Audio signal processing method, device, equipment and storage medium

Abstract

The present disclosure relates to a method, apparatus, device and storage medium for processing an audio signal. The present disclosure obtains a target audio signal, processes the target audio signal according to a noise covariance matrix, and obtains a target audio signal after noise suppression. Feature extraction is performed on the target audio signal after noise suppression to obtain multiple feature vectors, and the multiple feature vectors are synthesized into a single vector. Furthermore, it can be determined whether the target audio signal includes a speech signal based on the single vector. If the target audio signal does not include a speech signal, the noise covariance matrix is ​​updated so that the updated noise covariance matrix does not contain speech components as much as possible, and suppresses noise as much as possible without suppressing useful speech components. Therefore, when there is large noise in the audio signal collected by the electronic device, the noise in the audio signal can be effectively suppressed, so that the user voice can be recognized and human-computer voice interaction can be performed.

Description

Technical Field

[0001] The present disclosure relates to the field of information technology, and in particular to a method, apparatus, device, and storage medium for processing an audio signal. Background Art

[0002] With the continuous advancement of technology, a wide variety of voice-interactive electronic devices have emerged. Specifically, users can control electronic devices through voice interaction. For example, electronic devices can use voice recognition technology to convert user voice commands into text, and use semantic understanding technology to understand the intent of the commands and provide corresponding feedback. Therefore, the prerequisite for human-computer voice interaction is that the electronic device can clearly hear the voice commands.

[0003] However, the inventors of this application found that some electronic devices themselves emit relatively loud noises, such as sweeping robots, which results in relatively loud noises in the audio signals collected by the electronic devices and makes it impossible to recognize the user's voice commands, thus preventing human-computer voice interaction from proceeding normally. Summary of the Invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a method, apparatus, device and storage medium for processing audio signals, which can effectively suppress the noise in the audio signal through the updated noise covariance matrix, so that the electronic device can recognize the user's voice and then conduct normal human-computer voice interaction.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for processing an audio signal, comprising:

[0006] Acquire target audio signal;

[0007] Processing the target audio signal according to the noise covariance matrix to obtain a noise-suppressed target audio signal;

[0008] Extracting features from the noise-suppressed target audio signal to obtain multiple feature vectors;

[0009] Combining the plurality of feature vectors into a single vector;

[0010] determining whether the target audio signal includes a speech signal according to the single vector;

[0011] If the target audio signal does not include a speech signal, the noise covariance matrix is ​​updated, and the updated noise covariance matrix is ​​used to process the next audio signal of the target audio signal.

[0012] In a second aspect, an embodiment of the present disclosure provides an audio signal processing device, comprising:

[0013] An acquisition module, used to acquire a target audio signal;

[0014] a processing module, configured to process the target audio signal according to the noise covariance matrix to obtain a noise-suppressed target audio signal;

[0015] A feature extraction module is used to extract features from the target audio signal after noise suppression to obtain multiple feature vectors;

[0016] A synthesis module, configured to synthesize the plurality of feature vectors into a single vector;

[0017] a determination module, configured to determine whether the target audio signal includes a speech signal according to the single vector;

[0018] An updating module is configured to update the noise covariance matrix when the target audio signal does not include a speech signal, and the updated noise covariance matrix is ​​used to process an audio signal next to the target audio signal.

[0019] In a third aspect, an embodiment of the present disclosure provides an electronic device, including:

[0020] Memory;

[0021] processor; and

[0022] computer programs;

[0023] The computer program is stored in the memory and is configured to be executed by the processor to implement the method as described in the first aspect.

[0024] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method described in the first aspect.

[0025] The audio signal processing method, apparatus, device and storage medium provided by the embodiments of the present disclosure obtain a target audio signal, process the target audio signal according to the noise covariance matrix, and obtain a target audio signal after noise suppression. Feature extraction is performed on the target audio signal after noise suppression to obtain multiple feature vectors, and the multiple feature vectors are synthesized into a single vector. Further, based on the single vector, it can be determined whether the target audio signal includes a speech signal. If the target audio signal does not include a speech signal, it means that the target audio signal includes non-speech signals such as silence and noise. At this time, the noise covariance matrix is ​​updated so that the updated noise covariance matrix does not contain speech components as much as possible, so that the updated noise covariance matrix suppresses noise as much as possible without suppressing useful speech components. Therefore, when there is a large amount of noise in the audio signal collected by the electronic device, the noise in the audio signal can be effectively suppressed by the updated noise covariance matrix, so that the electronic device can recognize the user's voice and then perform normal human-computer voice interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0027] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0028] Figure 1 A flow chart of a method for processing an audio signal provided in an embodiment of the present disclosure;

[0029] Figure 2 A schematic diagram of an application scenario provided by an embodiment of the present disclosure;

[0030] Figure 3 A flowchart of a method for processing an audio signal provided by another embodiment of the present disclosure;

[0031] Figure 4 A flowchart of a method for processing an audio signal provided by another embodiment of the present disclosure;

[0032] Figure 5 A flowchart of a method for processing an audio signal provided by another embodiment of the present disclosure;

[0033] Figure 6 A schematic diagram of the structure of an audio signal processing and positioning device provided by an embodiment of the present disclosure;

[0034] Figure 7 A schematic diagram of the structure of an electronic device embodiment provided by the present disclosure. DETAILED DESCRIPTION

[0035] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features therein can be combined with each other in the absence of conflict.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0037] With the development of computers, the internet, mobile internet, and the Internet of Things, as well as advances in related science and technology and the accumulation of data in specific business fields, the electronic devices people commonly use are evolving from computers and mobile phones to more diverse, miniaturized, and wearable devices. The traditional human-computer interaction methods based on keyboards, mice, and remote controls are no longer sufficient to meet the needs of this increasingly diverse electronic device. Voice is the most natural way for people to interact with each other, and is therefore expected to become one of the most effective human-computer interaction methods of the future.

[0038] Specifically, users can control electronic devices through voice interaction. For example, electronic devices can convert user voice commands into text using voice recognition technology and understand the intent of the commands through semantic understanding technology, thereby providing corresponding feedback. Therefore, the premise of human-computer voice interaction is that the electronic device can clearly hear the voice commands. However, some electronic devices themselves emit a lot of noise, such as sweeping robots, which causes the audio signals collected by the electronic devices to contain a lot of noise, making it impossible to recognize the user's voice commands, resulting in the human-computer voice interaction not being able to proceed normally.

[0039] Normally, the signal-to-noise ratio of the signal collected by the microphone on the electronic device is relatively low, which is not conducive to subsequent recognition and wake-up processing. The role of speech enhancement is to improve the signal-to-noise ratio and speech intelligibility of the target speech through algorithm processing, so as to facilitate subsequent human-computer interaction. For example, the sweeping robot itself will emit noise, such as motor sound, various mechanical friction sounds, etc. Since the sound pickup device is installed on the sweeping robot, its distance from the noise source is much smaller than the distance from the target speaker, so the original signal collected is often relatively low, which is not conducive to subsequent processing. Therefore, it is necessary to suppress the self-noise of the sweeping robot, that is, ego-noise suppression, so as to enhance the target speech. Therefore, the embodiments of the present disclosure can not only be applied to the self-noise suppression of the sweeping robot, but also to other electronic devices working in a higher noise environment, such as drones, shopping guide robots, transmission robots, etc. The signal-to-noise ratio of the audio signals collected by these electronic devices is usually low. That is to say, the embodiments of the present disclosure can also be applied to some electronic devices with low signal-to-noise ratio, for example, electronic devices whose signal-to-noise ratio of the collected audio signal is lower than the threshold. In addition, since the method provided by the embodiment of the present disclosure requires relatively low computing power, it can also be applied to various low-resource embedded systems.

[0040] To address this problem, an embodiment of the present disclosure provides a method for processing an audio signal, which is described below in conjunction with specific embodiments.

[0041] Figure 1 Flowchart of the method for processing audio signals provided in the embodiment of the present disclosure. The method can be executed by the electronic device with low signal-to-noise ratio as described above, or by a server. In the embodiment of the present disclosure, the electronic device with low signal-to-noise ratio can be some electronic devices working in a relatively high noise environment, such as drones, mobile robots, and some smart home devices in a relatively high noise environment. Among them, some smart home devices are also gradually developing in the direction of voice and intelligence, such as sweeping robots. The following is a schematic illustration taking a sweeping robot as an example, for example Figure 2 The cleaning robot 21 shown can execute the audio signal processing method provided by the embodiment of the present disclosure. Alternatively, the cleaning robot 21 and the server 22 can exchange information so that the server 22 can execute the audio signal processing method provided by the embodiment of the present disclosure. Figure 1 As shown, the specific steps of this method are as follows:

[0042] S101: Acquire a target audio signal.

[0043] like Figure 2As shown, the sweeping robot 21 may be provided with an audio collection device, which may also be referred to as a sound pickup device, specifically a microphone. In addition, the sweeping robot 21 may be provided with one or more microphones. For example, the microphone 23 may be provided at the edge of the sweeping robot 21 to effectively collect audio signals.

[0044] In one feasible implementation, a processing device within the sweeping robot 21, such as a processor, can obtain an audio signal collected by the microphone 23. The audio signal can be a long audio signal. At this time, the processor can divide the audio signal into multiple segments and assign an identifier to each segment, such as a sequence number or index number. The sequence number or index number of each segment can be related to the time when the segment was collected. For example, the earlier the collection time, the smaller the sequence number of the segment; the later the collection time, the larger the sequence number of the segment. Any one of the multiple segments can be recorded as the target audio signal.

[0045] In another feasible implementation, the audio signal collected by microphone 23 is relatively short. For example, microphone 23 can collect an audio signal in a very short time. In this case, each audio signal collected by microphone 23 can be recorded as a target audio signal. In this case, each target audio signal will also be associated with identification information, such as a sequence number, which can also be related to the collection time.

[0046] It is understandable that in some scenarios, the microphone 23 can collect audio signals in real time. At this time, the processor in the sweeping robot 21 can treat each target audio signal as a unit and process each unit in turn to realize the processing of real-time audio signals.

[0047] S102: Process the target audio signal according to the noise covariance matrix to obtain a noise-suppressed target audio signal.

[0048] Currently, the sweeping robot 21 itself emits some noise when in operation, such as mechanical noise, motor noise, vacuum cleaner noise, etc., and these noises are generally quite loud. The microphone 23 is mounted on the body of the sweeping robot 21, close to these noise sources. Therefore, the audio signal collected by the microphone 23 is always mixed with these noises. Therefore, after the processor acquires the target audio signal, it can perform noise suppression on the target audio signal based on the noise covariance matrix. In other words, it suppresses the noise components in the target audio signal to obtain a noise-suppressed target audio signal. It is understood that in some cases, the target audio signal may include noise but no useful speech components, such as user speech. In other cases, the target audio signal may contain both noise and useful speech components. Or in some other embodiments, the target audio signal may include silence. Therefore, to distinguish, the useful speech components can be recorded as speech signals, and other components other than speech signals, such as noise and silence, can be recorded as non-speech signals.

[0049] S103: Perform feature extraction on the noise-suppressed target audio signal to obtain multiple feature vectors.

[0050] For example, the processor in the cleaning robot 21 can perform feature extraction on the target audio signal after noise suppression to obtain multiple feature vectors.

[0051] S104: Combining the multiple feature vectors into a single vector.

[0052] For example, the processor may combine multiple feature vectors into a single vector.

[0053] S105: Determine whether the target audio signal includes a speech signal according to the single vector.

[0054] For example, the processor may determine whether the target audio signal includes a speech signal based on the single vector.

[0055] S106: If the target audio signal does not include a speech signal, update the noise covariance matrix, and use the updated noise covariance matrix to process the next audio signal of the target audio signal.

[0056] If the target audio signal does not include a speech signal, it indicates that the target audio signal may include non-speech signals such as silence and noise. In this case, the noise covariance matrix can be updated so that the updated noise covariance matrix can process the next audio signal after the target audio signal. In addition, if the target audio signal includes a speech signal, the noise covariance matrix is ​​not updated.

[0057] In addition, in other embodiments, the above S101-S106 may also be executed by a server, for example Figure 2 As shown, the cleaning robot 21 can send the audio signal it collects to the server 22, so that the server can execute the method described in S101-S106.

[0058] The embodiment of the present disclosure obtains a target audio signal, processes the target audio signal according to a noise covariance matrix, and obtains a target audio signal after noise suppression. Feature extraction is performed on the target audio signal after noise suppression to obtain multiple feature vectors, and the multiple feature vectors are synthesized into a single vector. Further, based on the single vector, it can be determined whether the target audio signal includes a speech signal. If the target audio signal does not include a speech signal, it means that the target audio signal includes non-speech signals such as silence and noise. At this time, the noise covariance matrix is ​​updated so that the updated noise covariance matrix does not contain speech components as much as possible, so that the updated noise covariance matrix suppresses noise as much as possible without suppressing useful speech components. Therefore, when there is a large noise in the audio signal collected by the electronic device, the noise in the audio signal can be effectively suppressed by the updated noise covariance matrix, so that the electronic device can recognize the user's voice and then perform normal human-computer voice interaction.

[0059] It is understandable that, since the sweeping robot 21 or the server 22 processes the target audio signal as a unit, the noise covariance matrix can be continuously updated in the process of processing each target audio signal one by one by the sweeping robot 21. In other words, the updating of the noise covariance matrix described in S106 is an update of the noise covariance matrix, and the updated noise covariance matrix can be used as the object of the next update, thereby continuously iterating. In the process of continuous iteration, the noise covariance matrix will have a stronger and stronger ability to suppress noise.

[0060] like Figure 3As shown, the sweeping robot 21 or the server 22 can implement functions such as voice enhancement and keyword detection. Taking the sweeping robot as an example, the sweeping robot may include an update module, a noise suppression module, a feature extraction module, a self-attention mechanism module, an acoustic model, a decoder, and a confidence calculation module. Among them, the update module can determine whether to update the noise covariance matrix based on the feedback information of the decoder. For example, when the feedback information indicates that the target audio signal does not include a voice signal, the update module can update the noise covariance matrix. When the feedback information indicates that the target audio signal includes a voice signal, the update module does not update the noise covariance matrix. x represents the input target audio signal. For example, the sweeping robot 21 can be provided with 6 microphones, and the 6 microphones can collect audio signals at the same time, thereby obtaining 6-channel audio signals. Therefore, the target audio signal can be a 6-channel audio signal. When x is input to the update module, the update module can determine whether to update the noise covariance matrix based on the feedback information of the decoder. The update module can output x and Φ, where Φ represents the noise covariance matrix. The noise covariance matrix output by the update module may be the updated noise covariance matrix or the unupdated noise covariance matrix. y represents the output of the noise suppression module, i.e., the target audio signal after noise suppression. f1,…,f N Represents the output of the feature extraction module.

[0061] Assume that the target audio signal input to the update module at the current time is Indicates the identifier of the target audio signal input at the current time, which can be a sequence number or an index number. In addition, It can also represent the current time or the acquisition time of the target audio signal. The noise covariance matrix output by the current time update module is The noise suppression module can be right Perform noise suppression to obtain the target audio signal after noise suppression The value of n can be 1, 2...N, where N is the number of outputs of the noise suppression module. For example, if the noise suppression module can output 3 signals, the value of n is 1, 2, or 3. The signal-to-noise ratio is higher than The noise suppression module is based on and Calculated The process can be expressed by the following formulas (1) and (2).

[0062]

[0063]

[0064] in, Represents the beamformer. When the noise suppression module uses the multi-look minimum variance distortionless response (MVDR) beamforming algorithm to calculate hour, It can be a Multi-look MVDR beamformer. ∈ represents a known number. I represents the identity matrix. represents the steering vector. represents the azimuth of the sweeping robot. θ represents the elevation angle of the steering vector. γ represents the parameter.

[0065] Output of the noise suppression module It can be used as the input of the feature extraction module. Since the noise suppression module can output three signals, the feature extraction module can output the feature vector of each signal in the three signals, that is, the feature extraction module can output three feature vectors, which can be recorded as f1, ..., f N , N=3. When the input is In the case of f1, ..., f N It can be recorded as The value of n is 1, 2, or 3. Furthermore, the self-attention mechanism module can synthesize the three feature vectors into one feature vector, and the synthesized feature vector is recorded as a single vector That is, when the input is In the case of Equivalent to Figure 3 in Specifically, the self-attention mechanism module can Perform weighted summation to obtain For example, It can be obtained by the following formulas (3), (4), and (5):

[0066]

[0067]

[0068]

[0069] Where v represents a known vector. express The weighting coefficient of .

[0070] On the basis of the above embodiment, determining whether the target audio signal includes a speech signal according to the single vector may include: Figure 4 The following steps are shown:

[0071] S401. Input the single vector into an acoustic model, wherein the acoustic model includes multiple sub-models, and the acoustic model outputs multiple first conditional probabilities, wherein the multiple first conditional probabilities correspond one-to-one to the multiple sub-models, and each first conditional probability is used to represent the probability of observing the single vector under the condition of any sub-model, and the multiple sub-models include models for detecting speech signals and non-speech signals, respectively.

[0072] like Figure 3 As shown, a single vector Input an acoustic model, which may include multiple sub-models, including models for detecting voice signals and non-voice signals respectively. Under normal circumstances, a user can wake up an electronic device through keyword voice. After waking up, the user can send some voice commands to the electronic device so that the electronic device executes the voice command. Therefore, the process of the user waking up the electronic device through keyword voice and the process of controlling the electronic device to execute the voice command can be regarded as a human-computer voice interaction process. Therefore, the model used to detect voice signals in the acoustic model can be a model for detecting keyword voice or a model for detecting voice commands. The following is a schematic explanation using the model for detecting keyword voice as an example. For example, the keyword used to wake up the sweeping robot 21 is "ABCD", which can also be called a wake-up word. The acoustic model may include 5 sub-models, which are, in order, a modeling model for non-voice signals, a modeling model for "A", a modeling model for "B", a modeling model for "C", and a modeling model for "D", where the non-voice signal can be recorded as "Filler", for example, Filler can be noise, silence, non-wake-up word voice, etc. A single vector Input the acoustic model, and the acoustic model can output 5 conditional probabilities. Here, the conditional probability output by the acoustic model is recorded as the first conditional probability. For example, the 5 first conditional probabilities output by the acoustic model are recorded as in, Indicates that under the condition of the modeling model of non-speech signal, it is observed probability. Indicates that under the condition of the modeling model of "A", the observation probability. Indicates that under the condition of the modeling model of "B", the observation probability. Indicates that under the condition of the modeling model of "C", the observation probability. Indicates that under the condition of the modeling model of "D", the observation The probability of a message arriving at the destination is 0. Specifically, the acoustic model can be based on a feedforward sequential memory neural network (FSMN). Specifically, the acoustic model can be a neural network model. Furthermore, the acoustic model can use not only neural network models based on the FSMN architecture but also networks based on other architectures, such as convolutional neural networks (CNNs).

[0073] Specifically, the acoustic model is based on Output The process can be obtained by the following formulas (6), (7), and (8):

[0074]

[0075]

[0076]

[0077] Where L represents the number of layers in the neural network model. h0(τ) in formula (6) represents the input of the first layer in the neural network model. Represents the output of the lth layer in the neural network model, h in formula (7) l-1 (τ) represents the output of the l-1th layer in the neural network model, for example, L = 5. h in formula (8) L (τ) represents the output of the Lth layer in the neural network model. is an array, They are Elements in .

[0078] S402: Determine whether the target audio signal includes a speech signal according to the multiple first conditional probabilities and one or more historical audio signals before the target audio signal.

[0079] For example, It can be used as the input of the decoder, and the decoder can determine the target audio signal according to the multiple first conditional probabilities and one or more historical audio signals before the target audio signal. Whether to include voice signals.

[0080] Optionally, determining whether the target audio signal includes a speech signal based on the multiple first conditional probabilities and one or more historical audio signals before the target audio signal includes: inputting the multiple first conditional probabilities into a decoder, and the decoder updating the probability transfer matrix corresponding to the decoder based on the multiple first conditional probabilities; determining whether the target audio signal includes a speech signal based on the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix.

[0081] Optionally, the decoder updates the probability transfer matrix corresponding to the decoder based on the multiple first conditional probabilities, including: the decoder calculates the updated forward probability based on the multiple first conditional probabilities, the historical forward probability and the historical probability transfer matrix; and calculates the updated probability transfer matrix based on the updated forward probability and the historical probability transfer matrix.

[0082] For example, After inputting the decoder, the decoder can The probability transfer matrix corresponding to the decoder is updated. Specifically, the decoder can use a Hidden Markov Model (HMM) decoder, which corresponds to a probability transfer matrix, which can also be called a state transfer matrix. The decoder can use Algorithm 1, which can implement online HMM transfer matrix update. For example, the probability transfer matrix before the update, that is, the historical probability transfer matrix, can be recorded as The updated probability transfer matrix can be recorded as The element in row i and column k of T(τ-1) can be written as The element in row i and column k can be written as In addition, the update process involves the forward probability and matrix The historical forward probability can be recorded as The i-th element in can be recorded as The size of is the same as the size of the probability transfer matrix, The element in row i and column j can be written as The specific update process can be obtained using the following formulas (9)-(14):

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089] Among them, p in formula (9) j (τ) represents The jth element in , according to formula (9) can be calculated as the matrix Each element in .

[0090] Formula (10) represents the historical forward probability The updated forward probability is recorded as In formula (10) express The jth element in . According to formula (10), we can calculate Each element in .

[0091] In formula (11) It is calculated according to formula (9). The numerator on the right side of the equal sign in formula (11) is It is calculated by formula (9) The matrix formed by formula (11) represents The standardization process, the numerator on the right side of the equation (11) is the matrix before normalization, the left side of the equal sign is is the matrix after normalization.

[0092] Formula (12) expresses The standardization process, the numerator on the right side of the equation (12) is the updated forward probability before normalization, It is calculated according to formula (10) constituted, Indicates that The i-th element in . The left side of the equal sign in formula (12) is the updated forward probability after normalization.

[0093] In formula (13) Indicates the left side of the equal sign in formula (12) The i-th element in η (2) Represents a fixed forgetting factor, i.e. η (2) Can be a constant. η i Represents the updated probability transfer matrix The forgetting factor corresponding to the i-th row in .

[0094] In formula (14) Represents the updated probability transfer matrix The element in row i and column j in formula (14) Indicates the left side of the equal sign in formula (11) The element in row i and column j in formula (14) Indicates the left side of the equal sign in formula (12) The i-th element in .

[0095] Furthermore, the decoder can One or more previous historical audio signals and the updated probability transfer matrix Sure Whether to include voice signals.

[0096] Optionally, determining whether the target audio signal includes a speech signal based on the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix includes: Figure 5 The following steps are shown:

[0097] S501: Output a recognition result according to the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and an updated probability transfer matrix.

[0098] For example, the target audio signal The previous historical audio signal can be recorded as x(τ-1). The previous multiple historical audio signals can be recorded as x(τ-m), x(τ-m+1), ... x(τ-1), where m represents a number greater than or equal to 1. It can be understood that when the input of the updated model is x(τ-1), the acoustic model can output 5 first conditional probabilities, which are similar to Similarly, when the input of the updated model is x(τ-m), x(τ-m+1), ..., or x(τ-2), the acoustic model can output similar 5 first conditional probabilities. Specifically, the decoder can be based on The corresponding 5 first conditional probabilities, and the 5 first conditional probabilities corresponding to x(τ-m), x(τ-m+1), ...x(τ-1), and the updated probability transfer matrix Output the recognition results.

[0099] S502: If the target audio signal includes a voice signal of a keyword, the recognition result is an identifier of the keyword.

[0100] If the decoder determines the target audio signal If the speech signal includes a keyword, the output recognition result is the identifier of the keyword.

[0101] For example, the keyword is "ABCD", "A", "B", "C", and "D" are recorded as keywords respectively. For example, the identifier of "A" is 1, the identifier of "B" is 2, the identifier of "C" is 3, and the identifier of "D" is 4. If the decoder determines that the target audio signal If the speech signal of "D" is included in x(τ-m), x(τ-m+1), ... or x(τ-1), the output recognition result is 4. Similarly, in the historical time, the decoder also determines whether the speech signal of a keyword is included in x(τ-m), x(τ-m+1), ... or x(τ-1), for example Figure 3 As shown, the decoder's recognition results for x(τ-m), x(τ-m+1), ...x(τ-1) are 00112223344, where 0 is the identifier of "Filler".

[0102] S503: If the target audio signal does not include a speech signal, the recognition result is a preset identifier.

[0103] For example, if the decoder determines that the target audio signal If the target audio signal does not contain any keyword, then the target audio signal is determined to be The speech signal is not included in the decoder. In this case, the recognition result output by the decoder is a preset flag, such as 0.

[0104] In addition, the decoder can also output the forward probability while outputting the recognition result. For example, when the input of the update module is When , the decoder outputs a recognition result and the forward probability corresponding to the recognition result include Here yes

[0105] When the input of the update module is x(τ-1), the decoder outputs a recognition result and the forward probability corresponding to the recognition result include Here yes And so on.

[0106] Specifically, the updated forward probability includes a second conditional probability, and the second conditional probability is used to represent the probability of observing a non-speech signal under the condition of the single vector.

[0107] For example, you can Denoted as the second conditional probability, the second conditional probability is expressed in a single vector The probability of observing the non-speech signal “Filler” under the condition of . When the input of the update module is different, the single vector Also different.

[0108] Correspondingly, if the target audio signal does not include a speech signal, updating the noise covariance matrix includes: if the target audio signal does not include a speech signal, updating the noise covariance matrix according to the second conditional probability.

[0109] For example Figure 3 As shown, the decoder can convert the second conditional probability Feedback to the update module as feedback information. Specifically, the update module can Determines whether to update the noise covariance matrix. For example, when the input of the update module is When the decoder outputs a recognition result of 1, 2, 3, or 4, it means Including the voice signal, at this time, yes The update module receives When the next audio signal x(τ+1) is received, the noise covariance matrix Φ(τ) is not updated, that is, Φ(τ+1)=Φ(τ), and Φ(τ+1) can be used to suppress the noise of x(τ+1). When the input of the update module is If the recognition result output by the decoder is 0, it means does not include speech signals, at this time, yes The update module receives When the next audio signal x(τ+1) is received, the noise covariance matrix Φ(τ) is updated to obtain an updated noise covariance matrix Φ(τ+1), where Φ(τ+1) is not equal to Φ(τ).

[0110] Similarly, when the input of the update module is x(τ-1), if the recognition result output by the decoder is a value among 1, 2, 3, and 4, it means that x(τ-1) contains the speech signal. At this time, yes The update module receives When the noise covariance matrix Φ(τ-1) is not updated, that is, Φ(τ)=Φ(τ-1), Φ(τ) can be used to When the input of the update module is x(τ-1), if the recognition result output by the decoder is 0, it means that x(τ-1) does not contain speech signals. At this time, yes The update module receives When , the noise covariance matrix x(τ-1) is updated. The updating process can be implemented by the following formulas (15) and (16), thereby obtaining the updated noise covariance matrix Φ(τ), where Φ(τ) is not equal to Φ(τ-1).

[0111]

[0112]

[0113] Among them, η (1) represents a fixed forgetting factor, η (1) and the above η (2) Different. η represents the dynamic forgetting factor. In formula (16) yes

[0114] Optionally, the confidence of the keyword is calculated based on the recognition result corresponding to the target audio signal and the recognition results corresponding to the one or more historical audio signals, and the keyword includes one or more keywords; if the confidence of the keyword is greater than a threshold, the electronic device is woken up.

[0115] For example Figure 3 As shown in , for each target audio signal input, the decoder can output a recognition result and the forward probability corresponding to the recognition result, which is composed of 5 conditional probabilities. Figure 3 Each value in 0011222334440 shown corresponds to a forward probability. Specifically, some continuous values ​​can be selected from 0011222334440, for example, 1122233444 can be selected through a sliding window. The confidence calculation module can calculate the confidence of the keyword "ABCD" based on the forward probability corresponding to each value in 1122233444. If the confidence is greater than the threshold, the sweeping robot is awakened. The confidence calculation process can be implemented by the following formulas (17), (18), and (19).

[0116]

[0117]

[0118]

[0119] Wherein, i in formula (17) = A, B, C, D, Indicates that the input is , the decoder outputs the recognition result. The identifier of the target audio signal x(τ0) input at the current time. τ0-Z+2, ..., τ0-1 represent the identifiers of multiple historical audio signals before x(τ0). Z represents the number of values ​​selected from 0011222334440, that is, the size of the sliding window. For example, when i=A in formula (17), since the identifier corresponding to A is 1, 1122233444 selected from 0011222334440 includes two 1s, and each 1 corresponds to a forward probability, and the forward probability corresponding to each 1 includes 5 conditional probabilities. For example, the forward probability corresponding to the first 1 in 1122233444 includes yes The forward probability corresponding to the second 1 includes yes At this time, q in formula (17) i Indicates that the two 1s correspond to The maximum value in q i It can be recorded as the maximum posterior probability, q i It represents the probability of the i-th keyword among the four keywords "A", "B", "C", and "D" appearing in the longer audio signal corresponding to 1122233444. Since each value in 1122233444 corresponds to an input, which is a shorter audio signal, the shorter audio signals corresponding to 1122233444 constitute a longer audio signal. Since i = A, B, C, D in formula (17), the four q can be calculated according to formula (17). i q in formula (19) j Compared with q in formula (17) i The meaning of is the same, and the formula (19) Indicates that when 4 q j A q in i When it is not equal to 0, the 4 q j Multiply and divide by q i When the 4 q j A q in i When it is equal to 0, is 0. The q in formula (18) represents the confidence of the keyword, and the meaning of q is 4 q j After sorting in descending order, the first three q jWhen the confidence level of the keyword is greater than the threshold, it indicates that the processor in the robot vacuum cleaner has detected the keyword, i.e., the wake-up word. At this time, the processor can wake up the robot vacuum cleaner so that it can further detect other user voices.

[0120] In traditional beamforming algorithms, estimating the steering vector (or steering vector) is often a challenge, and the accuracy of the estimation directly affects the noise reduction effect. However, in the disclosed embodiments, a fixed steering vector can be used, thus avoiding the problem of steering vector estimation. Instead, the noise covariance matrix is ​​iteratively updated to minimize the inclusion of speech components in the updated noise covariance matrix, thereby minimizing the amount of noise suppressed without suppressing the useful speech components.

[0121] In addition, if Figure 3 As shown, feedback information is sent to the update module through the decoder, so that the update module can determine whether to update the noise covariance matrix according to the feedback information, so that Figure 3 The iterative processes of the speech enhancement (SE) part and the keyword spotting (KWS) part shown in FIG are combined together to achieve better results than the individual iterations. Figure 3 The overall structure shown here iterates during the usage or inference phase. Keyword spotting (KWS), commonly known as voice wake-up, involves an electronic device continuously listening for a predefined keyword, such as "ABCD." Detecting the keyword indicates a need for human-computer interaction, and the corresponding interaction process is initiated.

[0122] Therefore, the noise in the audio signal collected by the sweeping robot can be effectively suppressed by the updated noise covariance matrix, thereby improving the signal-to-noise ratio of the audio signal after noise suppression. As a result, the method described in the embodiment of the present disclosure can be applied to scenarios where the sweeping robot has multiple noise sources, such as motor sounds, various mechanical friction sounds, etc., and the sound pickup device is close to the multiple noise sources. That is, in this scenario, the noise in the original signal collected by the sound pickup device can be effectively suppressed by continuously iteratively updating the noise covariance matrix, thereby greatly improving the signal-to-noise ratio. In addition, the sweeping robot is moving during operation, while the user may be stationary, resulting in the audio signal collected by the sweeping robot being real-time and dynamic. However, by continuously updating the probability transfer matrix by the decoder, the decoder can be dynamically adjusted according to different environments, thereby achieving better noise reduction and wake-up effects. Therefore, the embodiment of the present disclosure can effectively suppress the noise emitted by the electronic device itself under conditions of low signal-to-noise ratio, multiple interference sources, and moving sound sources, thereby achieving effective voice wake-up.

[0123] Furthermore, the method described in the embodiments of the present disclosure has a relatively small computational load, making it applicable to chips with low computing power and low resources, thereby reducing hardware costs and being applicable to various low-resource embedded systems.

[0124] Figure 6 The structure diagram of the audio signal processing device provided by the embodiment of the present disclosure is as follows. The audio signal processing device provided by the embodiment of the present disclosure can execute the processing flow provided by the embodiment of the audio signal processing method, such as Figure 6 As shown, the audio signal processing device 60 includes:

[0125] An acquisition module 61 is configured to acquire a target audio signal;

[0126] a processing module 62, configured to process the target audio signal according to the noise covariance matrix to obtain a noise-suppressed target audio signal;

[0127] A feature extraction module 63 is used to extract features from the noise-suppressed target audio signal to obtain multiple feature vectors;

[0128] a synthesis module 64, configured to synthesize the plurality of feature vectors into a single vector;

[0129] a determination module 65, configured to determine whether the target audio signal includes a speech signal according to the single vector;

[0130] The updating module 66 is configured to update the noise covariance matrix when the target audio signal does not include a speech signal, and the updated noise covariance matrix is ​​used to process the next audio signal of the target audio signal.

[0131] Optionally, when the determination module 65 determines whether the target audio signal includes a speech signal according to the single vector, it is specifically configured to:

[0132] Inputting the single vector into an acoustic model, the acoustic model including a plurality of sub-models, the acoustic model outputting a plurality of first conditional probabilities, the plurality of first conditional probabilities corresponding one-to-one to the plurality of sub-models, each first conditional probability being used to represent a probability of observing the single vector under the condition of any sub-model, the plurality of sub-models including models for detecting speech signals and non-speech signals respectively;

[0133] It is determined whether the target audio signal includes a speech signal according to the multiple first conditional probabilities and one or more historical audio signals before the target audio signal.

[0134] Optionally, when the determination module 65 determines whether the target audio signal includes a speech signal based on the multiple first conditional probabilities and one or more historical audio signals before the target audio signal, it is specifically configured to:

[0135] Inputting the multiple first conditional probabilities into a decoder, wherein the decoder updates a probability transfer matrix corresponding to the decoder according to the multiple first conditional probabilities;

[0136] It is determined whether the target audio signal includes a speech signal according to the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix.

[0137] Optionally, when the decoder updates the probability transfer matrix corresponding to the decoder according to the multiple first conditional probabilities, it is specifically used to:

[0138] The decoder calculates an updated forward probability based on the multiple first conditional probabilities, the historical forward probability and the historical probability transfer matrix;

[0139] An updated probability transfer matrix is ​​calculated based on the updated forward probability and the historical probability transfer matrix.

[0140] Optionally, the updated forward probability includes a second conditional probability, where the second conditional probability is used to represent the probability of observing a non-speech signal under the condition of the single vector; when the target audio signal does not include a speech signal, the updating module 66 updates the noise covariance matrix, specifically for:

[0141] If the target audio signal does not include a speech signal, the noise covariance matrix is ​​updated according to the second conditional probability.

[0142] Optionally, when the determination module 65 determines whether the target audio signal includes a speech signal based on the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix, it is specifically configured to:

[0143] outputting a recognition result according to the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix;

[0144] If the target audio signal includes a voice signal of a keyword, the recognition result is an identifier of the keyword;

[0145] If the target audio signal does not include a speech signal, the recognition result is a preset identifier.

[0146] Optionally, the audio signal processing device 60 also includes a calculation module 67, which is used to calculate the confidence of the keyword based on the recognition result corresponding to the target audio signal and the recognition results corresponding to the one or more historical audio signals, where the keyword includes one or more keywords; if the confidence of the keyword is greater than a threshold, the electronic device is woken up.

[0147] Figure 6 The audio signal processing device of the illustrated embodiment can be used to execute the technical solution of the above-mentioned method embodiment. Its implementation principle and technical effects are similar and will not be described in detail here.

[0148] The internal functions and structure of the audio signal processing device are described above. The device can be implemented as an electronic device. Figure 7 This is a schematic diagram of the structure of an electronic device embodiment provided by the present disclosure. Figure 7 As shown, the electronic device includes a memory 71 and a processor 72 .

[0149] The memory 71 is used to store programs. In addition to the aforementioned programs, the memory 71 may also be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device, contact data, phone book data, messages, images, videos, and the like.

[0150] The memory 71 can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0151] The processor 72 is coupled to the memory 71 and executes the program stored in the memory 71 to:

[0152] Acquire target audio signal;

[0153] Processing the target audio signal according to the noise covariance matrix to obtain a noise-suppressed target audio signal;

[0154] Extracting features from the noise-suppressed target audio signal to obtain multiple feature vectors;

[0155] Combining the plurality of feature vectors into a single vector;

[0156] determining whether the target audio signal includes a speech signal according to the single vector;

[0157] If the target audio signal does not include a speech signal, the noise covariance matrix is ​​updated, and the updated noise covariance matrix is ​​used to process the next audio signal of the target audio signal.

[0158] Further, if Figure 7 As shown, the electronic device may further include: a communication component 73, a power component 74, an audio component 75, a display 76 and other components. Figure 7 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 7 Components shown.

[0159] The communication component 73 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 WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 73 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 73 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0160] The power supply assembly 74 provides power to various components of the electronic device. The power supply assembly 74 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.

[0161] The audio component 75 is configured to output and / or input audio signals. For example, the audio component 75 includes a microphone (MIC), which is configured to receive external audio signals when the electronic device is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 71 or transmitted via the communication component 73. In some embodiments, the audio component 75 also includes a speaker for outputting audio signals.

[0162] The display 76 includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). 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 touches, slides, 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.

[0163] It is understood that the above embodiment is an audio signal processing method introduced using a sweeping robot as an example. In other embodiments, the method can also be applied to a mobile platform, which can be a device operating in a relatively noisy environment, for example, a drone, a mobile robot, etc. The following describes an audio signal processing method for a mobile platform in a noisy environment, which specifically includes the following steps:

[0164] S1. Acquire an audio signal collected by the movable platform in the noisy environment.

[0165] S2. Process the audio signal according to the noise covariance matrix to obtain a noise-suppressed audio signal.

[0166] S3. Perform feature extraction on the noise-suppressed audio signal to obtain multiple feature vectors.

[0167] S4. Combining the multiple feature vectors into a single vector.

[0168] S5. Determine whether the audio signal includes a speech signal according to the single vector.

[0169] S6. If the audio signal does not include a speech signal, update the noise covariance matrix, and use the updated noise covariance matrix to process the next audio signal of the audio signal.

[0170] Specifically, the implementation process and specific principles of S1-S6 can refer to the relevant content in the embodiment of the audio signal processing method described in the above embodiment, and will not be repeated here.

[0171] In addition, the embodiment of the present disclosure also provides a mobile platform in a noisy environment, which includes a drone, a mobile robot, etc. The mobile platform in a noisy environment includes: a fuselage, a power system, an audio signal acquisition device, and an audio signal processing device; wherein the power system is installed in the fuselage to provide mobile power; the audio signal acquisition device is used to collect audio signals, and the audio signal acquisition device can be a microphone; the audio signal processing device is used to process the audio signals using the methods described in S1-S6, and the structural diagram of the audio signal processing device can be referred to. Figure 6 The structural diagram of the audio signal processing device is not described here in detail.

[0172] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by a processor to implement the audio signal processing method described in the above embodiment.

[0173] 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 actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0174] The foregoing description is intended only to provide specific embodiments of the present disclosure, intended to enable those skilled in the art to understand and implement the present disclosure. 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 the present disclosure. Therefore, the present disclosure is not intended to be limited to the embodiments described herein, but rather to be construed in the broadest manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for processing audio signals on a mobile platform in a noisy environment, wherein: The method comprises: Acquiring an audio signal collected by the movable platform in the noisy environment; Processing the audio signal according to a noise covariance matrix to obtain a noise-suppressed audio signal, wherein the noise is generated by the noisy environment movable platform itself; Performing feature extraction on the noise-suppressed audio signal to obtain a plurality of feature vectors; Combining the plurality of feature vectors into a single vector; determining whether the audio signal includes a speech signal according to the single vector; If the audio signal does not include the speech signal, updating the noise covariance matrix, and using the updated noise covariance matrix to process a next audio signal of the audio signal to suppress the noise generated by the noisy environment movable platform itself in the next audio signal; Wherein, determining whether the audio signal includes a speech signal based on the single vector includes: inputting the single vector into an acoustic model, the acoustic model includes multiple sub-models, the acoustic model outputs multiple first conditional probabilities, the multiple first conditional probabilities correspond to the multiple sub-models one-to-one, each first conditional probability is used to represent the probability of observing the single vector under the condition of any sub-model, and the multiple sub-models include models for detecting the speech signal and non-speech signal respectively; determining whether the audio signal includes the speech signal based on the multiple first conditional probabilities and one or more historical audio signals before the audio signal.

2. A mobile platform in a noisy environment, wherein: include: body; A power system, installed on the fuselage, for providing movement power; An audio signal acquisition device, used for acquiring audio signals; as well as An audio signal processing device, configured to process the audio signal using the method according to claim 1.

3. A method for processing an audio signal, wherein: The method comprises: Acquire target audio signal; Processing the target audio signal according to a noise covariance matrix to obtain a noise-suppressed target audio signal, wherein the noise is generated by an electronic device that collects the target audio signal; Extracting features from the noise-suppressed target audio signal to obtain multiple feature vectors; Combining the plurality of feature vectors into a single vector; determining whether the target audio signal includes a speech signal according to the single vector; If the target audio signal does not include the speech signal, updating the noise covariance matrix, where the updated noise covariance matrix is ​​used to process an audio signal next to the target audio signal to suppress the noise generated by the electronic device itself in the next audio signal; Wherein, determining whether the target audio signal includes a speech signal based on the single vector includes: inputting the single vector into an acoustic model, the acoustic model includes multiple sub-models, the acoustic model outputs multiple first conditional probabilities, the multiple first conditional probabilities correspond to the multiple sub-models one-to-one, each first conditional probability is used to represent the probability of observing the single vector under the condition of any sub-model, and the multiple sub-models include models for detecting the speech signal and non-speech signal respectively; determining whether the target audio signal includes the speech signal based on the multiple first conditional probabilities and one or more historical audio signals before the target audio signal.

4. The method according to claim 3, wherein: Determining whether the target audio signal includes the speech signal according to the multiple first conditional probabilities and one or more historical audio signals before the target audio signal includes: Inputting the multiple first conditional probabilities into a decoder, wherein the decoder updates a probability transfer matrix corresponding to the decoder according to the multiple first conditional probabilities; It is determined whether the target audio signal includes the speech signal according to the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix.

5. The method according to claim 4, wherein The decoder updates a probability transfer matrix corresponding to the decoder according to the multiple first conditional probabilities, including: The decoder calculates an updated forward probability based on the multiple first conditional probabilities, the historical forward probability and the historical probability transfer matrix; An updated probability transfer matrix is ​​calculated based on the updated forward probability and the historical probability transfer matrix.

6. The method according to claim 5, wherein: The updated forward probability includes a second conditional probability, wherein the second conditional probability is used to represent the probability of observing a non-speech signal under the condition of the single vector; Correspondingly, if the target audio signal does not include the speech signal, updating the noise covariance matrix includes: If the target audio signal does not include the speech signal, the noise covariance matrix is ​​updated according to the second conditional probability.

7. The method according to claim 4, wherein: Determining whether the target audio signal includes the speech signal according to the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix includes: outputting a recognition result according to the multiple first conditional probabilities, one or more historical audio signals before the target audio signal, and the updated probability transfer matrix; If the target audio signal includes the speech signal of the keyword, the recognition result is the identifier of the keyword; If the target audio signal does not include the voice signal, the recognition result is a preset identifier.

8. The method according to claim 7, wherein: The method further comprises: Calculating the confidence of a keyword based on the recognition result corresponding to the target audio signal and the recognition results corresponding to the one or more historical audio signals, where the keyword includes one or more keywords; If the confidence level of the keyword is greater than a threshold, the electronic device is woken up.

9. An audio signal processing device, wherein: include: An acquisition module, used to acquire a target audio signal; a processing module, configured to process the target audio signal according to a noise covariance matrix to obtain a noise-suppressed target audio signal, wherein the noise is generated by an electronic device that collects the target audio signal; A feature extraction module is used to extract features from the target audio signal after noise suppression to obtain multiple feature vectors; A synthesis module, configured to synthesize the plurality of feature vectors into a single vector; a determination module, configured to determine whether the target audio signal includes a speech signal according to the single vector; an updating module, configured to update the noise covariance matrix when the target audio signal does not include the speech signal, wherein the updated noise covariance matrix is ​​used to process an audio signal next to the target audio signal to suppress the noise generated by the electronic device itself in the next audio signal; In which, the determination module is also used to perform the following process to determine whether the target audio signal includes a speech signal based on the single vector: input the single vector into the acoustic model, the acoustic model includes multiple sub-models, and the acoustic model outputs multiple first conditional probabilities, the multiple first conditional probabilities and the multiple sub-models correspond one-to-one, each first conditional probability is used to represent the probability of observing the single vector under the condition of any sub-model, and the multiple sub-models include models for detecting the speech signal and non-speech signal respectively; determine whether the target audio signal includes the speech signal based on the multiple first conditional probabilities and one or more historical audio signals before the target audio signal.

10. An electronic device, wherein: include: Memory; processor; as well as computer programs; The computer program is stored in the memory and configured to be executed by the processor to implement the method according to any one of claims 1 and 3-8.

11. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 and 3 to 8 is implemented.

Images