expression set of the target user; and if the content of the speech input signals is a commonly-used expression of the target user, a language model is constructed based on commonly-used expressions is adopted to decode the acoustic features of the speech input signals, so that a speech recognition result can be obtained. With the method and device provided by the embodiments adopted, speech recognition efficiency can be improved.

The invention relates to a speech recognition method and device, a computer readable storage medium and a computer device. The method comprises steps of acquiring multiple word sequences obtained by decoding the speech data and first scores corresponding t the first word sequences; extracting preset quantity of word sequences whose first scores rank highly from the word sequences to serve as candidate word sequences; recognizing the field where the candidate word sequences are located; according to the field where the candidate word sequences are located, inputting the candidate word sequencesto a neural network corresponding to the field; carrying out re-scoring on the candidate word sequences through the neural network so as to obtain a second score corresponding to each candidate wordsequence; according to the first scores and the second scores corresponding to the candidate word sequences, obtaining the final scores of the candidate word sequences; and using the candidate word sequence with the highest final score as a speech recognition result of the speech data. In this way, the time for training the neural network of a special field is greatly shortened; and efficiency ofthe speech recognition is further improved.

The invention discloses a speech recognition method of vehicle equipment. The vehicle equipment includes a speech acquisition module and stores a local command recognition word library. The speech recognition method includes the steps as follows: receiving a speech signal which is collected by the speech acquisition; recognizing a controlled module corresponding to the speech signal according to the local command recognition word library; detecting the current internet connection status of the controlled module; when the current internet connection status of the controlled module conforms to presupposition condition, the speech signal is recognized by a cloud speech recognition method, and the controlled module works according to the feedback result of speech recognition. The invention also discloses the vehicle equipment and a computer readable storage medium. The technical scheme of the vehicle equipment, the speech recognition and the readable storage medium aims at solving the disadvantage of low efficiency of the speech recognition.

The invention discloses a method and device for adding punctuation in voice recognition, a computer device and a storage medium. The method comprises: performing voice recognition on an acquired to-be-recognized voice, synchronously detecting a mute segment in the to-be-recognized voice, and determining whether the time length of the mute segment exceeds a first time length; if so, outputting a text sequence before the mute segment, and inserting a comma or period into a corresponding position in the text sequence according to the time length of the mute segment; performing voice recognition on the acquired to-be-recognized voice after the mute segment, and correcting the comma or period inserted into the text sequence according to a preset discrimination model. The method and device are used for improving the output efficiency and the accuracy of the punctuation in voice recognition, so as to achieve a purpose of improving voice recognition efficiency, accurately breaking sentences and accurately expressing emotions.

The invention discloses a speaker-dependent voice recognizing method. The speaker-dependent voice recognizing method is characterized in that a voice data cache and a benchmark voice database are defined preliminarily, and original voice signals are stored in the voice data cache and sent to a voice signal recognition engine to be processed and recognized to acquire recognition results with acoustic features and recognition factors; the benchmark voice database is judged whether to have benchmark voice data or not; if the benchmark voice database has no benchmark voice data, the benchmark voice data are initialized according to voice signal recognition engine results; if the benchmark voice database has the benchmark voice data, corresponding processing methods are selected to process according to different voice signal recognition engine results, so that the benchmark voice data are updated or recognition results of the voice signal recognition engine are corrected; and finally a speaker dependent pronounces the same recognition word repeatedly, and the previous steps are performed repeatedly to update the benchmark voice data continuously to make the benchmark voice data to be optimal. The speaker-dependent voice recognizing method has the advantages that speaker-dependent voice recognition efficiency is improved, and voice false recognition and recognition rejection are reduced; and with the increase of the number of use of users, the benchmark voice data in the benchmark voice database are reliable, recognition accuracy and efficiency are increased, and user experience effects are improved well.

The invention provides a speech recognition method and device. The method comprises the steps that speech data are received; whether the speech data belong to the command word class used for trigging the specific function of a client side is judged, wherein the command word class includes multiple command words composed of words or the combination of words; when the speech data belong to the command word class, the command words corresponding to the speech data are determined in the command word class, and the specific function corresponding to the command words is triggered; and when the speech data do not belong to the command word class, an operation instruction including the speech data is uploaded to a network side, the character content corresponding to the speech data is recognized by the network side and corresponding operation is performed according to the operation instruction. According to the technical scheme, the situation that the user requires to switch between different speech recognition modes when the user continuously inputs different types of speech data can be avoided so that the use experience of the client side speech recognition function can be greatly enhanced for the user.

The present invention discloses a multiplayer collaborative recording system and identification method based on instant communication. The system comprises a collaborative recording module which is used for collaborating a recording process and comprises room creating, member adding, session starting, voice identifying as a session record and realizing email sharing, and a message processing module for processing the real-time message generated by the system. The method comprises the steps of using the real-time communication technology to realize the real-time communication of different clients, parallelly uploading the voice by using a synchronous uploading method and reducing the network time delay of transmission, using a Xunfei voice identifier to support the voice identification of multiple languages, carrying out time division identification of different voices based on an asynchronous identification method, and organizing a voice identification result as the session record according to a given format. According to the system and the method, a traditional recording function is extended, the non-face-to-face multiplayer collaborative recording can be realized, and the instant messaging and voice noise reduction technology are used to improve the voice identification speed and accuracy of collaborative recording.

The embodiment of the invention provides a speech recognition method, device and system. The method comprises the following steps: receiving a wake-on-voice word; identifying a first dialect to whichthe wake-on-voice word belongs; sending a service request to a server to request the server to select an ASR (Automatic Speech Recognition) model corresponding to the first dialect from ASR models corresponding to different dialects, so as to perform speech recognition on a speech signal to be identified by the server by using the ASR model corresponding to the first dialect. The method provided by the embodiment can automatically perform the speech recognition on multiple dialects and improve the speech recognition efficiency for multiple dialects.

The invention discloses a wireless voice control method and system for a mobile platform. The method comprises the following steps: creating a wireless communication connection between a mobile control terminal and the mobile platform; acquiring a voice signal through the mobile control terminal; performing voice recognition for the acquired voice signal to obtain a recognized text; performing semantics analysis for the recognized text and generating an executing order through the mobile control terminal; sending the executing order to the mobile platform through the mobile control terminal; receiving and executing the order through the mobile platform. With the adoption of the wireless voice control method and system for the mobile platform, the inconvenience of manual control mode of the traditional mobile device is overcome, the operation is simple, the design concept of man-machine integration is interpreted well, the efficiency is high, the accuracy is ensured, and the voice recognizing efficiency and precision are greatly raised.

The invention relates to a speech recognition method. The method comprises that a speech data stream formed by collecting speech data in real time is obtained, and the speech data stream is processedsuccessively to generate a corresponding data packet flow; the data packet flow is transmitted to a server in the form of flow so that the server carries out real-time speech analysis on a received data packet to generate a corresponding recognition result, and the total length of the analyzed data packet is counted; recognition results and corresponding total lengths of the analyzed data packetsreturned by the server according to the recognition progress are received from the server; when the speech data stream is ended, the total data stream length corresponding to the speech data stream isobtained; and when a comparison result by comparing the total length of the analyzed data packet at present with the total data stream length satisfies a condition to end speech recognition in advance, a target speech recognition result of the speech data stream is obtained according to the present recognition result corresponding to the total length of the analyzed data packet at present. The speech recognition efficiency is improved.

The invention relates to a computer voice control method and an intelligent voice assistant system. The intelligent voice assistant system comprises a display interface used for receiving a first operation instruction which is inputted by a user and is used for starting the intelligent voice assistant system, a memory used for storing voice configuration files and mouse and keyboard configuration files, a voice acquisition device used for acquiring voice commands inputted by the user and transmitting the voice commands to a processor, and a processor used for converting the voice commands into corresponding voice command entries, calling the voice configuration files of the memory and matching the voice command entries with entries of the voice configuration files, if matching succeeds, program operation sequences of the mouse and keyboard configuration files of the memory corresponding to the voice command entries are called to control program operation, and the display interface is further used for displaying a performing result to be success. The intelligent voice assistant system is advantaged in that a computer is controlled for operation through the voice commands instead of mouse and keyboard operation, and thereby the computer is made to be more concise and convenient for use.

The invention relates to a voice recognition method and device, a storage medium and electronic equipment and aims to reduce complexity of a voice recognition system and improve voice recognition efficiency. The voice recognition method comprises steps of obtaining a to-be-recognized target audio; performing feature extraction on the target audio to obtain a voice feature sequence; and inputting the voice feature sequence into a voice recognition model to obtain a target text with punctuations corresponding to the target audio, the voice recognition model being obtained by training a sample text with punctuations marked with punctuation information and a sample audio corresponding to the sample text with punctuations.

The invention discloses a model library establishing method and device, a voice recognition method and device, a facility and a medium. The model library establishing method comprises the following steps: acquiring a training sample set; if the sample number of original voice data in the training sample set is greater than a preset threshold value, establishing a current grading model according totraining voice characteristics extracted by the original voice data, and dividing the original voice data into at least two current training subsets according to the model grading logic; if the sample number of the current training subsets is greater than the preset threshold value, updating the current training subsets to the training sample set; and if the sample number of the current trainingsubsets is not greater than the preset threshold value, determining the current training subsets as identification data sets. With the method, a recognition server can rapidly match the identificationdata set in which to-be-detected voice data are located according to the model grading logic of a grading model library, so that the voice recognition efficiency is improved.

The invention relates to a language depth neural network-based language recognition method. The method includes the following steps that: S11, biorthogonal wavelet transformation is adopted to filterinputted speech information; S12, the Mel frequency cepstrum coefficient is adopted to extract to a feature sequence formed by key feature parameters reflecting speech signal features; S13, the feature parameters of a training speech library are adopted to train an acoustic model; S14, the feature parameters of a text database are adopted to a train language model; S15, a decoder is established, as for inputted speech signals, a recognition network is established according to the trained HMM acoustic model, the language model, and a dictionary; and S16, the decoder retrieves words matched withto-be-recognized speech feature parameters in the text library according to the step S15, and judges front and behind commonly used words corresponding to the words, and sorting is performed according to grammars and speech recognized in the step S14, and finally the speech signals are converted into text information. The method of the invention has the advantages of high recognition efficiency and accurate voice conversion.

The invention discloses a speech recognition achieving method and system for third-party applications. The method comprises that an auxiliary client arranged at a first terminal obtains a speech input instruction initiated by a main client arranged at a second terminal; the auxiliary client produces a background recording requirement according to the speech input instruction and transmits the background recording requirement to an operation system of the first terminal to request the first terminal to call recording equipment of the second terminal to perform recording; the auxiliary client recognizes speech information obtained through recording by controlling the second terminal through the first terminal, so that the main client processes speech recognition results. By means of the method and the system, third-party speech recognition user operations can be simplified, and the third-party speech recognition efficiency is improved.

The invention discloses a speech recognition processing method and device and a storage medium. The method comprises the following steps of acquiring original audio data; inputting the original audiodata into a first speech recognition engine, and performing speech recognition to obtain first reference text information carrying speech mark data, wherein the first reference text information is text information corresponding to a speech segment recognized from the original audio data, the speech mark data are used for indicating the position of the speech segment in the original audio data, andthe first speech recognition engine is a universal speech recognition engine; determining a keyword in the first reference text information; and inputting the original audio data into a second speechrecognition engine matched with the keyword to obtain target text information, wherein the second speech recognition engine is a speech recognition engine of the field to which the keyword belongs. According to the method, the technical problem that the operation of obtaining text data corresponding to audio data is relatively complex in an audio recognition method provided by related technologies is solved.

The invention discloses a speech recognition method, which comprises the following steps: when an intelligent terminal enters a man-machine interaction mode, the intelligent terminal outputs questioning information and collects the speech information input by a user based on the questioning information; the intelligent terminal analyzes the speech information to obtain its semantic information; the intelligent terminal judges whether the number of characters corresponding to the semantic information is smaller than a preset value; when the number of characters corresponding to the semantic information is smaller than the preset value, the intelligent terminal determines the first operation instruction output by the terminal according to the number of characters. The invention also discloses an intelligent terminal, and a computer-readable storage medium. The speech recognition efficiency and the error rate of the speech recognition are improved.

The embodiment of the invention discloses a voice recognition method and device, a computer readable storage medium and computer equipment. The method comprises the steps: carrying out the feature extraction of to-be-recognized voice information, and obtaining a plurality of feature vectors; calculating the sparseness value of each feature vector, wherein the sparseness value is the relative entropy between the distribution of the self-attention score sequence of each feature vector and the uniform distribution of the self-attention score sequence; determining a first feature vector of which the sparseness value is greater than a preset threshold value and a second feature vector of which the sparseness value is not greater than the preset threshold value; determining a target matrix according to the self-attention calculation result of the first feature vector and the second feature vector; and inputting the target matrix and the feature matrix corresponding to the tag sequence into a classification network for classification processing to obtain a recognition result corresponding to the to-be-recognized voice information. Therefore, the deep learning method is adopted, the calculation amount of the self-attention mechanism in the speech recognition process is reduced, and the speech recognition efficiency is improved.

The invention provides a dual-microphone noise reduction method and device, and a mobile terminal wherein the method comprises the following steps: S10) acquiring a first sound signal by the main microphone; acquiring a second sound signal by the sub-microphone; S20) comparing the amplitude difference value between the first sound signal and the second sound signal with a preset amplitude difference value; S30) based on the comparison result of S20, comparing the signal-to-noise ratio of the first sound signal with a preset signal-to-noise ratio; and S40) according to the comparison result of S30, adjusting the preset amplitude difference value to reduce the noise level of the dual-microphone device and to avoid the fault phenomenon of the uplink voice signal distorted by the position change of a user holding the handset, the monomer microphone difference, the airtight difference of the cavity structure, path attenuation and other factors, therefore, improving the efficiency of sound recognition.

The embodiment of the invention provides a voice recognition method and device and a device for voice recognition. The method specifically comprises the steps: determining voice features of to-be-recognized voice; determining a voice recognition result corresponding to the voice feature by using an acoustic model, wherein the acoustic model comprises a neural network model and a hidden Markov model, the neural network model comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises a feedforward neural network layer and a self-attention neural network layer; and outputting the voice recognition result. The voice recognition method and device and the device for voice recognition can improve the voice recognition efficiency.

The invention provides a voice language recognition method and system based on the confidence degree, aiming to solve the problem that in the existing voice recognition, the language recognition efficiency is low. The method comprises the following steps: S1, extracting a voice segment from each voice segment as a preset voice segment, comparing with a preset language database, acquiring the language information matched with the preset voice segment; S2, acquiring the language confidence degree and the confidence degree average value of each voice segment according to the language information,judging whether the confidence degree mean value is larger than a preset credibility threshold value or not, if yes, the current language is used as the default language of the voice information; S3,if not, screening all the voice segments through a preset screening condition, until the mean value of the language confidence degree is larger than the preset threshold value, and acquiring the voice fragment obtained by the screening, and turning to the step S1. By adopting the voice language recognition method and device, the voice recognition efficiency is improved, and meanwhile, the recognition accuracy of the multi-language voice information is further improved.

The invention relates to the field of artificial intelligence, and provides a speech recognition model training method and device, equipment and a medium, and the method comprises the steps: obtaining a speech sample set containing speech samples; inputting the speech sample into an initial recognition model; obtaining a to-be-processed audio clip through audio enhancement processing; performing teacher acoustic feature extraction through a teacher network in the initial recognition model to obtain a first feature vector, and performing student acoustic feature extraction through a student network in the initial recognition model to obtain a second feature vector; performing alignment comparison processing in combination with a dynamic queue in a teacher network to obtain a loss value; and when the loss value does not reach a preset convergence condition, carrying out iterative updating until convergence, and obtaining a trained speech recognition model. According to the invention, common speech recognition through the teacher network and the student network is realized, and the training efficiency is improved. The method is suitable for the field of artificial intelligence, and can further promote the construction of a smart city.

An embodiment of the invention provides a speech recognition method and a speech recognition device. The speech recognition method comprises the steps of: acquiring a plurality of speech data packets,wherein the packet size of the plurality of speech data packets is less than a first preset size; determining feature information of a speech signal of each frame in each speech data packet respectively; and recognizing the speech signal of each frame according to the feature information. According to the speech recognition method and the speech recognition device, the plurality of speech data packets with small size are acquired, so that the response speed of recognition is accelerated when a decoder is used for recognizing each speech data packet, thereby being conductive to acquire the recognition results quickly, and improving the speech recognition efficiency.