Speech recognition processing method and device, computer equipment and storage medium

Abstract

The invention relates to a speech recognition processing method and device, computer equipment and a storage medium. The method comprises the following steps: preprocessing user speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result; performing text analysis on the user speech spectrum data according to the speech spectrum preprocessing result and a preset recognition mapping relation to obtain a text analysis result; under the condition that the text analysis result is that the user speech spectrum data does not need to be subjected to backflow analysis, performing intention recognition on the text analysis result according to a preset intention recognition model to obtain an intention recognition result; and outputting a target intention corresponding to the user speech spectrum data under the condition that the intention recognition result is that the reflow recognition is not needed. By adopting the method, the voice processing efficiency and accuracy can be improved, the information loss and delay are reduced, the recognition effect is optimized, the voice which fails to explain and is wrongly recognized can be effectively and timely found, and the voice is analyzed and recognized again, so that the stability of voice recognition processing is improved.

Description

Technical Field

[0001] The present application relates to the field of speech recognition technology, and in particular to a speech recognition processing method, apparatus, computer equipment and storage medium. Background Art

[0002] The voice AI platform uses smart remote controls and speakers as the main carrier terminals, and realizes voice search for live broadcasts, on-demand content, and playback of content in large- and medium-screen scenarios in vertical fields, and voice control of smart home devices as the core business scenarios. At the same time, it realizes voice interaction of large-screen value-added applications and skills, creating a capability-intensive voice AI platform to provide users with a more efficient and natural interactive experience, and support the large-scale development of smart remote control and smart speaker businesses across the country.

[0003] In related technologies, after the user's voice command audio for live broadcast, on-demand, playback search, large-screen operation, smart home control, etc. is delivered to the voice AI platform, it goes through the speech recognition technology ASR and natural language processing technology NLP to implement the intent classification and entity extraction of the user's voice command, and then executes the subsequent process processing. The voice AI platform provides multi-modal ASR and NLP capabilities, supporting the analysis and understanding of multiple dialects and various types of scenarios. When users need to use personalized and regionalized automatic speech recognition (ASR) engines and natural language processing (NLP) engines for specific scenarios, they must manually set and switch on the large-screen device.

[0004] However, in the related art, manual setting switching on large-screen devices cannot achieve seamless switching, which not only lacks flexibility but also greatly affects the user experience. In addition, in the related art, a batch execution architecture is adopted, which means that each instruction needs to enter all engines for processing, rather than pre-dividing the instructions in the home scene at the initial stage, which not only prolongs the processing time, but also reduces the success rate of matching, and cannot meet the usage requirements. Summary of the invention

[0005] Based on this, it is necessary to provide a speech recognition processing method, apparatus, computer device, computer-readable storage medium and computer program product that can enhance the accuracy of speech recognition and text classification in order to address the above technical problems.

[0006] In a first aspect, the present application provides a speech recognition processing method. The method comprises:

[0007] Preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result;

[0008] Perform text parsing on the user's spectrogram data according to the spectrogram preprocessing result and the preset recognition mapping relationship to obtain a text parsing result;

[0009] When the text parsing result is user grammar data and does not require reflux parsing, perform intent recognition on the text parsing result according to a preset intent recognition model to obtain an intent recognition result;

[0010] When the intent recognition result is that the intent does not require reflow recognition, the target intent corresponding to the user's speech spectrum data is output.

[0011] In one embodiment, preprocessing the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result includes:

[0012] Generate spectral data based on user spectral data;

[0013] Encoding the spectral data according to the speech feature extraction algorithm to obtain the spectral encoding result;

[0014] Perform difference judgment on the spectrum encoding results according to a preset discriminator to obtain multiple speech recognition labels;

[0015] A plurality of user spectrum data corresponding to the same speech recognition tag is set as a recognition result cluster, and a plurality of recognition result clusters are set as spectrum preprocessing results.

[0016] In one embodiment, encoding the spectral data according to the speech feature extraction algorithm to obtain the spectral encoding result includes:

[0017] Extracting features from the spectral data according to a speech feature extraction algorithm to obtain spectral feature data;

[0018] Performing position coding on the spectral feature data to obtain spectral position coding data;

[0019] Perform intention recognition on the spectrogram position coding data to obtain the spectrogram coding result.

[0020] In one embodiment, text parsing is performed on the user's speech spectrum data according to the speech spectrum preprocessing result and the preset recognition mapping relationship to obtain the text parsing result, including:

[0021] The recognition engine is screened according to the speech recognition tag and the preset recognition mapping relationship to obtain a screened engine;

[0022] Perform speech analysis on the speech recognition result cluster according to the filtered engine to obtain a speech analysis result;

[0023] When the speech analysis result is that all speech recognition results are analyzed successfully, a plurality of analyzed texts are obtained and all analyzed texts are set as text analysis results.

[0024] In one embodiment, after performing speech analysis on the speech recognition result cluster according to the post-screening engine to obtain the speech analysis result, the method further includes:

[0025] When the speech analysis result is at least one speech recognition result of the reflow analysis, obtaining the speech that failed to be analyzed;

[0026] Preprocess the speech that failed to be parsed according to the preset speech recognition algorithm to obtain a secondary preprocessing result;

[0027] Perform text analysis on the speech that failed to be analyzed according to the secondary preprocessing result and the preset recognition mapping relationship to obtain a secondary text analysis result;

[0028] If the secondary text parsing result is a successful parsing, mark the secondary text parsing result;

[0029] When the secondary text analysis result is analysis failure, the analysis failure speech is stored in the analysis failure library and analysis failure alarm information is output.

[0030] In one embodiment, when the text parsing result is user grammar data and does not require reflow parsing, the text parsing result is subjected to intent recognition according to a preset intent recognition model to obtain an intent recognition result, including:

[0031] When the text analysis result is that the user's speech data does not need reflux analysis, feature extraction is performed on the speech analysis text according to the text feature extraction algorithm to obtain the extracted text;

[0032] Perform extraction operations on the extracted text according to the semantic extraction network to obtain semantic extraction results;

[0033] When the semantic extraction result is a correct recognition of the text semantics, the target text semantics is obtained and the target text semantics is set as the intent recognition result.

[0034] In one embodiment, after performing an extraction operation on the extracted text according to the semantic extraction network to obtain a semantic extraction result, the method further includes:

[0035] When the semantic extraction result is text semantic reflux recognition, obtaining the incorrectly recognized text;

[0036] Perform feature extraction on the wrongly recognized text according to the text feature extraction algorithm to obtain a secondary extracted text;

[0037] Perform an extraction operation on the secondary extraction text according to the semantic extraction network to obtain the secondary extraction result;

[0038] When the secondary extraction result is successful recognition of text semantics, a secondary intent recognition result is obtained and marked;

[0039] In the case where the secondary extraction result is a text semantic error recognition, the error recognition text is stored in an error recognition library and a recognition error alarm message is output.

[0040] In a second aspect, the present application also provides a speech recognition processing device. The device comprises:

[0041] The preprocessing module is used to preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain the speech spectrum preprocessing result;

[0042] The text parsing module is used to perform text parsing on the user's speech spectrum data according to the speech spectrum preprocessing result and the preset recognition mapping relationship to obtain the text parsing result;

[0043] The intention recognition module is used to perform intention recognition on the text parsing result according to a preset intention recognition model to obtain an intention recognition result when the text parsing result is a successful parsing of the user's lexical data;

[0044] The intent output module is used to output the target intent corresponding to the user's speech spectrum data when the intent recognition result is a successful intent recognition.

[0045] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0046] Preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result;

[0047] Perform text parsing on the user's spectrogram data according to the spectrogram preprocessing result and the preset recognition mapping relationship to obtain a text parsing result;

[0048] When the text parsing result is a successful parsing of the user's lexical data, the text parsing result is subjected to intent recognition according to a preset intent recognition model to obtain an intent recognition result;

[0049] When the intention recognition result is that the intention is successfully recognized, the target intention corresponding to the user's speech spectrum data is output.

[0050] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0051] Preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result;

[0052] Perform text parsing on the user's spectrogram data according to the spectrogram preprocessing result and the preset recognition mapping relationship to obtain a text parsing result;

[0053] When the text parsing result is a successful parsing of the user's lexical data, the text parsing result is subjected to intent recognition according to a preset intent recognition model to obtain an intent recognition result;

[0054] When the intention recognition result is that the intention is successfully recognized, the target intention corresponding to the user's speech spectrum data is output.

[0055] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0056] Preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result;

[0057] Perform text parsing on the user's spectrogram data according to the spectrogram preprocessing result and the preset recognition mapping relationship to obtain a text parsing result;

[0058] When the text parsing result is a successful parsing of the user's lexical data, the text parsing result is subjected to intent recognition according to a preset intent recognition model to obtain an intent recognition result;

[0059] When the intention recognition result is that the intention is successfully recognized, the target intention corresponding to the user's speech spectrum data is output.

[0060] The above-mentioned speech recognition processing method, device, computer equipment, storage medium and computer program product, after obtaining the user's spectrum data, use a preset speech recognition algorithm to perform feature extraction, encoding processing, speech division and other pre-processing operations on the user's spectrum data, and then obtain the spectrum pre-processing result, and then screen the recognition engine through the preset recognition mapping relationship and the speech recognition label in the spectrum pre-processing result, and use the post-recognition engine to perform text recognition and analysis on the spectrum pre-processing result to obtain the text analysis result, and then use the preset intention recognition model to perform semantic recognition on the text analysis result, and then obtain the post-recognition semantics, that is, the target intention corresponding to the user's spectrum data, which can improve the processing efficiency and accuracy of speech, reduce information loss and delay, and optimize the recognition effect; at the same time, there are reflux parsing schemes and reflux recognition schemes in the parsing and recognition processes, which can effectively and timely discover speech that fails to be interpreted or recognized incorrectly, and re-parse and recognize it, thereby improving the stability of speech recognition processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] FIG1 is a schematic diagram of a flow chart of a speech recognition processing method in one embodiment;

[0062] FIG2 is a flow chart of a step of encoding spectrum data in one embodiment;

[0063] FIG3 is a schematic diagram of a flow chart of a semantic recognition step in one embodiment;

[0064] FIG4 is a flowchart of a speech recognition processing method in one embodiment;

[0065] FIG5 is a structural block diagram of a speech recognition processing device in one embodiment;

[0066] FIG. 6 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0068] In one embodiment, as shown in FIG1 , a speech recognition processing method is provided. This embodiment is illustrated by applying the method to a terminal. It is understandable that the method can also be applied to a server, or to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0069] Step 102, preprocessing the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result.

[0070] Among them, the preset speech recognition algorithm includes a speech feature extraction algorithm.

[0071] Exemplarily, after the user issues a voice command, the voice command is obtained and converted into corresponding user spectrum data, and then the user spectrum data is subjected to feature extraction using a voice feature extraction algorithm. And a position encoding operation is performed on the features extracted from each user spectrum data to characterize the position of each word in the sentence and the positional relationship between them. Afterwards, the position-encoded spectrum feature data is input into the encoder to obtain a spectrum encoding result. Finally, the spectrum encoding result is judged and recognized, and a plurality of speech recognition tags and their corresponding recognition result clusters are obtained, and the plurality of speech recognition tags and their corresponding recognition result clusters are set as spectrum preprocessing results.

[0072] Step 104: perform text analysis on the user's spectrogram data according to the spectrogram preprocessing result and the preset recognition mapping relationship to obtain a text analysis result.

[0073] Among them, the preset recognition mapping relationship is the mapping relationship between the speech recognition tag and the recognition engine, and the spectrum preprocessing result includes the speech recognition tag and its corresponding speech recognition result cluster.

[0074] Exemplarily, after obtaining the spectrum preprocessing results including multiple speech recognition tags and their corresponding recognition result clusters, a suitable recognition engine is selected according to the mapping relationship between each speech recognition tag and a preset recognition, and then the speech recognition result cluster corresponding to each speech recognition tag is input into the corresponding recognition engine for text parsing, thereby obtaining the parsed text corresponding to each user's spectrum data.

[0075] During the parsing process, if there is user spectral data that cannot be parsed, the user spectral data is reflowed to re-process the data, and then the speech recognition label is re-obtained. The reflowed user spectral data is re-parsed based on the obtained speech recognition label.

[0076] If the secondary analysis is successful, the text analysis result is marked; if it still cannot be analyzed, the user's spectrum data is stored in the database and an email is sent to notify the operator for manual processing.

[0077] Step 106, when the text parsing result is user grammar data that does not require reflow parsing, perform intent recognition on the text parsing result according to a preset intent recognition model to obtain an intent recognition result.

[0078] Among them, the preset intent recognition model includes a text feature extraction algorithm and a semantic extraction network, and the text analysis results include speech analysis text.

[0079] For example, after obtaining the text parsing results, feature extraction is performed on the parsed text corresponding to each user's lexical data to obtain the feature-extracted text. Then, the individual words in the feature-extracted text are stacked and extracted, and the dependency between words is captured by combining the semantic extraction network, thereby obtaining the intent recognition results corresponding to each user's lexical data.

[0080] If there is a parsed text corresponding to the user's speech data that cannot be recognized, the parsed text corresponding to the user's speech data is reflowed to re-perform intent recognition, thereby obtaining a secondary recognition result corresponding to the user's speech data.

[0081] If the second recognition is successful, the intent recognition result is marked; if it is still unrecognizable, the user's speech data and its corresponding parsed text are recorded in the database, and an email is sent to notify the operator for manual processing.

[0082] Step 108, when the intention recognition result is that the intention does not require reflow recognition, output the target intention corresponding to the user's speech spectrum data.

[0083] Exemplarily, after the parsed text corresponding to the user's spectrogram data is successfully recognized, the target intent corresponding to the user's spectrogram data is output.

[0084] In the above-mentioned speech recognition processing method, after obtaining the user's spectrum data, the preset speech recognition algorithm is used to perform feature extraction, encoding processing, speech division and other preprocessing operations on the user's spectrum data, so as to obtain the spectrum preprocessing result, and then the recognition engine is screened by the preset recognition mapping relationship and the speech recognition label in the spectrum preprocessing result, and the post-recognition engine is used to perform text recognition and analysis on the spectrum preprocessing result to obtain the text analysis result, and then the preset intention recognition model is used to perform semantic recognition on the text analysis result, so as to obtain the post-recognition semantics, that is, the target intention corresponding to the user's spectrum data, which can improve the processing efficiency and accuracy of speech, reduce information loss and delay, and optimize the recognition effect; at the same time, there are reflux parsing schemes and reflux recognition schemes in the parsing and recognition processes, respectively, which can effectively and timely discover the speech that fails to be interpreted and recognized incorrectly, and re-parse and recognize it, thereby improving the stability of speech recognition processing.

[0085] In an exemplary embodiment, preprocessing the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result includes:

[0086] The user's spectrogram data is encoded and processed according to the speech feature extraction algorithm to obtain the spectrogram encoding result; the spectrogram encoding result is judged by the preset discriminator to obtain multiple speech recognition labels; multiple user spectrogram data corresponding to the same speech recognition label are set as a recognition result cluster, and multiple recognition result clusters are set as spectrogram preprocessing results.

[0087] Among them, the preset speech recognition algorithm includes a speech feature extraction algorithm.

[0088] Exemplarily, after the user issues a voice command, the voice command is acquired and converted into corresponding user spectrogram data, and then a voice feature extraction algorithm is used to perform feature extraction and encoding processing on the user spectrogram data to obtain a spectrogram encoding result.

[0089] The obtained spectrogram encoding results are then input into the discriminator to judge the difference between the two spectra, and then obtain the recognition result clusters corresponding to different types of speech streams. The discriminator calculation is as follows:

[0090]

[0091] Among them, H is the spectrum encoding result.

[0092] The above steps capture global information, improve flexibility and computational efficiency, while capturing long-distance dependencies.

[0093] In an exemplary embodiment, as shown in FIG2 , encoding the user's speech spectrum data according to the speech feature extraction algorithm to obtain the speech spectrum encoding result includes:

[0094] Step 202: extract features from the user's speech spectrum data according to a speech feature extraction algorithm to obtain speech spectrum feature data.

[0095] For example, after obtaining the user's speech spectrogram data corresponding to the voice command, the convolutional neural network (CNN) is used to extract features from the speech spectrogram, which can help the model automatically learn the local features of the audio signal, such as the pitch, frequency pattern, and speech features in the audio. Then, the GELU (Gaussian Error Linear Unit) activation function is combined to add nonlinearity to the convolution output to avoid information loss.

[0096] Step 204: Position-code the spectral feature data to obtain spectral position-coded data.

[0097] For example, after extracting features, we need to add positional encoding to each word in order to understand their temporal order and relationship. The specific encoding calculation method is as follows:

[0098]

[0099]

[0100] Among them, pos represents the position of each word in the sentence, 2i and 2i+1 represent the dimensions of the position encoding, and their value range is .

[0101] Step 206: perform intent recognition on the spectrogram position coding data to obtain a spectrogram coding result.

[0102] Exemplarily, after position coding, the preliminarily processed spectrum data is input into the encoder for intention recognition. The specific calculation method is:

[0103]

[0104]

[0105]

[0106]

[0107] Among them, Q, K and V are the same spectral input, n is the number of attention heads, W is the corresponding parameter, and A is the adjacency matrix.

[0108] In an exemplary embodiment, text parsing is performed on the user's speech spectrum data according to the speech spectrum preprocessing result and the preset recognition mapping relationship to obtain the text parsing result, including:

[0109] The recognition engines are screened according to the speech recognition tags and the preset recognition mapping relationship to obtain the screened engines; speech analysis is performed on the speech recognition result cluster according to the screened engines to obtain the speech analysis results; when the speech analysis results are successful analysis of all speech recognition results, multiple analysis texts are obtained and all the analysis texts are set as text analysis results.

[0110] The spectrum preprocessing result includes speech recognition tags and their corresponding speech recognition result clusters.

[0111] Exemplarily, the recognition engines are screened by voice recognition tags and preset recognition mapping relationships. Through this screening process, the system can select the engine that best suits the current task. For example, among multiple ASR (automatic speech recognition) engines, the best engine is selected for subsequent processing based on the information provided by the voice recognition tags. The screened engine parses the voice recognition result cluster to obtain more accurate voice parsing results. In this process, the system first ensures that all voice recognition results are successfully parsed. If the parsing is successful, multiple parsed texts will be obtained and integrated into the final text parsing result.

[0112] In an exemplary embodiment, after performing speech analysis on the speech recognition result cluster according to the post-screening engine to obtain the speech analysis result, the method further includes:

[0113] In the case where the speech analysis result is at least one speech recognition result of the reflux analysis, the speech that failed to be analyzed is obtained; the speech that failed to be analyzed is preprocessed according to the preset speech recognition algorithm to obtain a secondary preprocessing result; the speech that failed to be analyzed is text analyzed according to the secondary preprocessing result and the preset recognition mapping relationship to obtain a secondary text analysis result; in the case where the secondary text analysis result is a successful analysis, the secondary text analysis result is marked; in the case where the secondary text analysis result is a analysis failure, the speech that failed to be analyzed is stored in a analysis failure library and a analysis failure alarm message is output.

[0114] For example, if the selected engine cannot parse successfully, the failed parsing voice needs to be reflowed for reprocessing, that is, re-preprocessing, text parsing and other operations. Generally, the system log file is checked regularly to find the failed parsing instructions recorded therein. Once the failed parsing instruction is found, the voice reflow system will reprocess it.

[0115] The text parsing results are then parsed and judged. If the reprocessing is successful, the processing results will be marked; if the reprocessing fails, the data reflux module will process the storage record and send an email to notify the operator for manual processing.

[0116] In an exemplary embodiment, as shown in FIG3 , when the text parsing result is user grammar data that does not require reflow parsing, the text parsing result is subjected to intent recognition according to a preset intent recognition model to obtain an intent recognition result, including:

[0117] Step 302: When the text analysis result indicates that the user's speech data does not require reflow analysis, feature extraction is performed on the speech analysis text according to a text feature extraction algorithm to obtain an extracted text.

[0118] Among them, the preset intent recognition model includes a text feature extraction algorithm and a semantic extraction network, and the text analysis results include speech analysis text.

[0119] For example, when the text parsing result is that the user's speech data does not need to be parsed back and forth, that is, the parsed text is successfully parsed, and text feature extraction is performed on the successfully parsed speech parsed text, mainly using a feature extraction module based on a convolutional neural network to capture local features, as shown below:

[0120]

[0121]

[0122]

[0123] Among them, MaxPool represents the pooling operation, CNN represents the convolution operation, and Dropout represents the discarding operation.

[0124] Step 304: perform an extraction operation on the extracted text according to the semantic extraction network to obtain a semantic extraction result.

[0125] For example, semantic recognition processing is performed on the text after feature extraction, and individual words are stacked at intervals. The calculation method is as follows:

[0126]

[0127] in, Indicates stacking the data in the same dimension.

[0128] Through this processing method, for M1 and M nThere is not enough data for stacking, so it is randomly initialized and the semantics is extracted through a bidirectional recurrent neural network. In actual operation, a gated recurrent neural network is used, and the calculation method is as follows:

[0129]

[0130]

[0131]

[0132]

[0133] Among them, H n-1 represents the hidden layer data, A represents the static adjacency matrix, represents the superposition operation, is the activation function sigmoid, is a tangent function. And when n=1, H0 is randomly initialized. Through the above calculation method, we extract the corresponding intention purpose.

[0134] Step 306, when the semantic extraction result is a correct recognition of the text semantics, the target text semantics is obtained and the target text semantics is set as the intention recognition result.

[0135] Exemplarily, when the semantic extraction result is a correct recognition of the text semantics, the extracted intent, i.e., the target text semantics, is output as the intent recognition result.

[0136] In an exemplary embodiment, after performing an extraction operation on the extracted text according to the semantic extraction network to obtain a semantic extraction result, the method further includes:

[0137] In the case where the semantic extraction result is text semantic reflux recognition, the incorrectly recognized text is obtained; feature extraction is performed on the incorrectly recognized text according to the text feature extraction algorithm to obtain a secondary extracted text; extraction operation is performed on the secondary extracted text according to the semantic extraction network to obtain a secondary extraction result; in the case where the secondary extraction result is a successful recognition of text semantics, a secondary intent recognition result is obtained and marked; in the case where the secondary extraction result is an incorrect recognition of text semantics, the incorrectly recognized text is stored in an error recognition library and recognition error alarm information is output.

[0138] For example, if the semantics cannot be recognized through the above calculation method, the incorrectly recognized text will be reflowed for re-semantic recognition processing, that is, the text feature extraction, semantic recognition calculation and other operations will be re-performed to obtain the result of secondary semantic recognition. If the reprocessing is successful, the processing result will be marked; if the reprocessing fails, the data reflow module will process the storage record and send an email to notify the operator to manually process it.

[0139] In an exemplary embodiment, a speech recognition processing method is provided, the method comprising the following steps:

[0140] The user's speech spectrum data is subjected to feature extraction according to a speech feature extraction algorithm to obtain speech spectrum feature data.

[0141] Position encoding is performed on the spectral feature data to obtain spectral position encoding data.

[0142] Perform intention recognition on the spectrogram position coding data to obtain the spectrogram coding result.

[0143] The spectrum encoding results are judged differently according to the preset discriminator to obtain multiple speech recognition labels.

[0144] A plurality of user spectrum data corresponding to the same speech recognition tag is set as a recognition result cluster, and a plurality of recognition result clusters are set as spectrum preprocessing results.

[0145] The recognition engines are filtered according to the speech recognition tags and the preset recognition mapping relationship to obtain the filtered engines.

[0146] The speech recognition result cluster is subjected to speech analysis according to the filtered engine to obtain a speech analysis result.

[0147] When the speech analysis result is at least one speech recognition result of the reflow analysis, the analysis failure speech is obtained.

[0148] The speech that failed to be parsed is preprocessed according to the preset speech recognition algorithm to obtain a secondary preprocessing result.

[0149] According to the secondary preprocessing result and the preset recognition mapping relationship, text analysis is performed on the speech that failed to be analyzed to obtain a secondary text analysis result.

[0150] When the secondary text parsing result is a successful parsing, the secondary text parsing result is marked.

[0151] When the secondary text analysis result is analysis failure, the analysis failure speech is stored in the analysis failure library and analysis failure alarm information is output.

[0152] When the speech analysis result is that all speech recognition results are analyzed successfully, a plurality of analyzed texts are obtained and all analyzed texts are set as text analysis results.

[0153] When the text analysis result is that the user's speech data does not need reflow analysis, feature extraction is performed on the speech analysis text according to the text feature extraction algorithm to obtain the extracted text.

[0154] The extracted text is extracted according to the semantic extraction network to obtain the semantic extraction result.

[0155] When the semantic extraction result is text semantic reflow recognition, the misrecognition text is obtained.

[0156] The feature extraction of the wrongly recognized text is performed according to the text feature extraction algorithm to obtain the secondary extracted text.

[0157] The secondary extraction text is extracted according to the semantic extraction network to obtain the secondary extraction result.

[0158] When the secondary extraction result is a successful recognition of text semantics, a secondary intent recognition result is obtained and marked.

[0159] In the case where the secondary extraction result is a text semantic error recognition, the error recognition text is stored in an error recognition library and a recognition error alarm message is output.

[0160] When the semantic extraction result is a correct recognition of the text semantics, the target text semantics is obtained and the target text semantics is set as the intent recognition result.

[0161] When the intent recognition result is that the intent does not require reflow recognition, the target intent corresponding to the user's speech spectrum data is output.

[0162] The above specific steps are applicable to the process framework diagram shown in FIG4 .

[0163] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0164] Based on the same inventive concept, the embodiment of the present application also provides a speech recognition processing device for implementing the speech recognition processing method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more speech recognition processing device embodiments provided below can refer to the limitations on the speech recognition processing method above, and will not be repeated here.

[0165] In one embodiment, as shown in FIG. 5 , a speech recognition processing device is provided, including: a preprocessing module 502 , a text parsing module 504 , an intention recognition module 506 and an intention output module 508 , wherein:

[0166] The preprocessing module 502 is used to preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result.

[0167] The text analysis module 504 is used to perform text analysis on the user's speech spectrum data according to the speech spectrum preprocessing result and the preset recognition mapping relationship to obtain a text analysis result.

[0168] The intention recognition module 506 is used to perform intention recognition on the text analysis result according to a preset intention recognition model to obtain an intention recognition result when the text analysis result is user speech spectrum data that does not require reflow analysis.

[0169] The intention output module 508 is used to output the target intention corresponding to the user speech data when the intention recognition result is that the intention does not require reflow recognition.

[0170] In an exemplary embodiment, the preprocessing module 502 is also used to encode the user spectrogram data according to the speech feature extraction algorithm to obtain the spectrogram encoding result; perform difference judgment on the spectrogram encoding result according to a preset discriminator to obtain multiple speech recognition labels; set multiple user spectrogram data corresponding to the same speech recognition label as a recognition result cluster, and set the multiple recognition result clusters as spectrogram preprocessing results.

[0171] In an exemplary embodiment, the preprocessing module 502 is also used to perform feature extraction on the user's spectrogram data according to a speech feature extraction algorithm to obtain spectrogram feature data; perform position encoding on the spectrogram feature data to obtain spectrogram position encoding data; perform intent recognition on the spectrogram position encoding data to obtain a spectrogram encoding result.

[0172] In an exemplary embodiment, the text parsing module 504 is also used to filter the recognition engine according to the speech recognition tag and the preset recognition mapping relationship to obtain the filtered engine; perform speech analysis on the speech recognition result cluster according to the filtered engine to obtain the speech analysis result; when the speech analysis result is a successful analysis of all speech recognition results, obtain multiple parsed texts and set all the parsed texts as text parsing results.

[0173] In an exemplary embodiment, the speech recognition processing device is also used to obtain a speech that failed to be parsed when the speech analysis result is at least one speech recognition result of a reflux analysis; preprocess the speech that failed to be parsed according to a preset speech recognition algorithm to obtain a secondary preprocessing result; perform text analysis on the speech that failed to be parsed according to the secondary preprocessing result and a preset recognition mapping relationship to obtain a secondary text analysis result; when the secondary text analysis result is a successful analysis, mark the secondary text analysis result; when the secondary text analysis result is a parsing failure, store the speech that failed to be parsed in a parsing failure library and output a parsing failure alarm message.

[0174] In an exemplary embodiment, the intent recognition module 506 is also used to perform feature extraction on the speech analysis text according to a text feature extraction algorithm to obtain an extracted text when the text parsing result is that the user's speech data does not require reflow parsing; perform extraction operations on the extracted text according to a semantic extraction network to obtain a semantic extraction result; and when the semantic extraction result is a correct recognition of the text semantics, obtain the target text semantics and set the target text semantics as the intent recognition result.

[0175] In an exemplary embodiment, the intent recognition module 506 is also used to obtain incorrectly recognized text when the semantic extraction result is text semantic reflux recognition; perform feature extraction on the incorrectly recognized text according to a text feature extraction algorithm to obtain secondary extracted text; perform extraction operations on the secondary extracted text according to a semantic extraction network to obtain secondary extraction results; when the secondary extraction result is successful recognition of text semantics, obtain a secondary intent recognition result and mark the secondary intent recognition result; when the secondary extraction result is incorrect recognition of text semantics, store the incorrectly recognized text in an error recognition library and output recognition error alarm information.

[0176] Each module in the above-mentioned speech recognition processing device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0177] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG6. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store voice data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a speech recognition processing method is implemented.

[0178] Those skilled in the art will understand that the structure shown in FIG. 6 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0179] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0180] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0181] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0183] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0184] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0185] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A speech recognition processing method, characterized in that: The method comprises: Preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain a speech spectrum preprocessing result; Performing text parsing on the user's speech spectrum data according to the speech spectrum preprocessing result and the preset recognition mapping relationship to obtain a text parsing result; When the text parsing result is user speech data and does not require reflow parsing, performing intent recognition on the text parsing result according to a preset intent recognition model to obtain an intent recognition result; When the intention recognition result is that the intention does not require reflow recognition, the target intention corresponding to the user spectrum data is output.

2. The speech recognition processing method according to claim 1, characterized in that: The preset speech recognition algorithm includes a speech feature extraction algorithm; The preprocessing of the user's speech spectrum data according to the preset speech recognition algorithm to obtain the speech spectrum preprocessing result includes: Encode the user's spectral data according to the speech feature extraction algorithm to obtain the spectral encoding result; Performing difference judgment on the spectrum encoding results according to a preset discriminator to obtain multiple speech recognition labels; The plurality of user speech spectrum data corresponding to the same speech recognition tag are set as a recognition result cluster, and the plurality of recognition result clusters are set as speech spectrum preprocessing results.

3. The speech recognition processing method according to claim 2, characterized in that: The encoding process of the user's speech spectrum data according to the speech feature extraction algorithm to obtain the speech spectrum encoding result includes: Extracting features from the user's speech spectrum data according to the speech feature extraction algorithm to obtain speech spectrum feature data; Performing position coding on the spectral feature data to obtain spectral position coding data; Intention recognition is performed on the spectrogram position coding data to obtain a spectrogram coding result.

4. The speech recognition processing method according to claim 1, characterized in that: The spectrum preprocessing result includes a speech recognition tag and its corresponding speech recognition result cluster; The performing text parsing on the user's speech spectrum data according to the speech spectrum preprocessing result and the preset recognition mapping relationship to obtain the text parsing result includes: Filter the recognition engines according to the speech recognition tags and the preset recognition mapping relationship to obtain filtered engines; Performing speech analysis on the speech recognition result cluster according to the filtered engine to obtain a speech analysis result; When the speech analysis result is that all speech recognition results are analyzed successfully, a plurality of analyzed texts are obtained and all analyzed texts are set as text analysis results.

5. The speech recognition processing method according to claim 4, characterized in that: After performing speech analysis on the speech recognition result cluster according to the post-screening engine to obtain the speech analysis result, the method further includes: When the speech analysis result is at least one speech recognition result of the reflow analysis, obtaining the speech in which the analysis failed; Preprocessing the failed-to-parse speech according to a preset speech recognition algorithm to obtain a secondary preprocessing result; Performing text analysis on the speech that failed to be analyzed according to the secondary preprocessing result and the preset recognition mapping relationship to obtain a secondary text analysis result; If the secondary text parsing result is a successful parsing, marking the secondary text parsing result; When the secondary text analysis result is analysis failure, the analysis failure speech is stored in a analysis failure library and analysis failure alarm information is output.

6. The speech recognition processing method according to claim 1, characterized in that: The preset intention recognition model includes a text feature extraction algorithm and a semantic extraction network; the text analysis result includes a speech analysis text; when the text analysis result is user speech data and does not require reflux analysis, the text analysis result is subjected to intention recognition according to the preset intention recognition model to obtain an intention recognition result, including: When the text analysis result is that the user's speech data does not need reflow analysis, feature extraction is performed on the speech analysis text according to a text feature extraction algorithm to obtain an extracted text; Performing an extraction operation on the extracted text according to a semantic extraction network to obtain a semantic extraction result; When the semantic extraction result is a correct recognition of the text semantics, the target text semantics is obtained and the target text semantics is set as the intention recognition result.

7. The speech recognition processing method according to claim 6, characterized in that: After performing an extraction operation on the extracted text according to the semantic extraction network to obtain a semantic extraction result, the method further includes: When the semantic extraction result is text semantic reflux recognition, obtaining an erroneous recognition text; Perform feature extraction on the error-recognized text according to the text feature extraction algorithm to obtain a secondary extracted text; Performing an extraction operation on the secondary extracted text according to the semantic extraction network to obtain a secondary extraction result; When the secondary extraction result is a successful recognition of text semantics, obtaining a secondary intent recognition result and marking the secondary intent recognition result; In the case where the secondary extraction result is a text semantic error recognition, the error recognition text is stored in an error recognition library and recognition error alarm information is output.

8. A speech recognition processing device, characterized in that: The device comprises: The preprocessing module is used to preprocess the user's speech spectrum data according to a preset speech recognition algorithm to obtain the speech spectrum preprocessing result; A text parsing module, used to perform text parsing on the user's speech spectrum data according to the speech spectrum preprocessing result and a preset recognition mapping relationship to obtain a text parsing result; An intention recognition module is used to perform intention recognition on the text analysis result according to a preset intention recognition model to obtain an intention recognition result when the text analysis result is user speech data and does not require reflow analysis; The intention output module is used to output the target intention corresponding to the user speech spectrum data when the intention recognition result is that the intention does not require reflow recognition.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images