Intention recognition method and device, electronic equipment and readable medium

Abstract

This application relates to an intent recognition method, apparatus, electronic device, and readable medium. The method includes: acquiring audio data of a target object; inputting the audio data into a preset recognition model for intent recognition to obtain a recognition result, wherein the preset recognition model is pre-trained using a first training set and a second training set, and the training data used in the first and second training sets are different; and determining the intent of the audio data using the recognition result. By directly inputting the user's voice data into the preset recognition model for intent recognition to determine the user's intent, this solves the problem that the voice recognition process must first be converted into text, which leads to a cumbersome recognition process and affects recognition accuracy.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an intent recognition method, apparatus, electronic device, and readable medium. Background Technology

[0002] With the development of intelligent devices, users can control devices directly through voice. Currently, the recognition of user voice intent is basically done by first processing the user's voice and then converting it into text. Then, the user's intent is recognized by performing operations such as word segmentation and keyword extraction on the text, and then issuing corresponding commands. This recognition method of converting speech into text first not only increases the recognition steps, but also affects the recognition accuracy when there are deviations in the speech-to-text conversion.

[0003] There is currently no effective solution to the problem that “converting speech to text first not only increases the number of recognition steps, but also affects recognition accuracy when there are deviations in the speech-to-text conversion process.” Summary of the Invention

[0004] This application provides an intent recognition method, apparatus, electronic device, and readable medium to solve the aforementioned technical problem that "the recognition method of first converting speech to text not only increases the number of recognition steps, but also affects the recognition accuracy when there are deviations in the speech-to-text conversion."

[0005] According to one aspect of the embodiments of this application, this application provides an intent recognition method, including: acquiring audio data of a target object; inputting the audio data into a preset recognition model for intent recognition and obtaining a recognition result, wherein the preset recognition model is obtained by pre-training using a first training set and a second training set, and the training data used in the first training set and the second training set are different; and determining the intent of the audio data using the recognition result.

[0006] Optionally, the audio data is input into a preset recognition model for intent recognition, and the recognition result is obtained by: extracting speech features of each frame in the audio data according to preset dimensions to obtain a feature matrix; determining the matching probability of each frame speech feature in the feature matrix with each preset feature in the preset feature library to obtain a probability matrix, wherein the preset feature library is obtained during the training of the preset recognition model; converting the probability matrix into target machine code, wherein the target machine code includes the machine code of each frame speech feature; decoding the target machine code to obtain the target intent code, and determining the target intent code as the recognition result.

[0007] Optionally, extracting speech features from each frame of the audio data according to a preset dimension to obtain a feature matrix includes: filtering out impurity data in the audio data to obtain first audio data; dividing the first audio data into frames to obtain multiple frames of second audio data; extracting features from the multiple frames of second audio data using a preset dimension, and integrating the extracted speech features from each frame to obtain a feature matrix.

[0008] Optionally, the matching probability of each frame speech feature in the feature matrix with each preset feature in the preset feature library is determined separately to obtain the probability matrix. This includes: determining multiple Jaccard coefficients for the speech features and each preset feature in the preset feature library, and determining the multiple Jaccard coefficients as the matching probability of the speech features with each preset feature; and integrating the multiple matching probabilities to obtain the probability matrix.

[0009] Optionally, converting the probability matrix into target machine code includes: extracting multiple matching probabilities corresponding to single-frame speech features in the probability matrix, and determining the target preset feature corresponding to the matching probability with the largest value, wherein the multiple preset features include the target preset feature; determining the preset machine code carried by the target preset feature as the single-frame machine code of the single-frame speech feature, wherein the preset machine code is obtained by encoding a preset intent code; and concatenating the multiple single-frame machine codes in chronological order to obtain the target machine code.

[0010] Optionally, determining the intent of the audio data using the recognition results includes: obtaining an intent association table, wherein the intent association table is obtained during the training of a preset recognition model, and the intent association table is used to represent the correspondence between intent codes and intents; obtaining the target intent code in the recognition results, and determining the intent data corresponding to the target intent code in the intent association table based on the correspondence; and determining the intent data as the intent of the audio data.

[0011] Optionally, after determining the intent data corresponding to the target intent code, the method further includes: acquiring device data and operation data from the intent data; and sending the operation data to the target device corresponding to the device data, so that the target device performs the target operation corresponding to the operation data.

[0012] According to another aspect of the embodiments of this application, this application provides an intent recognition device, including: an audio acquisition module for acquiring audio data of a target object; an audio recognition module for inputting the audio data into a preset recognition model for intent recognition and obtaining a recognition result, wherein the preset recognition model is obtained by pre-training using a first training set and a second training set, and the training data used in the first training set and the second training set are different; and an intent determination module for determining the intent of the audio data using the recognition result.

[0013] According to another aspect of the embodiments of this application, this application provides an electronic device, including a memory, a processor, a communication interface and a communication bus. The memory stores a computer program that can run on the processor. The memory and the processor communicate with each other through the communication bus and the communication interface. When the processor executes the computer program, it implements the steps of the above method.

[0014] According to another aspect of the embodiments of this application, this application also provides a computer-readable medium having processor-executable non-volatile program code that causes the processor to perform the above-described method.

[0015] Compared with related technologies, the technical solutions provided in this application have the following advantages:

[0016] This application provides an intent recognition method, comprising: acquiring audio data of a target object; inputting the audio data into a preset recognition model for intent recognition to obtain a recognition result, wherein the preset recognition model is pre-trained using a first training set and a second training set, and the training data used in the first and second training sets are different; and determining the intent of the audio data using the recognition result. By directly inputting the user's voice data into the preset recognition model for intent recognition to determine the user's intent, this method solves the problem that in the voice recognition process, converting speech to text first is cumbersome and affects recognition accuracy. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of an optional intent recognition method provided according to an embodiment of this application;

[0020] Figure 2 A flowchart illustrating another optional intent recognition method provided according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram illustrating an optional intent recognition method according to an embodiment of this application.

[0022] Figure 4 This is a block diagram of an optional intent recognition device provided according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of an optional electronic device structure provided in an embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustration and has no specific meaning in itself. Therefore, "module" and "part" may be used interchangeably.

[0026] With the development of intelligent devices, users can control devices directly through voice. Currently, the recognition of user voice intent is basically done by first processing the user's voice and then converting it into text. Then, the user's intent is recognized by performing operations such as word segmentation and keyword extraction on the text, and then issuing corresponding commands. This recognition method of converting speech into text first not only increases the recognition steps, but also affects the recognition accuracy when there are deviations in the speech-to-text conversion.

[0027] To address the problems mentioned in the background art, according to one aspect of an embodiment of this application, an intent recognition method is provided, such as... Figure 1 As shown, it includes:

[0028] Step 101: Obtain the audio data of the target object;

[0029] Step 103: Input the audio data into the preset recognition model for intent recognition and obtain the recognition result. The preset recognition model is trained in advance using the first training set and the second training set. The training data used in the first training set and the second training set are different.

[0030] Step 105: Use the recognition results to determine the intent of the audio data.

[0031] This application directly inputs the audio data of the target object into a preset recognition model for intent recognition, and then uses the recognition results to determine the intent of the audio data.

[0032] Before intent recognition, a machine model needs to be built and trained. This machine model can perform certain filtering operations on the user's speech, extract the useful parts of the speech through filtering, and then perform feature extraction and similarity matching based on the Jaccard coefficient to determine the intent contained in the speech.

[0033] The Jaccard coefficient, also known as the Jaccard similarity coefficient, is used to compare the similarity and differences between finite sample sets. In this application, it is used to calculate the similarity between features.

[0034] Traditionally, recognizing user voice intent has primarily involved converting speech to text and then performing intent recognition on the text. This application optimizes the process and steps for recognizing voice intent, eliminating the need for speech-to-text conversion and subsequent intent recognition. Instead, it utilizes a pre-defined recognition model to identify user intent from audio data.

[0035] As an optional embodiment, the audio data is input into a preset recognition model for intent recognition, and the recognition result is obtained by: extracting speech features of each frame in the audio data according to preset dimensions to obtain a feature matrix; determining the matching probability of each frame speech feature in the feature matrix with each preset feature in the preset feature library to obtain a probability matrix, wherein the preset feature library is obtained during the training of the preset recognition model; converting the probability matrix into target machine code, wherein the target machine code includes the machine code of each frame speech feature; decoding the target machine code to obtain the target intent code, and determining the target intent code as the recognition result.

[0036] Specifically, the first training set is a definite intent training set, and the second training set is a fuzzy intent training set. An initial machine learning model is pre-trained using both the definite and fuzzy intent training sets to obtain a preset recognition model. This process includes: first, training the initial machine learning model using the definite intent training set, enabling it to convert the corresponding intent information into machine code (e.g., 101001 or 102001, where the first three digits are the device identification code and the last three digits are the intent identification code). This yields an intermediate model with initial intent recognition capabilities. Then, the fuzzy intent training set is used to reinforce the intermediate model. By repeatedly verifying the training results, the machine learning model is repeatedly strengthened to improve its recognition accuracy, ultimately obtaining the preset recognition model.

[0037] Specifically, the preset dimensions can be set according to requirements (e.g., 12 dimensions).

[0038] As an optional embodiment, extracting speech features from each frame of audio data according to a preset dimension to obtain a feature matrix includes: filtering out impurity data in the audio data to obtain first audio data; dividing the first audio data into frames to obtain multiple frames of second audio data; extracting features from the multiple frames of second audio data using a preset dimension, and integrating the extracted speech features of each frame to obtain a feature matrix.

[0039] Optionally, the MFCC (Mel-scale Frequency Cepstral Coefficients) features of the audio data are extracted as follows: silence and noise data are filtered out using VAD endpoint detection; new audio data is pre-emphasized, framed, and windowed to enhance speech signal performance; for each short-time analysis window, the corresponding spectrum is obtained through Fourier transform to obtain the spectrum distributed across different time windows on the time axis; the above spectrum is passed through a Mel filter bank to obtain the Mel spectrum; the linear natural spectrum is converted into a Mel spectrum that reflects human auditory characteristics using the Mel spectrum; cepstral analysis is performed on the Mel spectrum (including taking the logarithm, performing an inverse transform, which is generally implemented using DCT discrete cosine transform, and taking the 2nd to 13th coefficients after DCT as MFCC coefficients) to obtain the Mel frequency cepstral coefficients (MFCC), which are the features of this frame of speech; the MFCC features of all frames are then obtained.

[0040] The MFCC features of all frames are integrated to obtain the feature matrix.

[0041] For example, the audio is first filtered and transcoded by detecting VAD endpoints and extracting MFCC features. At this point, the sound becomes an M-row (M is the acoustic feature) N-column (N is the total number of frames) matrix.

[0042] As an optional embodiment, the matching probability of each frame speech feature in the feature matrix with each preset feature in the preset feature library is determined respectively. The probability matrix is ​​obtained by: determining multiple Jaccard coefficients of the speech feature with each preset feature in the preset feature library, and determining the multiple Jaccard coefficients as the matching probability of the speech feature with each preset feature; and integrating the multiple matching probabilities to obtain the probability matrix.

[0043] The speech features in the feature matrix are matched with each preset feature in the preset feature library to obtain the total matching probability. The matching method is to calculate the Jaccard coefficient between the speech features and the preset features. For example, matching the speech features of the first frame with 30 preset features in the preset feature library includes: calculating the Jaccard coefficient between the speech features of the first frame and the first preset feature to obtain P1, and determining P1 as the matching probability between the speech features of the first frame and the first preset feature. The Jaccard coefficients of the speech features of the first frame and other preset features are then calculated to obtain the probability (P1, P2...P30) between the speech features of the first frame and all preset features.

[0044] As an optional embodiment, converting the probability matrix into target machine code includes: extracting multiple matching probabilities corresponding to single-frame speech features in the probability matrix, and determining the target preset feature corresponding to the matching probability with the largest value, wherein the multiple preset features include the target preset feature; determining the preset machine code carried by the target preset feature as the single-frame machine code of the single-frame speech features, wherein the preset machine code is obtained by encoding a preset intent code; and concatenating the multiple single-frame machine codes in chronological order to obtain the target machine code.

[0045] Optionally, if there is more than one maximum value in the matching probability of the speech features in the same frame, the first maximum value is taken as the highest matching probability.

[0046] Specifically, the state machine code corresponding to the preset feature with the highest matching probability with the speech feature is the machine code of that speech feature. Therefore, after obtaining the probability matrix, the machine code of the speech feature in each frame can be determined by the highest matching probability corresponding to the speech feature in each frame.

[0047] Multiple single-frame machine codes are concatenated in chronological order to ensure the continuity and accuracy of speech features.

[0048] As an optional embodiment, determining the intent of audio data using the recognition result includes: obtaining an intent association table, wherein the intent association table is obtained during the training of a preset recognition model, and the intent association table is used to represent the correspondence between intent codes and intents; obtaining the target intent code in the recognition result, and determining the intent data corresponding to the target intent code in the intent association table based on the correspondence; and determining the intent data as the intent of the audio data.

[0049] Previously, converting intent information into machine code during the training of the preset recognition model was actually an encoding process, and an intent association table was generated during the encoding process. Therefore, in the process of audio intent recognition, after obtaining the target machine code, the corresponding target intent code can be determined according to the intent association table.

[0050] Optionally, this application constructs a state network using a Hidden Markov Model and then obtains the corresponding intent code from the state network.

[0051] As an optional embodiment, after determining the intent data corresponding to the target intent code, the method further includes: acquiring device data and operation data from the intent data; and sending the operation data to the target device corresponding to the device data, so that the target device performs the target operation corresponding to the operation data.

[0052] The target intent code includes an encoding representing the intent and an encoding representing the device. After obtaining the device data and operation data from the intent data, it is possible to determine which device the target object wants to perform which operation. Therefore, the operation data can be sent to the corresponding device.

[0053] For example, after the audio data of the phrase "Please turn on the air conditioner" is input into the preset recognition model, the preset recognition model first performs initial filtering and transcoding, then converts it into an M-row N-column feature matrix. Then, by calculating the Jaccard coefficients of the speech features in the feature matrix and the preset features in the preset database, a probability matrix (P01, P02, ...), (P10, P12, ...), ...) is obtained. The machine code for each frame can be obtained from the probability matrix, such as (0, 0, 1, 0, 0, ...). The intent code is obtained by decoding the machine code, such as 101001 (101 represents the "air conditioner" device, and 001 represents the "turn on" intent). Finally, the 101001 instruction is directly sent to the corresponding device.

[0054] Figure 2 A flowchart of an intent recognition method provided in this application includes:

[0055] Step 201, Voice input (same as "inputting audio data into the preset recognition model" in the above steps);

[0056] Step 203, speech processing, including VAD endpoint detection, speech segmentation, noise reduction filtering, framing and MFCC feature extraction (the same as "extracting speech features of each frame in the audio data according to preset dimensions to obtain the feature matrix" in the above steps);

[0057] Step 205: Calculate the Jaccard coefficient, which includes training the model in advance using the training set, generating a voiceprint matrix corresponding to the intent in the model, and directly using the Jaccard coefficient to calculate the similarity between the two voiceprint matrices when detecting user intent (same as "determining the matching probability of each frame speech feature in the feature matrix and each preset feature in the preset feature library" in the above steps).

[0058] Step 207: Obtain the probability matrix. Examples of the matrix include: ((P01, P02, P03, ...), (P11, P12, P13, ...)...);

[0059] Step 209: Obtain the machine code through the probability matrix (same as "convert the probability matrix into the target machine code" in the above steps). For example, (P01, P02, P03, ...) corresponds to 0, and (P11, P12, P13, ...) corresponds to 1.

[0060] Step 211: Decode the machine code to obtain the intent code (same as "decode the target machine code to obtain the target intent code" in the above steps). For example, the machine code (0, 1, 0, 0, ...) is decoded to correspond to the intent code 101001.

[0061] Step 213: Issue the instruction (same as "send the operation data to the target device corresponding to the device data" in the above steps).

[0062] Figure 3 This is a schematic diagram of intent recognition provided in this application, wherein s1029, s124,... represent the initial intent; and ay,... represent the final intent.

[0063] This application provides an intent recognition method, comprising: acquiring audio data of a target object; inputting the audio data into a preset recognition model for intent recognition to obtain a recognition result, wherein the preset recognition model is pre-trained using a first training set and a second training set, and the training data used in the first and second training sets are different; and determining the intent of the audio data using the recognition result. By directly inputting the user's voice data into the preset recognition model for intent recognition to determine the user's intent, this method solves the problem that the voice recognition process must first be converted into text, which leads to a cumbersome recognition process and affects recognition accuracy.

[0064] According to another aspect of the embodiments of this application, this application provides an intent recognition device, such as... Figure 4 As shown, it includes:

[0065] Audio acquisition module 402 is used to acquire audio data of the target object;

[0066] The audio recognition module 404 is used to input audio data into a preset recognition model for intent recognition and obtain recognition results. The preset recognition model is obtained by training a first training set and a second training set in advance. The training data used in the first training set and the second training set are different.

[0067] The intent determination module 406 is used to determine the intent of the audio data using the recognition results.

[0068] It should be noted that the audio acquisition module 402 in this embodiment can be used to execute step 101 in this application embodiment, the audio recognition module 404 in this embodiment can be used to execute step 103 in this application embodiment, and the intent determination module 406 in this embodiment can be used to execute step 105 in this application embodiment.

[0069] Optionally, the audio recognition module 404 includes:

[0070] The feature acquisition submodule is used to extract speech features from each frame of the audio data according to preset dimensions to obtain a feature matrix;

[0071] The probability determination submodule is used to determine the matching probability of each frame speech feature in the feature matrix with each preset feature in the preset feature library to obtain the probability matrix. The preset feature library is obtained during the training of the preset recognition model.

[0072] The probability transformation submodule converts the probability matrix into target machine code, where the target machine code includes the machine code of the speech features of each frame.

[0073] The machine code decoding submodule is used to decode the target machine code, obtain the target intent code, and determine the target intent code as the recognition result.

[0074] Optionally, the feature acquisition submodule is also used to filter out impurity data in the audio data to obtain first audio data; to divide the first audio data into frames to obtain multiple frames of second audio data; to extract features from the multiple frames of second audio data using a preset dimension, and to integrate the extracted speech features of each frame to obtain a feature matrix.

[0075] Optionally, the probability determination submodule is also used to determine multiple Jaccard coefficients between the speech features and each preset feature in the preset feature library, and to determine the multiple Jaccard coefficients as the matching probability between the speech features and each preset feature; and to integrate the multiple matching probabilities to obtain a probability matrix.

[0076] Optionally, the probability transformation submodule is also used to extract multiple matching probabilities corresponding to single-frame speech features in the probability matrix, and determine the target preset feature corresponding to the matching probability with the largest value, wherein the multiple preset features include the target preset feature; determine the preset machine code carried by the target preset feature as the single-frame machine code of the single-frame speech feature, wherein the preset machine code is obtained by encoding the preset intent code; and concatenate the multiple single-frame machine codes in chronological order to obtain the target machine code.

[0077] Optionally, the intent determination module 406 is further configured to obtain an intent association table, wherein the intent association table is obtained during the training of the preset recognition model, and the intent association table is used to represent the correspondence between intent codes and intents; obtain the target intent code in the recognition result, and determine the intent data corresponding to the target intent code in the intent association table based on the correspondence; and determine the intent data as the intent of the audio data.

[0078] Optionally, the device further includes a sending module, which, after determining the intent data corresponding to the target intent code, acquires device data and operation data from the intent data; and sends the operation data to the target device corresponding to the device data, so that the target device performs the target operation corresponding to the operation data.

[0079] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments.

[0080] According to another aspect of the embodiments of this application, this application provides an electronic device, such as... Figure 5 As shown, the system includes a memory 501, a processor 503, a communication interface 505, and a communication bus 507. The memory 501 stores a computer program that can run on the processor 503. The memory 501 and the processor 503 communicate through the communication interface 505 and the communication bus 507. When the processor 503 executes the computer program, it implements the steps of the above method.

[0081] The memory and processor in the aforementioned electronic devices communicate with each other via a communication bus and a communication interface. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0082] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0083] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0084] According to another aspect of the embodiments of this application, a computer-readable medium having processor-executable non-volatile program code is also provided.

[0085] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0086] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.

[0087] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0088] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0089] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0091] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0094] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0095] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An intent recognition method, characterized in that, include: Obtain the audio data of the target object; Speech features of each frame in the audio data are extracted according to preset dimensions to obtain a feature matrix; Determine multiple Jaccard coefficients between the speech feature and each preset feature in the preset feature library, and determine the multiple Jaccard coefficients as the matching probability between the speech feature and each preset feature; The multiple matching probabilities are integrated to obtain a probability matrix, wherein the preset feature library is obtained during the training of the preset recognition model; The probability matrix is ​​converted into target machine code, wherein the target machine code includes the machine code of the speech features in each frame; The target machine code is decoded to obtain the target intent code, and the target intent code is determined as the recognition result. The preset recognition model is obtained by pre-training using a first training set and a second training set. The first training set is a definite intent training set, and the second training set is a fuzzy intent training set. Pre-training the preset recognition model using the definite intent training set and the fuzzy intent training set specifically includes: training an initial machine model using the definite intent training set to obtain an intermediate model, enabling the intermediate model to convert the corresponding intent information into machine code; using the fuzzy intent training set to reinforce the intermediate model, repeatedly reinforcing the intermediate model by verifying the training results multiple times, and finally obtaining the preset recognition model. The intent of the audio data is determined using the recognition results.

2. The method according to claim 1, characterized in that, The step of extracting speech features from each frame of the audio data according to a preset dimension to obtain a feature matrix includes: Filter out impurity data from the audio data to obtain the first audio data; The first audio data is divided into frames to obtain multiple frames of second audio data; The preset dimensions are used to extract speech features from multiple frames of the second audio data, and the extracted speech features from each frame are integrated to obtain the feature matrix.

3. The method according to claim 1, characterized in that, The step of converting the probability matrix into target machine code includes: Extract multiple matching probabilities corresponding to the speech features of a single frame in the probability matrix, and determine the target preset feature corresponding to the matching probability with the largest value, wherein the multiple preset features include the target preset feature; The preset machine code carried by the target preset feature is determined as the single-frame machine code of the speech feature in a single frame, wherein the preset machine code is obtained by encoding the preset intent code; The target machine code is obtained by concatenating multiple single-frame machine codes in chronological order.

4. The method according to claim 1, characterized in that, The intent to determine the audio data using the recognition result includes: Obtain an intent association table, wherein the intent association table is obtained during the training of the preset recognition model, and the intent association table is used to represent the correspondence between intent codes and intents; Obtain the target intent code from the recognition result, and determine the intent data corresponding to the target intent code in the intent association table based on the correspondence; The intent data is determined as the intent of the audio data.

5. The method according to claim 4, characterized in that, After determining the intent data corresponding to the target intent code, the method further includes: Obtain the device data and operation data from the intent data; The operation data is sent to the target device corresponding to the device data, so that the target device performs the target operation corresponding to the operation data.

6. An intent recognition device, characterized in that, include: The audio acquisition module is used to acquire the audio data of the target object; An audio recognition module is used to extract speech features from each frame of the audio data according to a preset dimension to obtain a feature matrix; The process involves determining multiple Jaccard coefficients between the speech features and various preset features in a preset feature library, and defining these multiple Jaccard coefficients as the matching probabilities between the speech features and each preset feature; integrating these matching probabilities to obtain a probability matrix, wherein the preset feature library is obtained during the training of a preset recognition model; converting the probability matrix into target machine code, wherein the target machine code includes the machine code of each frame of the speech features; decoding the target machine code to obtain a target intent code, and defining the target intent code as the recognition result, wherein the preset recognition model is pre-trained using a first training set and a second training set, the first training set being an explicit intent training set and the second training set being an ambiguous intent training set; specifically, pre-training the preset recognition model using the explicit intent training set and the ambiguous intent training set includes: training an initial machine model using the explicit intent training set to obtain an intermediate model, enabling the intermediate model to convert the corresponding intent information into machine code; using the ambiguous intent training set to reinforce the intermediate model, repeatedly reinforcing the intermediate model by verifying the training results multiple times, and finally obtaining the preset recognition model; An intent determination module is used to determine the intent of the audio data using the recognition results.

7. An electronic device comprising a memory, a processor, a communication interface, and a communication bus, wherein the memory stores a computer program executable on the processor, and the memory and the processor communicate via the communication bus and the communication interface, characterized in that... When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable medium having processor-executable non-volatile program code, characterized in that, The program code causes the processor to execute the method of any one of claims 1 to 5.

Images