Audio recognition method, device, equipment, and storage medium based on machine learning

Abstract

The present application relates to artificial intelligence, and provides an audio recognition method and apparatus, device, and storage medium based on machine learning, the method comprising: obtaining a semantically-free first audio sample emitted by a living being, inputting the first audio sample into an intention recognition model to predict the intention of the first audio sample and obtain first intention data; and recognizing the timbre of the first audio sample to obtain first timbre data; determining a first target audio based on the first intention data and the first timbre data; training the intention recognition model based on the comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, and recognizing the intention of the audio to be recognized based on the trained intention recognition model. The technical solution of the embodiment of the present application can reduce the influence of parameters such as timbre on intention recognition and improve the accuracy of intention recognition.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically, to an audio recognition method and apparatus, device, and storage medium based on machine learning. Background Art

[0002] In real life, we hear non-semantic sounds made by living things, and ordinary people cannot understand their meaning or recognize their intentions. For example, when infants and young children have something to ask for, they will make non-semantic "ahhh", "oh oh oh" or crying sounds, but ordinary people cannot understand the intention of their sounds and thus cannot meet their demands; for another example, when animals have something to ask for, they will also make sounds, but ordinary people cannot understand the intention of their sounds.

[0003] With the development of artificial intelligence and speech technology, it has become possible to identify the intention of non-semantic sounds. However, the accuracy of the existing methods for identifying the intention of non-semantic sounds is low and it is impossible to accurately identify the intention of non-semantic sounds. Summary of the invention

[0004] To solve the above technical problems, the embodiments of the present application provide an audio recognition method and apparatus, device, and storage medium based on machine learning.

[0005] According to one aspect of an embodiment of the present application, a method for audio recognition based on machine learning is provided, the method comprising:

[0006] Obtaining a first non-semantic audio sample emitted by the creature;

[0007] Inputting the first audio sample into an intention recognition model to predict the intention of the first audio sample to obtain first intention data; and recognizing the timbre of the first audio sample to obtain first timbre data; wherein the intention recognition model is a model established based on machine learning;

[0008] Determine a first target audio based on the first intention data and the first timbre data;

[0009] The intention recognition model is trained based on the comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, and the intention of the audio to be recognized is recognized based on the trained intention recognition model.

[0010] Furthermore, the step of inputting the first audio sample into an intention recognition model to predict the intention of the first audio sample to obtain first intention data includes:

[0011] Inputting the first audio sample into an intention recognition model so that the intention recognition model predicts the intention of the first audio sample;

[0012] During the prediction process of the intention recognition model, obtaining data output by a hidden layer of the intention recognition model;

[0013] The acquired data is used as the first intention data.

[0014] Further, determining the first target audio based on the first intention data and the first timbre data includes:

[0015] Acquire second intention data and second timbre data;

[0016] generating a first audio based on the first intention data and the second timbre data, and generating a second audio based on the second intention data and the first timbre data;

[0017] Inputting the first audio into the intention recognition model to predict the intention of the first audio to obtain third intention data; and recognizing the timbre of the second audio to obtain third timbre data;

[0018] The first target audio is generated based on the third intention data and the third timbre data.

[0019] Further, the obtaining of the second intention data and the second timbre data includes:

[0020] Obtaining a second non-semantic audio sample emitted by the creature;

[0021] Inputting the second audio sample into the intention recognition model to predict the intention of the second audio sample to obtain second intention data; and recognizing the timbre of the second audio sample to obtain second timbre data;

[0022] After generating the first audio based on the first intent data and the second timbre data, and generating the second audio based on the second intent data and the first timbre data, and before obtaining the trained intent recognition model, the method further includes:

[0023] Inputting the second audio into the intention recognition model to predict the intention of the second audio to obtain fourth intention data; and recognizing the timbre of the first audio to obtain fourth timbre data;

[0024] generating a second target audio based on the fourth intention data and the fourth timbre data;

[0025] The intention recognition model is trained based on a comparison result of the second target audio and the second audio sample.

[0026] Furthermore, the first audio is generated by an audio generation model based on the first intention data and the second timbre data; the first timbre data is obtained by recognizing a first audio sample by a timbre recognition model; the audio generation model and the timbre recognition model are models established based on machine learning;

[0027] After generating the first target audio based on the third intention data and the third timbre data, the method further includes:

[0028] The audio generation model and the timbre recognition model are trained based on the comparison result of the first target audio and the first audio sample.

[0029] Furthermore, before obtaining the trained intent recognition model, the method further includes:

[0030] Acquire a third audio sample without semantics emitted by a living being and a preset intention of the third audio sample;

[0031] Inputting the third audio sample into the intention recognition model to predict the intention of the third audio sample and obtain intention data corresponding to the third audio sample;

[0032] The intention recognition model is trained based on the intention data corresponding to the third audio sample and the preset intention.

[0033] Further, after the intention recognition model is trained based on the comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, the method further includes:

[0034] Get the audio to be recognized;

[0035] Inputting the audio to be recognized into the trained intention recognition model to recognize the intention of the audio to be recognized, thereby obtaining the intention of the audio to be recognized;

[0036] Output the intention of the audio to be recognized.

[0037] According to one aspect of an embodiment of the present application, there is provided an audio recognition device based on machine learning, the device comprising:

[0038] An acquisition module configured to acquire a first non-semantic audio sample emitted by a living being;

[0039] a recognition module configured to input the first audio sample into an intention recognition model to predict the intention of the first audio sample to obtain first intention data; and recognize the timbre of the first audio sample to obtain first timbre data; wherein the intention recognition model is a model established based on machine learning;

[0040] a determination module, configured to determine a first target audio based on the first intention data and the first timbre data;

[0041] The training module is configured to train the intention recognition model based on the comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, and recognize the intention of the audio to be recognized based on the trained intention recognition model.

[0042] According to one aspect of an embodiment of the present application, there is provided an electronic device, including:

[0043] one or more processors;

[0044] The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the method as described above.

[0045] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a computer, the computer executes the method as described above.

[0046] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the method as described above.

[0047] In the technical solution provided in the embodiments of the present application, a first semantic-free audio sample emitted by a biological being is obtained, and the first audio sample is input into an intention recognition model to predict the intention of the first audio sample to obtain first intention data; and the timbre of the first audio sample is recognized to obtain first timbre data; based on the first intention data and the first timbre data, a first target audio is determined; based on the comparison result of the first target audio and the first audio sample, the intention recognition model is trained to obtain a trained intention recognition model, and based on the trained intention recognition model, the intention of the audio to be recognized is recognized, thereby reducing the influence of parameters such as timbre on intention recognition and improving the accuracy of intention recognition.

[0048] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and together with the specification, are used to explain the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0050] Figure 1 is a flowchart of an audio recognition method based on machine learning shown in an exemplary embodiment of the present application;

[0051] Figure 2 It is a structural diagram of the intent recognition model established using the BiRNN mechanism;

[0052] Figure 3 yes Figure 1 A flowchart of step S130 in the illustrated embodiment in an exemplary embodiment;

[0053] Figure 4 is a flowchart of an audio recognition method based on machine learning shown in another exemplary embodiment of the present application;

[0054] Figure 5 is a flowchart of an audio recognition method based on machine learning shown in another exemplary embodiment of the present application;

[0055] Figure 6 is a process diagram of obtaining a first target audio and a second target audio, shown in an exemplary embodiment of the present application;

[0056] Figure 7 is a block diagram of an audio recognition device based on machine learning shown in an exemplary embodiment of the present application;

[0057] Figure 8 It is a schematic diagram of the structure of an electronic device shown in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0058] Here, exemplary embodiments will be described in detail, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0059] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0060] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0061] It should also be noted that the "multiple" mentioned in this application refers to two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship.

[0062] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0063] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0064] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. Deep Learning (DL) is a new research direction in the field of machine learning. The concept of deep learning originates from the study of artificial neural networks. Multilayer perceptrons with multiple hidden layers are a type of deep learning structure. Deep learning combines low-level features to form more abstract high-level representations of attribute categories or features to discover distributed feature representations of data. The motivation for studying deep learning is to establish a neural network that simulates the human brain for analysis and learning. It imitates the mechanism of the human brain to interpret data.

[0065] The solution provided in the embodiments of the present application involves machine learning technology of artificial intelligence, which is specifically illustrated by the following embodiments.

[0066] See also Figure 1 , Figure 1 1 is a flow chart of an audio recognition method based on machine learning according to an exemplary embodiment. The method may include steps S110 to S140, which are described in detail as follows:

[0067] Step S110, obtaining a first non-semantic audio sample emitted by a living being.

[0068] The first audio sample is a non-semantic audio emitted by a living being, which is a sample for training the intent recognition model. For the relevant content of the intent recognition model, please refer to the subsequent records.

[0069] The non-semantic audios emitted by living things include but are not limited to the calls of animals, the voices of infants and young children, the cries of adults, etc. For example, in order to understand the intention of the sound emitted by infants and young children, the voices of infants and young children can be collected in advance as the first audio samples; in order to understand the intention of the sound emitted by animals, the sounds of animals can be collected in advance as the first audio samples.

[0070] It should be noted that the number and type of the first audio samples can be flexibly set according to actual needs.

[0071] Step S120: input the first audio sample into the intention recognition model to predict the intention of the first audio sample to obtain first intention data; and recognize the timbre of the first audio sample to obtain first timbre data.

[0072] The intent recognition model is a model based on machine learning, which is used to recognize the intent of non-semantic audio emitted by organisms. In order to improve the recognition accuracy, the intent recognition model can be established based on deep learning. In order to further improve the recognition accuracy, the BiRNN (Bidirectional Recurrent Neural Networks) + Attention mechanism can be used to establish the intent recognition model. Among them, the Attention mechanism is a solution to the problem proposed by imitating human attention. Simply put, it is to quickly filter out high-value information from a large amount of information. Of course, other neural networks or other machine learning algorithms can also be used to establish the intent recognition model, for example, a convolutional neural network can be used to establish the intent recognition model, or a recurrent neural network can be used to establish the intent recognition model.

[0073] Intent is the intention to achieve a certain purpose. Intent can be set according to actual needs, including but not limited to "want to eat", "want to sleep", "need comfort", etc. Intent data is data that represents intent.

[0074] In this embodiment, the first audio sample is input into the intention recognition model so that the intention recognition model predicts the intention of the first audio sample, and the data output by the intention recognition model for the first audio sample is used as the first intention data corresponding to the first audio sample.

[0075] In one example, the implicit intent of the first audio sample may be used as the first intent data, that is, the data output by the hidden layer of the intent recognition model for the first audio sample may be used as the first intent data. Under this condition, step S120 may include steps S121-S123, which are described in detail as follows:

[0076] Step S121: input the first audio sample into the intention recognition model, so that the intention recognition model predicts the intention of the first audio sample.

[0077] The first audio sample is input into the intention recognition model, and the intention recognition model processes the first audio sample to predict the intention of the first audio sample.

[0078] Step S122, during the prediction process of the intent recognition model, obtain the data output by the hidden layer of the intent recognition model.

[0079] It should be noted that the machine learning model includes an input layer, a hidden layer and an output layer (i.e., a fully connected layer), wherein the input layer is used to receive data input into the machine learning model, the output of the input layer is connected to the input of the hidden layer, the input of the hidden layer is connected to the output layer, and the output layer is used to output the recognition result of the machine learning model. In order to facilitate the subsequent generation of audio based on intent data, in the process of the intent recognition model predicting the intent of the first audio sample, the data output by the hidden layer of the intent recognition model can be obtained as the first intent data. For example, see Figure 2 As shown, for the intention recognition model established by BiRNN, it includes an input layer, a hidden layer and an output layer, wherein the input layer receives data X1-X T , the hidden layer includes two layers of loops with opposite directions, respectively and The output of the recurrent layer (i.e. a t，1 -a t，T ) is input to the output layer (i.e. S t-1 and S t The output layer is Y t-1 and Y t , then a t，1 -a t，TAs the first intention data.

[0080] Step S123: Use the acquired data as first intention data.

[0081] After obtaining the data output by the hidden layer of the intent recognition model for the first audio sample, the obtained data is used as the first intent data. Since the data output by the hidden layer of the intent recognition model can characterize the intent of the audio to a certain extent and the data volume is large, using the data output by the hidden layer of the intent recognition model as the first intent data reduces the difficulty of generating audio based on the first intent data.

[0082] In another example, the explicit intent of the first audio sample may also be used as the first intent data, that is, the data output by the output layer of the intent recognition model for the first audio sample may be used as the first intent data. Figure 2 As shown, Y t-1 and Y t As the first intention data.

[0083] In this embodiment, the timbre of the first audio sample needs to be identified to obtain first timbre data corresponding to the first audio sample. The specific method of timbre recognition can be flexibly set according to actual needs. For example, in one example, the timbre of the first audio sample can be identified based on a timbre recognition model to obtain first timbre data, wherein the timbre recognition model is a model established based on machine learning, and is used to identify the timbre of the audio, wherein in order to improve the recognition accuracy, a timbre recognition model can be established based on deep learning. The timbre recognition model can be established using the BiRNN+Attention mechanism, or it can also be established using other neural networks or other machine learning algorithms.

[0084] Step S130: determining a first target audio based on the first intention data and the first timbre data.

[0085] Among them, the method of determining the first target audio based on the first intention data and the first timbre data can be flexibly set according to actual needs. For example, audio can be directly generated based on the first intention data and the first timbre data, and the generated audio can be used as the first target audio.

[0086] Step S140: training the intention recognition model based on the comparison result between the first target audio and the first audio sample to obtain a trained intention recognition model, and recognizing the intention of the audio to be recognized based on the trained intention recognition model.

[0087] It should be noted that the trained intent recognition model is a model obtained after training the intent recognition model and meets the preset conditions. The preset conditions can be flexibly set according to actual needs, for example, the accuracy can be set to reach a preset accuracy threshold (such as 90%, 70%, etc.).

[0088] In order to obtain a trained intention recognition model, in this embodiment, the first target audio and the first audio sample are compared to train the intention recognition model based on the comparison result of the first target audio and the first audio sample. Subsequently, the intention of the audio to be recognized can be recognized based on the trained intention recognition model. Since the first target audio is determined based on the first predicted intention and the first timbre data corresponding to the first audio sample, if the first target audio and the first audio sample match, it indicates to a great extent that the intention recognition model recognizes the intention of the first audio sample accurately. If the first target audio and the first audio sample do not match, it indicates to a great extent that the intention recognition model recognizes the intention of the first audio sample incorrectly, and the parameters of the intention recognition model need to be adjusted. The first target audio and the first audio sample match can be that the similarity between the first target audio and the first audio sample is greater than the preset similarity threshold. The preset similarity threshold can be flexibly set according to actual needs, for example, set to 90%, 80%, etc.

[0089] Among them, the first target audio and the first audio sample can be compared first, and the comparison result can be input into the intention recognition model so that the intention recognition model adjusts internal parameters based on the comparison result; alternatively, the first target audio can be input into the intention recognition model, and the intention recognition model can compare the first target audio and the first audio sample to adjust internal parameters based on the comparison result.

[0090] In this embodiment, a first non-semantic audio sample emitted by a living being is obtained, and the first audio sample is input into an intention recognition model to predict the intention of the first audio sample to obtain first intention data; and the timbre of the first audio sample is recognized to obtain first timbre data; based on the first intention data and the first timbre data, a first target audio is determined; based on the comparison result of the first target audio and the first audio sample, the intention recognition model is trained to obtain a trained intention recognition model, and based on the trained intention recognition model, the intention of the audio to be recognized is recognized, thereby identifying the intention of the non-semantic audio emitted by the living being, and the intention recognition model established based on machine learning is used to perform intention recognition, which can improve the recognition accuracy. Furthermore, in the process of training the intention recognition model, the influence of parameters such as timbre on intention recognition is reduced, thereby further improving the accuracy of intention recognition.

[0091] See also Figure 3 , Figure 3 for Figure 1 The flowchart of step S130 in the illustrated embodiment is in an exemplary embodiment. In order to further reduce the interference of factors such as timbre on the intention recognition model, as Figure 3 As shown, the process of determining the first target audio based on the first intention data and the first timbre data may include steps S131-S134, which are described in detail as follows:

[0092] Step S131, obtaining second intention data and second timbre data.

[0093] The second intent data and the second timbre data may be features corresponding to the same audio, or features corresponding to different audios. The second intent data and the second timbre data may be pre-set, or may be obtained based on audio samples during the training of the intent recognition model.

[0094] Under the condition that the second intent data and the second timbre data belong to features corresponding to the same audio, and the second intent data and the second timbre data are obtained based on audio samples during the training process of the intent recognition model, step S131 may include steps S210 to S220, which are described in detail as follows:

[0095] Step S210, obtaining a second non-semantic audio sample emitted by the organism.

[0096] The second audio sample is a semantic-free audio emitted by a living being, which is a sample for training the intention recognition model. The specific interpretation of the second audio sample can be found in the first audio sample, which will not be repeated here.

[0097] The second audio sample may be a sample located before the first audio sample among the samples input to the intent recognition model; or, it may be a sample located after the first audio sample among the samples input to the intent recognition model; of course, it may also be a sample of other input intent recognition models.

[0098] Step S220: input the second audio sample into the intention recognition model to predict the intention of the second audio sample to obtain second intention data; and recognize the timbre of the second audio sample to obtain second timbre data.

[0099] In this embodiment, the second audio sample is input into the intention recognition model, so that the intention recognition model predicts the intention of the second audio sample, and the data output by the intention recognition model for the second audio sample is used as the second intention data corresponding to the second audio sample. The specific method is similar to the aforementioned step S120 and will not be repeated here.

[0100] In step S220, the timbre of the second audio sample needs to be identified to obtain second timbre data, wherein the specific process can refer to the aforementioned identification process of the first timbre data, which will not be repeated here.

[0101] Step S132: Generate a first audio based on the first intention data and the second timbre data, and generate a second audio based on the second intention data and the first timbre data.

[0102] The specific manner of generating the first audio based on the first intention data and the second timbre data and generating the second audio based on the second intention data and the first timbre data can be flexibly set according to actual needs.

[0103] For example, in one example, the audio generation model can generate the first audio based on the first intent data and the second timbre data, and the audio generation model can generate the second audio based on the second intent data and the first timbre data. Among them, the audio generation model is a model established based on machine learning, which is used to generate audio based on intent data and timbre data. For example, it can be a model established based on deep learning. The audio generation model can be established using one of the models such as CNN (Convolutional Neural Networks), RNN (Recurrent Neural Networks), BERT (Bidirectional Encoder Representation from Transformers), etc., wherein BERT is a pre-trained language representation model. When the audio generation model uses CNN, its internal convolution method can select hollow convolution, wherein the hollow convolution can increase the perception field of view and complete the convolution with smaller parameters; of course, other convolution methods can also be used.

[0104] Step S133, inputting the first audio into the intention recognition model to predict the intention of the first audio to obtain third intention data; and recognizing the timbre of the second audio to obtain third timbre data.

[0105] Among them, the specific implementation method of step S133 is similar to the specific implementation method of the aforementioned step S120, and will not be repeated here.

[0106] Step S134: Generate a first target audio based on the third intention data and the third timbre data.

[0107] Among them, the specific method of generating the first target audio can be flexibly set according to actual needs. For example, the third intention data and the third timbre data are input into the audio generation model so that the audio generation model generates the first target audio based on the third intention data and the third timbre data.

[0108] In this embodiment, second intent data and second timbre data are obtained, a first audio is generated based on the first intent data and the second timbre data, and a second audio is generated based on the second intent data and the first timbre data; the first audio is input into an intention recognition model to predict the intent of the first audio to obtain third intent data; and the timbre of the second audio is recognized to obtain third timbre data; a first target audio is generated based on the third intent data and the third timbre data, so that the first target audio is a reconstructed audio, and in the process of generating the first target audio, the timbre and the intent are exchanged, so that the intention recognition model trained based on the first target audio can reduce the influence of timbre, volume, etc. on intent recognition during the intention recognition process, thereby improving recognition accuracy.

[0109] In an exemplary embodiment, under the condition that step S131 includes steps S210 to S220, Figure 3 After step S132 shown, and before obtaining the trained intent recognition model, the method may further include steps S310-S330, which are described in detail as follows:

[0110] Step S310: input the second audio into the intention recognition model to predict the intention of the second audio to obtain fourth intention data; and recognize the timbre of the first audio to obtain fourth timbre data.

[0111] Among them, the specific implementation method of step S310 is similar to the specific implementation method of the aforementioned step S133, which will not be repeated here.

[0112] It should be noted that the second intent data, the third intent data, and the fourth intent data can be explicit intents or implicit intents. In order to maintain consistency, the first intent data, the second intent data, the third intent data, and the fourth intent data can all be explicit intents, or can all be implicit intents.

[0113] Step S320: Generate a second target audio based on the fourth intention data and the fourth timbre data.

[0114] Among them, the specific implementation method of step S320 is similar to the specific implementation method of the aforementioned step S134, which will not be repeated here.

[0115] That is to say, the intention data and timbre data of the first audio sample and the intention data and timbre data of the second audio sample are crossed to generate the first audio and the second audio, and the intention data and timbre data of the first audio and the intention data and timbre data of the second audio are crossed to generate the first target audio corresponding to the first audio sample and the second target audio corresponding to the second audio sample, so that the interference of timbre on intention recognition can be reduced.

[0116] It should be noted that, in other embodiments, the second target audio may also be determined based on the method for determining the first target audio.

[0117] Step S330: training the intention recognition model based on the comparison result of the second target audio and the second audio sample.

[0118] It should be noted that in order to obtain a trained intent recognition model, in this embodiment, in addition to training the intent recognition model based on the comparison results of the first target audio and the first audio sample, the intent recognition model can also be trained based on the comparison results of the second target audio and the second audio sample, thereby improving the training speed.

[0119] The specific implementation of step S330 is similar to the specific implementation of the aforementioned step S140, and will not be repeated here.

[0120] In an exemplary embodiment, under the condition that the first audio is generated by the audio generation model based on the first intention data and the second timbre data, and the first timbre data is obtained by recognizing the first audio sample by the timbre recognition model, Figure 3 After step S134 shown, the method further includes: training the audio generation model and the timbre recognition model based on the comparison result of the first target audio and the first audio sample.

[0121] In other examples, the audio generation model and the timbre recognition model may also be trained based on the comparison result between the second target audio and the second audio sample.

[0122] In an exemplary embodiment, see Figure 4 As shown, before obtaining the trained intent recognition model, the method further includes:

[0123] Step S410, obtaining a third audio sample without semantics emitted by a living being and a preset intention of the third audio sample.

[0124] In this embodiment, for the definition of the third audio sample, reference may be made to the related description of the first audio sample. A corresponding intent, ie, a preset intent, is configured for each third audio sample, and the preset intent may be regarded as the real intent of the third audio sample.

[0125] The third audio sample and the first audio sample may be the same sample or different samples; the third audio sample and the second audio sample may be the same sample or different samples.

[0126] Step S420: input the third audio sample into the intention recognition model to predict the intention of the first audio sample and obtain intention data corresponding to the third audio sample.

[0127] The third audio sample is input into the intent recognition model, the data output by the intent recognition model for the third audio sample is obtained, and the intent data corresponding to the third audio sample is obtained. The intent data may be an explicit intent or an implicit intent. For consistency, if the preset intent is an explicit intent, the intent data corresponding to the third audio sample is an explicit intent; if the preset intent is an implicit intent, the intent data corresponding to the third audio sample is an implicit intent.

[0128] Step S430: training the intent recognition model based on the intent data corresponding to the third audio sample and the preset intent.

[0129] It should be noted that in order to obtain a trained intent recognition model, in this embodiment, in addition to training the intent recognition model based on the comparison result of the first target audio and the first audio sample, the preset intent corresponding to the third audio sample can also be input into the intent recognition model, so that the intent recognition model adjusts internal parameters based on the intent data corresponding to the third audio sample and the preset intent. In this way, by training the intent recognition model in two different ways, not only the training speed can be improved, but also the recognition accuracy of the intent recognition model can be improved.

[0130] It should be noted that the types of semanticless audio included in the samples for training the intention recognition model (including the aforementioned first audio sample, the second audio sample and the third audio sample) can be flexibly set according to actual needs. For example, multiple types of semanticless audio samples can be included, so that after the intention recognition model is trained based on the samples, the trained model can recognize multiple types of semanticless audio. Alternatively, since different types of semanticless sounds contain different features and there are differences in the recognition methods, in order to improve the accuracy of model recognition, only one type of semanticless audio sample can be included, so that after the intention recognition model is trained based on the samples, the trained model has high recognition accuracy. Among them, the type classification method of semanticless audio can be set according to actual needs. For example, it can be divided according to the type of organism and the type of sound. For example, the calls of dogs, cats, and cows are respectively regarded as a type of semanticless audio, and the crying and laughing of infants and young children are respectively regarded as another type of semanticless audio. For example, if only an intention recognition model that can recognize the crying of infants and young children is needed, the samples corresponding to the intention recognition model can only include the crying of infants and young children.

[0131] In this embodiment, the third audio sample and the preset intent corresponding to the third audio sample are also used to train the intent recognition model, thereby further improving the accuracy of intent recognition based on the intent recognition model.

[0132] In an exemplary embodiment, see Figure 5 As shown, in Figure 1After step S140, the method further includes steps S510-S530, which are described in detail as follows:

[0133] Step S510, obtaining audio to be recognized.

[0134] The audio to be recognized is the audio to be recognized.

[0135] The audio to be recognized may be audio obtained by real-time collection of ambient sound. For example, when the method proposed in this embodiment is applied to infant care, the sound emitted by the infant may be collected in real time, the intention may be recognized, and the intention may be output, so that the caregiver can understand the intention of the infant in real time. The audio to be recognized may also be audio selected by the user, or other audio.

[0136] Step S520: input the audio to be recognized into the trained intention recognition model to recognize the intention of the audio to be recognized, thereby obtaining the intention of the audio to be recognized.

[0137] The audio to be recognized is input into the intent recognition model, the intent recognition model recognizes the intent of the audio to be recognized, and outputs the intent of the audio to be recognized.

[0138] Step S530: output the intent of the audio to be recognized.

[0139] The method of outputting the intent of the audio to be recognized can be flexibly set according to actual needs. For example, corresponding text information can be generated based on the intent of the audio to be recognized, and the generated text information can be output. Among them, a corpus can be pre-set, and the corpus includes texts corresponding to different intents, and the corresponding text information is obtained from the corpus based on the intent of the audio to be recognized.

[0140] Outputting generated text information includes but is not limited to the following three methods:

[0141] The first method: Generate corresponding audio based on text information and play the audio.

[0142] Among them, Fastspeech2 technology can be used to generate corresponding audio based on text information, thereby achieving good speech generation effects.

[0143] The second type: display text information.

[0144] For example, the corresponding text information is displayed on the display screen.

[0145] The third method: Send text messages to preset contacts.

[0146] The text message may be sent to a preset contact via SMS, email, interactive software, etc., so that the preset contact is aware of the intention of the audio to be recognized.

[0147] It should be noted that the above three methods can be adopted simultaneously.

[0148] In this embodiment, after the intention of the audio to be recognized is obtained, the intention of the audio to be recognized is output, so that the user knows the intention of the audio to be recognized.

[0149] In an exemplary embodiment, taking the recognition of infant crying as an example, the audio recognition method based on machine learning proposed in the embodiment of the present application is described, and the method includes:

[0150] Step S601, obtaining a first audio sample and a second audio sample without semantics emitted by a living being.

[0151] Among them, the first audio sample, the second audio sample and the subsequent third audio sample are all the crying sounds of infants.

[0152] Step S602: input the first audio sample into the intent recognition model, and obtain data output by the hidden layer of the intent recognition model for the first audio sample as first intent data.

[0153] Step S603: input the first audio sample into a timbre recognition model to recognize the timbre of the first audio sample to obtain first timbre data.

[0154] Step S604: input the second audio sample into the intent recognition model, and obtain data output by the hidden layer of the intent recognition model for the second audio sample as second intent data.

[0155] Step S605: input the second audio sample into a timbre recognition model to recognize the timbre of the second audio sample to obtain second timbre data.

[0156] Step S606: input the first intention data and the second timbre data into the audio generation model to obtain the first audio; input the second intention data and the first timbre data into the audio generation model to obtain the second audio.

[0157] Step S607: input the first audio into the intention recognition model, and obtain data output by the hidden layer of the intention recognition model for the first audio as third intention data.

[0158] Step S608: input the first audio into a timbre recognition model to recognize the timbre of the first audio to obtain fourth timbre data.

[0159] Step S609: input the second audio into the intent recognition model, and obtain data output by the hidden layer of the intent recognition model for the second audio as fourth intent data.

[0160] Step S610: Input the second audio into a timbre recognition model to recognize the timbre of the second audio to obtain third timbre data.

[0161] Step S611, input the third intention data and the third timbre data into the audio generation model to obtain the first target audio; input the fourth intention data and the fourth timbre data into the audio generation model to obtain the second target audio.

[0162] The specific process of obtaining the first target audio and the second target audio can be found in Figure 6 As shown, Figure 6 In order to facilitate viewing, three intent recognition models are shown. In essence, these three intent recognition models are the same model. Figure 6 The three audio generation models in are the same model, and the three timbre recognition models are the same model.

[0163] The first intention data, the second intention data, the third intention data, the fourth intention data, the first timbre data, the second timbre data, the third timbre data and the fourth timbre data may all be in a matrix form.

[0164] Step S612: training the intention recognition model, the timbre recognition model and the audio generation model based on the comparison result of the first target audio and the first audio sample.

[0165] The first target audio and the first audio sample may be compared based on a loss function.

[0166] Step S613: training the intention recognition model, the timbre recognition model and the audio generation model based on the comparison result of the second target audio and the second audio sample.

[0167] The second target audio and the second audio sample may be compared based on the loss function.

[0168] Step S614, obtaining a third audio sample without semantics emitted by the living being and a preset intention of the third audio sample.

[0169] Step S615: input the third audio sample into the intention recognition model to predict the intention of the first audio sample and obtain intention data corresponding to the third audio sample.

[0170] Step S616: training the intent recognition model based on the intent data corresponding to the third audio sample and the preset intent.

[0171] It should be noted that, in this embodiment, the order of steps S612, S613 and S616 can be flexibly set according to actual needs. For example, they can be performed synchronously or crosswise (for example, for the training of the intent recognition model, which requires a large number of samples, steps S612 and S613 can be performed based on a first audio sample and a second audio sample, and then step S616 can be performed based on a third audio sample; then, based on the next first audio sample and the next second audio sample, steps S612 and S613 can be performed, and then step S616 can be performed based on the next third audio sample, and the cycle can be repeated).

[0172] Step S617, when the intention recognition model meets the preset conditions, a trained intention recognition model is obtained.

[0173] The preset condition can be flexibly set according to actual needs.

[0174] Step S618, obtaining the audio to be recognized.

[0175] The audio to be recognized is the collected crying sound of an infant.

[0176] Step S619: input the audio to be recognized into the trained intention recognition model to recognize the intention of the audio to be recognized, thereby obtaining the intention of the audio to be recognized.

[0177] Step S620: output the intent of the audio to be recognized.

[0178] The corresponding text information can be generated based on the intent of the audio to be recognized, the corresponding audio can be generated based on the text information, and the audio can be played. For infants and young children, their crying sounds are usually hunger, sleepiness, fear, discomfort, etc. Therefore, hunger-related texts, sleepiness-related texts, fear-related texts, discomfort-related texts, etc. can be set in the corpus in advance, and then the corresponding text information can be obtained from the corpus based on the intent of the audio to be recognized.

[0179] It should be noted that the detailed processes involved in the above steps S601 to S620 have been described in the aforementioned embodiments, and therefore will not be repeated here.

[0180] It should be noted that the aforementioned steps S601-S617 are the training process of the intent recognition model; steps S618-S620 are the application process of the intent recognition model, and the execution subjects of the training process and the application process may be inconsistent; for example, the intent recognition model can be trained based on the server, and after the training is completed, the client (such as a mobile phone or other device) can download the intent recognition model to perform intent recognition based on the trained intent recognition model. Of course, the execution subjects of the training process and the intent recognition process can be the same.

[0181] In an exemplary embodiment, in order to facilitate the intention recognition model to process the audio, after acquiring the audio (including but not limited to the first audio sample, the second audio sample, the third audio sample and the audio to be recognized), if the audio is analog audio, the analog audio can be first converted to digital audio (if the acquired audio is already digital audio, no processing is required), and then the converted digital audio is input into the intention recognition model. The process of converting analog audio to digital audio can be: adopting, quantizing, and encoding the analog audio to obtain digital audio, that is, taking points in the continuous time domain waveform of the analog audio at a certain frequency to obtain discrete points, and quantizing and encoding the discrete points (that is, encoding them as 0 or 1 according to the amplitude of each discrete point), thereby obtaining digital audio.

[0182] See also Figure 7 , Figure 7 : is a block diagram of an audio recognition device based on machine learning shown in an exemplary embodiment of the present application. The device includes:

[0183] The acquisition module 701 is configured to acquire a first non-semantic audio sample emitted by a living being; the recognition module 702 is configured to input the first audio sample into an intention recognition model to predict the intention of the first audio sample and obtain first intention data; and recognize the timbre of the first audio sample to obtain first timbre data; wherein the intention recognition model is a model established based on machine learning; the determination module 703 is configured to determine the first target audio based on the first intention data and the first timbre data; the training module 704 is configured to train the intention recognition model based on the comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, and recognize the intention of the audio to be recognized based on the trained intention recognition model.

[0184] In another exemplary embodiment, the identification module 702 includes:

[0185] The intention prediction module is configured to input the first audio sample into the intention recognition model so that the intention recognition model predicts the intention of the first audio sample.

[0186] The first data acquisition module is configured to obtain data output by a hidden layer of the intention recognition model during the prediction process of the intention recognition model.

[0187] The intention acquisition module is configured to use the acquired data as the first intention data.

[0188] In another exemplary embodiment, the determining module 703 includes:

[0189] The second data acquisition module is configured to acquire second intention data and second timbre data.

[0190] The first generating module is configured to generate a first audio based on the first intention data and the second timbre data, and to generate a second audio based on the second intention data and the first timbre data.

[0191] The first data recognition module is configured to input the first audio into the intention recognition model to predict the intention of the first audio to obtain third intention data; and recognize the timbre of the second audio to obtain third timbre data.

[0192] The second generating module is configured to generate the first target audio based on the third intention data and the third timbre data.

[0193] In another exemplary embodiment, the second data acquisition module includes:

[0194] The first sample acquisition module is configured to acquire a second non-semantic audio sample emitted by a living being.

[0195] The second data recognition module is configured to input the second audio sample into the intention recognition model to predict the intention of the second audio sample to obtain second intention data; and recognize the timbre of the second audio sample to obtain second timbre data.

[0196] In another exemplary embodiment, the apparatus further comprises:

[0197] The third data recognition module is configured to input the second audio into the intention recognition model to predict the intention of the second audio to obtain fourth intention data; and recognize the timbre of the first audio to obtain fourth timbre data.

[0198] a third generating module configured to generate a second target audio based on the fourth intention data and the fourth timbre data;

[0199] The first model training module is configured to train the intent recognition model based on the comparison result of the second target audio and the second audio sample.

[0200] In another exemplary embodiment, under the condition that the first audio is generated by the audio generation model based on the first intention data and the second timbre data, and the first timbre data is obtained by recognizing the first audio sample by the timbre recognition model, the apparatus further includes:

[0201] The second model training module is configured to train the audio generation model and the timbre recognition model based on the comparison result of the first target audio and the first audio sample.

[0202] In another exemplary embodiment, the apparatus further comprises:

[0203] The third data acquisition module is configured to acquire a third non-semantic audio sample emitted by a living being and a preset intention of the third audio sample.

[0204] The fourth data recognition module is configured to input the third audio sample into the intention recognition model to predict the intention of the third audio sample and obtain intention data corresponding to the third audio sample.

[0205] The third model training module is configured to train the intent recognition model based on the intent data corresponding to the third audio sample and the preset intent.

[0206] In another exemplary embodiment, the apparatus further comprises:

[0207] The fourth data acquisition module is configured to acquire audio to be recognized.

[0208] The intention recognition module is configured to input the audio to be recognized into the trained intention recognition model to recognize the intention of the audio to be recognized, thereby obtaining the intention of the audio to be recognized;

[0209] The output module is configured to output the intent of the audio to be recognized.

[0210] It should be noted that the device provided in the above embodiment and the method provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment and will not be repeated here.

[0211] An embodiment of the present application also provides an electronic device, including a storage device and one or more processors, wherein the storage device is used to store one or more programs, and when the one or more programs are executed by one or more processors, the electronic device implements the aforementioned method.

[0212] Figure 8 The diagram is a schematic structural diagram of an electronic device according to an exemplary embodiment.

[0213] It should be noted that the electronic device is only an example adapted to the present application and cannot be considered to provide any limitation on the scope of use of the present application. The electronic device cannot be interpreted as requiring dependence on or having Figure 8 One or more components of an exemplary electronic device shown in FIG.

[0214] like Figure 8 As shown, in an exemplary embodiment, the electronic device includes a processing component 801, a memory 802, a power component 803, a multimedia component 804, an audio component 805, a sensor component 807, and a communication component 808. Among them, the above components are not all necessary, and the electronic device can add other components or reduce certain components according to its own functional requirements, which is not limited in this embodiment.

[0215] The processing component 801 generally controls the overall operation of the electronic device, such as operations associated with display, data communication, and log data processing. The processing component 801 may include one or more processors 809 to execute instructions to complete all or part of the steps of the above operations. In addition, the processing component 801 may include one or more modules to facilitate the interaction between the processing component 801 and other components. For example, the processing component 801 may include a multimedia module to facilitate the interaction between the multimedia component 804 and the processing component 801.

[0216] The memory 802 is configured to store various types of data to support operations on the electronic device, examples of which include instructions for any application or method for operating on the electronic device. The memory 802 stores one or more modules, which are configured to be executed by the one or more processors 809 to complete all or part of the steps in the method described in the above embodiment.

[0217] The power supply component 803 provides power to various components of the electronic device. The power supply component 803 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device.

[0218] The multimedia component 804 includes a screen that provides an output interface between the electronic device and the user. In some embodiments, the screen may include a TP (Touch Panel) and an LCD (Liquid Crystal Display). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0219] The audio component 805 is configured to output and / or input audio signals. For example, the audio component 805 includes a microphone, and when the electronic device is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. In some embodiments, the audio component 805 also includes a speaker for outputting audio signals.

[0220] The sensor component 807 includes one or more sensors for providing status evaluation of various aspects for the electronic device. For example, the sensor component 807 can detect the on / off status of the electronic device and can also detect the temperature change of the electronic device.

[0221] The communication component 808 is configured to facilitate wired or wireless communication between the electronic device and other devices. The electronic device can access a wireless network based on a communication standard, such as Wi-Fi (Wireless-Fidelity).

[0222] Understandably, Figure 8 The structure shown is for illustration only. The electronic device may include Figure 8 More or fewer components as shown, or with Figure 8 The different components are shown. Figure 8 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0223] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0224] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.

[0225] Another aspect of the present application further provides a computer-readable storage medium having computer-readable instructions stored thereon, which, when executed by a processor of a computer, causes the computer to execute the aforementioned method. The computer-readable storage medium may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device.

[0226] It should be noted that the computer-readable storage medium of the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution system, device or device or used in combination with it. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0227] Another aspect of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the method provided in each of the above embodiments.

[0228] The above content is only a preferred exemplary embodiment of the present application and is not intended to limit the implementation scheme of the present application. A person skilled in the art can easily make corresponding changes or modifications based on the main concept and spirit of the present application. Therefore, the scope of protection of the present application shall be based on the scope of protection required by the claims.

Claims

1. An audio recognition method based on machine learning, characterized in that: The method comprises: Obtaining a first non-semantic audio sample emitted by the creature; Inputting the first audio sample into an intention recognition model to predict the intention of the first audio sample to obtain first intention data; and recognizing the timbre of the first audio sample to obtain first timbre data; wherein the intention recognition model is a model established based on machine learning; Determine a first target audio based on the first intention data and the first timbre data; Training the intention recognition model based on the comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, and recognizing the intention of the audio to be recognized based on the trained intention recognition model; The determining of the first target audio based on the first intention data and the first timbre data includes: Acquire second intention data and second timbre data; generating a first audio based on the first intention data and the second timbre data, and generating a second audio based on the second intention data and the first timbre data; Inputting the first audio into the intention recognition model to predict the intention of the first audio to obtain third intention data; and recognizing the timbre of the second audio to obtain third timbre data; The first target audio is generated based on the third intention data and the third timbre data.

2. The method according to claim 1, characterized in that The step of inputting the first audio sample into an intention recognition model to predict the intention of the first audio sample and obtain first intention data includes: Inputting the first audio sample into an intention recognition model so that the intention recognition model predicts the intention of the first audio sample; During the prediction process of the intention recognition model, obtaining data output by a hidden layer of the intention recognition model; The acquired data is used as the first intention data.

3. The method according to claim 2, characterized in that The obtaining of the second intention data and the second timbre data comprises: Obtaining a second non-semantic audio sample emitted by the creature; Inputting the second audio sample into the intention recognition model to predict the intention of the second audio sample to obtain second intention data; and recognizing the timbre of the second audio sample to obtain second timbre data; After generating the first audio based on the first intent data and the second timbre data, and generating the second audio based on the second intent data and the first timbre data, and before obtaining the trained intent recognition model, the method further includes: Inputting the second audio into the intention recognition model to predict the intention of the second audio to obtain fourth intention data; and recognizing the timbre of the first audio to obtain fourth timbre data; generating a second target audio based on the fourth intention data and the fourth timbre data; The intention recognition model is trained based on a comparison result of the second target audio and the second audio sample.

4. The method according to claim 2, characterized in that The first audio is generated by an audio generation model based on the first intention data and the second timbre data; the first timbre data is obtained by recognizing a first audio sample by a timbre recognition model; The audio generation model and the timbre recognition model are models established based on machine learning; After generating the first target audio based on the third intention data and the third timbre data, the method further includes: The audio generation model and the timbre recognition model are trained based on the comparison result of the first target audio and the first audio sample.

5. The method according to claim 1, characterized in that Before obtaining the trained intent recognition model, the method further includes: Acquire a third audio sample without semantics emitted by a living being and a preset intention of the third audio sample; Inputting the third audio sample into the intention recognition model to predict the intention of the third audio sample and obtain intention data corresponding to the third audio sample; The intention recognition model is trained based on the intention data corresponding to the third audio sample and the preset intention.

6. The method according to claim 1, characterized in that After training the intention recognition model based on the comparison result between the first target audio and the first audio sample to obtain a trained intention recognition model, the method further includes: Get the audio to be recognized; Inputting the audio to be recognized into the trained intention recognition model to recognize the intention of the audio to be recognized, thereby obtaining the intention of the audio to be recognized; Output the intention of the audio to be recognized.

7. An audio recognition device based on machine learning, characterized in that: The device comprises: An acquisition module configured to acquire a first non-semantic audio sample emitted by a living being; a recognition module configured to input the first audio sample into an intention recognition model to predict the intention of the first audio sample to obtain first intention data; and recognize the timbre of the first audio sample to obtain first timbre data; wherein the intention recognition model is a model established based on machine learning; a determination module, configured to determine a first target audio based on the first intention data and the first timbre data; a training module configured to train the intention recognition model based on a comparison result of the first target audio and the first audio sample to obtain a trained intention recognition model, and recognize the intention of the audio to be recognized based on the trained intention recognition model; The determining of the first target audio based on the first intention data and the first timbre data includes: Acquire second intention data and second timbre data; generating a first audio based on the first intention data and the second timbre data, and generating a second audio based on the second intention data and the first timbre data; Inputting the first audio into the intention recognition model to predict the intention of the first audio to obtain third intention data; and recognizing the timbre of the second audio to obtain third timbre data; The first target audio is generated based on the third intention data and the third timbre data.

8. An electronic device, characterized in that: include: one or more processors; A storage device, used to store one or more programs, when the one or more programs are executed by the one or more processors, enables the electronic device to implement the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: Computer-readable instructions are stored thereon, and when the computer-readable instructions are executed by a processor of a computer, the computer is caused to execute the method according to any one of claims 1 to 6.

Images