Audio signal processing method, device and computer equipment

Abstract

The application relates to an audio signal processing method, device and computer equipment. The method comprises the following steps: acquiring a frequency domain signal corresponding to a time domain audio signal to be processed; inputting the frequency domain signal into an audio signal separation model; outputting a mask corresponding to a first sub-signal in the frequency domain signal through the audio signal separation model; wherein the audio signal separation model is obtained by training according to the corresponding relationship between a time domain audio signal sample to be processed and a processed time domain audio signal; determining a first sub-signal and a second sub-signal in the frequency domain signal according to the frequency domain signal and the mask; and determining a processed time domain audio signal according to the first sub-signal, the second sub-signal and a preset intensity ratio of the first sub-signal and the second sub-signal. The distortion problem caused by the audio signal separation model can be greatly reduced.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and in particular to an audio signal processing method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] During audio playback, the separation or clarity of the vocal signal is a key factor affecting the audio experience. Multi-channel audio sources, such as 5.1-channel and 7.1-channel, can significantly improve the listening experience of vocal signals. However, multi-channel audio sources are difficult to produce, and most audio sources available on the market are currently two-channel. One related technology uses frequency band filtering enhancement, which filters out the vocal frequency band in the audio and then enhances the vocal signal by increasing the gain of that frequency band. However, this method also proportionally enhances the accompaniment signal within the same frequency band, resulting in poor realism. Another related technology uses neural network models to separate the vocal signal in the audio, but the separated vocal signal will have some degree of distortion, reducing the audio quality. Therefore, how to accurately separate the vocal signal without damaging it is a technical problem that urgently needs to be solved. Summary of the Invention

[0003] Therefore, it is necessary to provide an audio signal processing method, apparatus, computer equipment, storage medium, and computer program product to address the aforementioned technical problems.

[0004] Firstly, this application provides an audio signal processing method. The method includes:

[0005] Obtain the frequency domain signal corresponding to the time-domain audio signal to be processed;

[0006] The frequency domain signal is input into the audio signal separation model, and the audio signal separation model outputs the mask corresponding to the first sub-signal in the frequency domain signal; wherein, the audio signal separation model is trained based on the correspondence between the time domain audio signal sample to be processed and the processed time domain audio signal;

[0007] Based on the frequency domain signal and the mask, a first sub-signal and a second sub-signal in the frequency domain signal are determined; wherein the first sub-signal and the second sub-signal have different timbres;

[0008] The processed time-domain audio signal is determined based on the first sub-signal, the second sub-signal, and the preset intensity ratio between the first sub-signal and the second sub-signal.

[0009] In one possible implementation, the time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal. The step of inputting the frequency-domain signal to an audio signal separation model, and outputting a mask corresponding to the first sub-signal in the frequency-domain signal via the audio signal separation model, includes:

[0010] The first frequency domain signal corresponding to the first time domain audio signal, the second frequency domain signal corresponding to the second time domain audio signal, and the difference signal between the first frequency domain signal and the second frequency domain signal are input into the audio signal separation model; wherein, the first time domain audio signal and the second time domain audio signal have different channels;

[0011] The audio signal separation model outputs the mask corresponding to the first sub-signal in the first frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal.

[0012] In one possible implementation, determining the first sub-signal and the second sub-signal in the frequency domain signal based on the frequency domain signal and the mask includes:

[0013] Based on the first frequency domain signal and the mask corresponding to the first sub-signal in the first frequency domain signal, determine the first sub-signal and the second sub-signal in the first frequency domain signal;

[0014] Based on the second frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal, determine the first sub-signal and the second sub-signal in the second frequency domain signal;

[0015] The step of obtaining the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and a preset intensity ratio between the first sub-signal and the second sub-signal includes:

[0016] Obtain the updated first preset intensity ratio and second preset intensity ratio;

[0017] A first mixed-audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio; and a second mixed-audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio.

[0018] The first mixed audio domain signal and the second mixed audio domain signal are respectively converted from the frequency domain to the time domain to obtain the processed first time domain audio signal and second time domain audio signal.

[0019] In one possible implementation, the frequency domain signal includes a complex spectrum signal, and the step of acquiring the frequency domain signal corresponding to the time-domain audio signal to be processed includes:

[0020] Acquire the time-domain audio signal to be processed;

[0021] The time-domain audio signal is subjected to Z-transform to obtain a complex spectrum signal.

[0022] In one possible implementation, the audio signal separation model is obtained by:

[0023] A set of time-domain audio signal samples to be processed is obtained, the set including the time-domain audio signal samples to be processed, and the processed time-domain audio signal samples determined according to the first time-domain sub-signal sample, the second time-domain sub-signal sample and the preset intensity ratio in the time-domain audio signal samples;

[0024] The frequency domain signal sample corresponding to the time domain audio signal sample is input into the initial audio signal separation model to predict the mask corresponding to the first sub-signal;

[0025] Based on the frequency domain signal sample and the mask, determine the first sub-signal and the second sub-signal in the frequency domain signal sample;

[0026] The processed time-domain audio signal is determined based on the first sub-signal, the second sub-signal, and the preset intensity ratio.

[0027] Based on the difference between the time-domain audio signal and the processed time-domain audio signal sample, the training parameters are iteratively adjusted until the difference meets the preset requirements, thus obtaining the audio signal separation model.

[0028] In one possible implementation, the initial audio signal separation model includes a first recurrent neural network, an attention mechanism network, and a second recurrent neural network. The step of inputting the frequency domain signal sample corresponding to the time-domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal includes:

[0029] The frequency domain signal corresponding to the current frame in the time-domain audio signal sample is input into the first recurrent neural network, and the feature signal is output.

[0030] The attention mechanism network is used to filter the feature signals to obtain filtered feature signals; wherein, the attention mechanism network caches frequency domain signals of a preset number of frames prior to the current frame;

[0031] The filtered feature signals are input into the second recurrent neural network, which outputs the mask corresponding to the first sub-signal.

[0032] Secondly, this application also provides an audio signal processing apparatus. The apparatus includes:

[0033] The first acquisition module is used to acquire the frequency domain signal corresponding to the time domain audio signal to be processed;

[0034] A separation module is used to input the frequency domain signal into an audio signal separation model, and output a mask corresponding to the first sub-signal in the frequency domain signal through the audio signal separation model; wherein, the audio signal separation model is trained based on the correspondence between the time domain audio signal samples to be processed and the processed time domain audio signal;

[0035] The first determining module is used to determine a first sub-signal and a second sub-signal in the frequency domain signal based on the frequency domain signal and the mask; wherein the first sub-signal and the second sub-signal have different timbres;

[0036] The second determining module is used to determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and a preset intensity ratio between the first sub-signal and the second sub-signal.

[0037] In one possible implementation, the time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal, and the separation module includes:

[0038] The first input submodule is used to input the first frequency domain signal corresponding to the first time domain audio signal, the second frequency domain signal corresponding to the second time domain audio signal, and the differential signal between the first frequency domain signal and the second frequency domain signal into the audio signal separation model; wherein the first time domain audio signal and the second time domain audio signal have different channels;

[0039] The first output submodule is used to output, via the audio signal separation model, a mask corresponding to the first sub-signal in the first frequency domain signal and a mask corresponding to the first sub-signal in the second frequency domain signal.

[0040] In one possible implementation, the first determining module includes:

[0041] The first determining submodule is used to determine the first sub-signal and the second sub-signal in the first frequency domain signal based on the first frequency domain signal and the mask corresponding to the first sub-signal in the first frequency domain signal.

[0042] The second determining submodule is used to determine the first sub-signal and the second sub-signal in the second frequency domain signal based on the second frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal.

[0043] The second determining module includes:

[0044] The acquisition submodule is used to acquire the updated first preset intensity ratio and second preset intensity ratio;

[0045] The mixing submodule is used to mix the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio to obtain a first mixed audio domain signal; and to mix the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio to obtain a second mixed audio domain signal.

[0046] The first conversion submodule is used to perform frequency domain to time domain conversion processing on the first mixed audio domain signal and the second mixed audio domain signal respectively, to obtain the processed first time domain audio signal and second time domain audio signal.

[0047] In one possible implementation, the first acquisition module includes:

[0048] The acquisition submodule is used to acquire the time-domain audio signal to be processed;

[0049] The second conversion submodule is used to perform Z-transform on the time-domain audio signal to obtain a complex spectrum signal.

[0050] One possible implementation also includes:

[0051] The second acquisition module is used to acquire a set of time-domain audio signal samples to be processed. The set includes the time-domain audio signal samples to be processed, and the processed time-domain audio signal samples determined according to the first time-domain sub-signal sample, the second time-domain sub-signal sample and the preset intensity ratio in the time-domain audio signal samples.

[0052] The prediction module is used to input the frequency domain signal sample corresponding to the time domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal;

[0053] The third determining module is used to determine the first sub-signal and the second sub-signal in the frequency domain signal sample based on the frequency domain signal sample and the mask;

[0054] The fourth determining module is used to determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and the preset intensity ratio.

[0055] The generation module is used to iteratively adjust the training parameters based on the difference between the time-domain audio signal and the processed time-domain audio signal sample until the difference meets the preset requirements, thereby obtaining an audio signal separation model.

[0056] In one possible implementation, the initial audio signal separation model includes a first recurrent neural network, an attention mechanism network, and a second recurrent neural network, and the prediction module includes:

[0057] The second input submodule is used to input the frequency domain signal corresponding to the current frame in the time domain audio signal sample into the first recurrent neural network and output the feature signal.

[0058] The filtering submodule is used to filter the feature signals using the attention mechanism network to obtain filtered feature signals; wherein, the attention mechanism network caches frequency domain signals of a preset number of frames prior to the current frame;

[0059] The second output submodule is used to input the filtered feature signal into the second recurrent neural network and output the mask corresponding to the first sub-signal.

[0060] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the audio signal processing method described in any one of the embodiments of this disclosure.

[0061] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the audio signal processing method described in any one of the embodiments of this disclosure.

[0062] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the audio signal processing method described in any one of the embodiments of this disclosure.

[0063] The aforementioned audio signal processing methods, apparatus, computer equipment, storage media, and computer program products, in which the audio signal processing method uses an audio signal separation model trained on the correspondence between the time-domain audio signal samples to be processed and the processed time-domain audio signal, wherein both the time-domain audio signal to be processed and the processed time-domain audio signal belong to the time domain and both contain two or more timbre sub-signals. This scheme utilizes the time-domain audio signal samples to be processed and the processed time-domain audio signal to construct a loss function. Compared to the traditional model that uses the separated single-timbre signal and single-timbre signal samples to construct the loss function, this significantly reduces the distortion problem caused by the audio signal separation model. It improves the accuracy of the audio signal separation model and, by changing the preset intensity ratio of the first and second sub-signals, enhances the signal volume of the target sub-signal. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the first flow of an audio signal processing method in one embodiment;

[0065] Figure 2This is a schematic diagram of the second flow of an audio signal processing method in another embodiment;

[0066] Figure 3 This is a structural diagram of an audio signal separation model in an audio signal processing method of one embodiment;

[0067] Figure 4 This is a schematic diagram of the third process of an audio signal processing method in one embodiment;

[0068] Figure 5 This is a structural block diagram of an audio signal processing device in one embodiment;

[0069] Figure 6 This is an internal structural diagram of a computer device in one embodiment;

[0070] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0072] The audio signal processing method provided in this application can be applied to a terminal or a server. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0073] In one possible implementation, such as Figure 1 As shown, an audio signal processing method is provided, including the following steps:

[0074] Step S101: Obtain the frequency domain signal corresponding to the time domain audio signal to be processed.

[0075] Specifically, the time-domain audio signal can include audio signals over a period of time, such as a song containing vocals and accompaniment, a mixed audio of multiple instruments including cello, violin, and piano, or a recording containing the desired vocal content as well as some noise interference. The time-domain audio signal to be processed can include a mono, stereo, or multi-channel audio signal. The frequency-domain signal can be a frequency-domain signal obtained by converting the time-domain audio signal from the time domain to the frequency domain, such as a Laplace transform, Fourier transform, or Z-transform. The frequency-domain signal is used to describe the properties of the signal as a function of frequency.

[0076] Step S103: Input the frequency domain signal into the audio signal separation model, and output the mask corresponding to the first sub-signal in the frequency domain signal through the audio signal separation model; wherein, the audio signal separation model is trained based on the correspondence between the time domain audio signal sample to be processed and the processed time domain audio signal.

[0077] Specifically, the frequency domain signal may contain two or more sub-signals. The first sub-signal can describe a signal for which the volume is desired to be increased, or a signal for which the volume is desired to be decreased. When the first signal is a signal for which the volume is desired to be increased, the second signal is a signal for which the volume is desired to be decreased; conversely, when the first signal is a signal for which the volume is desired to be decreased, the second signal is a signal for which the volume is desired to be increased. For example, in a song, if a user wants the vocals to be increased, the vocals can be used as the first sub-signal; similarly, in a song, if a user wants the accompaniment to be decreased, the accompaniment can also be used as the first sub-signal. Therefore, it should be noted that the first sub-signal can contain a single type of signal or multiple types of signals.

[0078] In this embodiment, the mask, also known as a code, is used in deep learning to essentially cover the original input data with a mask, thereby shielding or selecting certain elements. In this embodiment, the mask corresponding to the first sub-signal may include setting the first sub-signal to 1 and the second sub-signal to 0 in the frequency domain signal, so that the first sub-signal can be filtered out by multiplying the mask and the frequency domain signal.

[0079] In this embodiment, the audio signal separation model is trained based on the correspondence between the time-domain audio signal sample to be processed and the processed time-domain audio signal sample. Specifically, the audio signal separation model may include an artificial intelligence neural network model based on deep learning. The time-domain audio signal sample to be processed may include a mixed audio signal sample of two or more signal types, each mixed audio signal sample labeled with an original first sub-signal sample and a second sub-signal sample. The processed time-domain audio signal sample may include the first sub-signal and the second sub-signal obtained by inputting the aforementioned time-domain audio signal sample to be processed into the audio signal separation model, mixing the first and second sub-signals according to a preset intensity ratio to obtain a mixed signal, and converting the mixed signal into a time-domain audio signal. The processed time-domain audio signal is then obtained. The original first sub-signal sample and the second sub-signal sample mixed at the same intensity ratio are used as the ground truth. A loss function is constructed using the processed time-domain audio signal and the ground truth, and the model is iteratively tuned based on the loss function.

[0080] Step S105: Determine the first sub-signal and the second sub-signal in the frequency domain signal based on the frequency domain signal and the mask; wherein the first sub-signal and the second sub-signal have different timbres.

[0081] Specifically, the frequency domain signal includes a first sub-signal and a second sub-signal, wherein the first sub-signal and the second sub-signal have different timbres. For example, the first sub-signal may include a human voice signal, and the second sub-signal may include an accompaniment signal or a noise signal. As another example, the first sub-signal may include a specific instrument in an audio segment, such as a violin, and the second sub-signal may include other sub-signals besides the violin. In this embodiment of the disclosure, the first sub-signal in the frequency domain signal can be obtained by multiplying the frequency domain signal by a mask of the first sub-signal; the second signal can be obtained by filtering out the first sub-signal from the frequency domain signal.

[0082] Step S107: Determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and the preset intensity ratio between the first sub-signal and the second sub-signal.

[0083] Specifically, based on the first sub-signal and the second sub-signal, and a preset intensity ratio between the first and second sub-signals, such as 5:3, the two sub-signals are superimposed to obtain a mixed frequency domain signal. The mixed frequency domain signal is then converted from the frequency domain to the time domain to obtain the corresponding processed time domain audio signal. Therefore, in this embodiment, by adjusting the preset intensity ratio, a sub-signal whose volume is desired to be increased can be enhanced, or a sub-signal whose volume is desired to be decreased can be reduced.

[0084] In the aforementioned audio signal processing method, the audio signal separation model is trained using the correspondence between the time-domain audio signal samples to be processed and the processed time-domain audio signal. Both the time-domain audio signal to be processed and the processed time-domain audio signal belong to the time domain and contain two or more timbre sub-signals. This scheme uses the time-domain audio signal samples to be processed and the processed time-domain audio signal to construct a loss function. Compared to the traditional model that uses the separated single-timbre signal and single-timbre signal samples to construct the loss function, this significantly reduces the distortion problem caused by the audio signal separation model. This improves the accuracy of the audio signal separation model. Furthermore, by changing the preset intensity ratio of the first and second sub-signals, the signal volume of the target sub-signal is enhanced.

[0085] In one possible implementation, the time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal. The step of inputting the frequency-domain signal to an audio signal separation model, and outputting a mask corresponding to the first sub-signal in the frequency-domain signal via the audio signal separation model, includes:

[0086] The first frequency domain signal corresponding to the first time domain audio signal, the second frequency domain signal corresponding to the second time domain audio signal, and the difference signal between the first frequency domain signal and the second frequency domain signal are input into the audio signal separation model; wherein, the first time domain audio signal and the second time domain audio signal have different channels;

[0087] The audio signal separation model outputs the mask corresponding to the first sub-signal in the first frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal.

[0088] Specifically, the time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal. The first time-domain audio signal and the second time-domain audio signal have different channels. For example, the first time-domain audio signal may include a left channel audio signal, and the second time-domain audio signal may include a right channel audio signal. The channels include independent audio signals collected or played back from different spatial locations during recording or playback.

[0089] In this embodiment of the disclosure, the first frequency domain signal corresponding to the first time-domain audio signal may include a first frequency domain signal obtained by converting the first time-domain audio signal from the time domain to the frequency domain. The second frequency domain signal corresponding to the second time-domain audio signal may include a second frequency domain signal obtained by converting the second time-domain audio signal from the time domain to the frequency domain. The difference signal between the first frequency domain signal and the second frequency domain signal may include the signal obtained by subtracting the first frequency domain signal from the second frequency domain signal.

[0090] In this embodiment of the disclosure, the first frequency domain signal, the second frequency domain signal, and the differential signal are input to the audio signal separation model, which can output the mask corresponding to the first sub-signal in the first frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal, respectively.

[0091] In this embodiment, the number of time-domain audio signal channels to be processed as input to the audio signal separation model is not limited. When the number of channels is a common sound source channel, such as the left and right channels, the difference signal between the two channels is used as another input signal to the audio signal separation model. Since the difference signal obtained by subtracting the two channel signals is more likely to retain a portion of the difference signal in the two channels (for example, it is easier to delete the close-up human voice signal in the left and right channels and retain the difference accompaniment signal), it helps the audio signal separation model learn the difference signal, thereby giving a more accurate mask prediction.

[0092] In one possible implementation, determining the first sub-signal and the second sub-signal in the frequency domain signal based on the frequency domain signal and the mask includes:

[0093] Based on the first frequency domain signal and the mask corresponding to the first sub-signal in the first frequency domain signal, determine the first sub-signal and the second sub-signal in the first frequency domain signal;

[0094] Based on the second frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal, determine the first sub-signal and the second sub-signal in the second frequency domain signal;

[0095] The step of obtaining the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and a preset intensity ratio between the first sub-signal and the second sub-signal includes:

[0096] Obtain the updated first preset intensity ratio and second preset intensity ratio;

[0097] A first mixed-audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio; and a second mixed-audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio.

[0098] The first mixed audio domain signal and the second mixed audio domain signal are respectively converted from the frequency domain to the time domain to obtain the processed first time domain audio signal and second time domain audio signal.

[0099] Specifically, the first sub-signal in the first frequency domain signal can be obtained by multiplying the first frequency domain signal with the mask corresponding to the first sub-signal. In an exemplary embodiment, the first sub-signal is subtracted from the first frequency domain signal to obtain the second sub-signal. Correspondingly, the first sub-signal in the second frequency domain signal can be obtained by multiplying the second frequency domain signal with the mask corresponding to the first sub-signal. In an exemplary embodiment, the second sub-signal is subtracted from the second frequency domain signal to obtain the second sub-signal.

[0100] In this embodiment of the disclosure, the updated first preset intensity ratio is different from the original intensity ratio of the first sub-signal and the second sub-signal in the first time-domain audio signal. For example, in the unprocessed left channel audio, the intensity ratio of the vocals to the accompaniment is 5:5. If the user wants to increase the volume of the vocals in the left channel, the updated first preset intensity ratio can be changed to 7:3. Similarly, the updated second preset intensity ratio is different from the original intensity ratio of the first sub-signal and the second sub-signal in the second time-domain audio signal.

[0101] In this embodiment of the present disclosure, a first mixed audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio; and a second mixed audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio; the first mixed audio domain signal and the second mixed audio domain signal are respectively converted from the frequency domain to the time domain to obtain the processed first time domain audio signal and the second time domain audio signal.

[0102] In the above embodiments, by changing the first preset intensity ratio and the second preset intensity ratio, the magnitudes of the first sub-signal and the second sub-signal in the first time-domain audio signal can be changed, and similarly, the magnitudes of the first sub-signal and the second sub-signal in the second time-domain audio signal can also be changed, thereby changing the position of the sound image of the first sub-signal.

[0103] In one possible implementation, the frequency domain signal includes a complex spectrum signal, and the step of acquiring the frequency domain signal corresponding to the time-domain audio signal to be processed includes:

[0104] Acquire the time-domain audio signal to be processed;

[0105] The time-domain audio signal is subjected to Z-transform to obtain a complex spectrum signal.

[0106] In this embodiment of the disclosure, the frequency domain signal includes a complex spectrum signal. A time-domain audio signal to be processed is acquired; a Z-transform is performed on the time-domain audio signal to obtain a complex spectrum signal. In an exemplary embodiment, when the time-domain audio signal to be processed includes time-domain audio signals from multiple channels, a Z-transform can be performed on the time-domain audio signal of each channel separately to obtain a complex spectrum signal.

[0107] In this embodiment, a complex spectrum signal is used as the modeling signal for the audio signal separation model and as the input signal when applying the model. The complex spectrum signal can simultaneously constrain the amplitude and phase in the spectrum, which can reduce phase distortion during signal reconstruction. Therefore, this embodiment can improve the accuracy of the audio signal separation model.

[0108] In one possible implementation, refer to Figure 2 As shown, the audio signal separation model is obtained in the following ways:

[0109] Step S201: Obtain a set of time-domain audio signal samples to be processed. The set includes the time-domain audio signal samples to be processed and the processed time-domain audio signal samples determined according to the first time-domain sub-signal sample, the second time-domain sub-signal sample and the preset intensity ratio in the time-domain audio signal samples.

[0110] Specifically, the time-domain audio signal sample includes a mixed first time-domain sub-signal and a second time-domain sub-signal. The first and second time-domain sub-signal samples in the time-domain audio signal sample are original signal samples ideally separated through other methods. In an exemplary embodiment, the time-domain audio signal sample may include first and second time-domain audio signal samples from different channels. Multiple new first time-domain sub-signal samples with different gains can be obtained by sampling the first time-domain sub-signal samples within a preset gain range. For example, uniform sampling with a step size of 0.1 within the [0,1] range can yield first time-domain sub-signal samples with different gain values ​​G. In another exemplary embodiment, the first time-domain sub-signal samples in the second time-domain audio signal sample can be selected with a gain value of 1-G, thus maintaining the overall volume of the first time-domain sub-signal samples in the original time-domain audio signal sample. In an exemplary embodiment, different intensity ratios can be set to mix the first and second time-domain sub-signal samples into multiple processed time-domain audio signal samples.

[0111] Step S203: Input the frequency domain signal sample corresponding to the time domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal.

[0112] Specifically, the frequency domain signal sample corresponding to the time-domain audio signal sample may include the frequency domain signal sample obtained by performing a time-domain to frequency-domain conversion on the time-domain audio signal sample. The initial audio signal separation model may include various network models, such as recurrent neural networks, variant networks of recurrent neural networks, or a combination of multiple networks. Among them, variant networks of recurrent neural networks may include LSTM networks, Bi-LSTM networks, GRU networks, etc.

[0113] In this embodiment of the disclosure, frequency domain signal samples are input into an initial audio signal separation model, and a mask corresponding to the first sub-signal is output.

[0114] Step S205: Determine the first sub-signal and the second sub-signal in the frequency domain signal sample based on the frequency domain signal sample and the mask.

[0115] Specifically, the first sub-signal can be obtained by multiplying the frequency domain signal sample with the mask corresponding to the first sub-signal. The second sub-signal is then obtained by filtering out the first sub-signal from the frequency domain signal sample.

[0116] Step S207: Determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and the preset intensity ratio.

[0117] Specifically, according to the preset intensity ratio, the first sub-signal and the second sub-signal are mixed to obtain a mixed frequency domain audio signal. The mixed frequency domain audio signal is then converted from the frequency domain to the time domain to obtain a time domain audio signal.

[0118] Step S209: Based on the difference between the time-domain audio signal and the processed time-domain audio signal sample, the training parameters are iteratively adjusted until the difference meets the preset requirements, thereby obtaining the audio signal separation model.

[0119] Based on the difference between the time-domain audio signal and the processed time-domain audio signal sample with the same intensity ratio, the training parameters are iteratively adjusted until the number of iterations meets the preset requirements, or the prediction result meets the preset requirements, to obtain the audio signal separation model.

[0120] This scheme constructs a loss function using the time-domain audio signal samples to be processed and the processed time-domain audio signal. Compared to the traditional model that uses the separated monophonic signals and monophonic signal samples to construct the loss function, this significantly reduces the distortion problem caused by the audio signal separation model and improves the accuracy of the audio signal separation model.

[0121] In one possible implementation, the initial audio signal separation model includes a first recurrent neural network, an attention mechanism network, and a second recurrent neural network. The step of inputting the frequency domain signal sample corresponding to the time-domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal includes:

[0122] The frequency domain signal corresponding to the current frame in the time-domain audio signal sample is input into the first recurrent neural network, and the feature signal is output.

[0123] The attention mechanism network is used to filter the feature signals to obtain filtered feature signals; wherein, the attention mechanism network caches frequency domain signals of a preset number of frames prior to the current frame;

[0124] The filtered feature signals are input into the second recurrent neural network, which outputs the mask corresponding to the first sub-signal.

[0125] For details, please refer to Figure 3 As shown, the first recurrent neural network 301 may include a GRU (Gate Recurrent Unit) network, and the second recurrent neural network 303 may also include a GRU network. The frequency domain signal (e.g., the complex spectrum of the left channel, the complex spectrum of the right channel, and the complex spectrum of the difference signal between the two complex spectra) corresponding to the current frame t of the time-domain audio signal sample is input to the first recurrent neural network 301 after passing through a fully connected layer. The first recurrent neural network 301 is used for feature extraction and analysis of the frequency domain signal. In an exemplary embodiment, the attention mechanism network buffers the frequency domain signal of a preset number of frames prior to the current frame. In this embodiment, one frame of audio signal represents the audio signal within a preset time period. The attention mechanism network can obtain useful information from the frequency domain signal of historical frames, thereby filtering the features of the first recurrent neural network 301. The filtered features are input into the second recurrent neural network 303, where the second recurrent neural network is used to synthesize the filtered features. After passing through two fully connected layers, it outputs the vocal mask corresponding to the left channel and the vocal mask corresponding to the right channel.

[0126] In this embodiment, the recurrent neural network selected is a GRU network, which can effectively reduce the memory burden of the algorithm itself when modeling temporal information. The attention control strategy through caching and convolution further enhances the receptive field range of feature analysis. This results in a relatively lightweight and accurate audio signal separation model.

[0127] In one possible implementation, Figure 4 This is a schematic diagram of the third process of an audio signal processing method in one embodiment, with reference to... Figure 4As shown, the audio signal separation model inputs the first frequency domain signal (left channel input signal), the second frequency domain signal (right channel input signal), and the differential signal between the first and second frequency domain signals. It outputs the mask corresponding to the first sub-signal (voice signal) in the first frequency domain signal and the mask corresponding to the first sub-signal (voice signal) in the second frequency domain signal, respectively. Multiplying the mask corresponding to the left channel voice signal and the first frequency domain signal yields the voice signal in the left channel input signal. Subtracting the voice signal from the left channel input signal yields the accompaniment signal. Similarly, the voice signal and accompaniment signal in the right channel input signal can be obtained. Based on the intensity ratio of the left channel voice signal to the accompaniment signal (L - voice-accompaniment ratio), the separated voice signal and accompaniment signal are mixed to obtain the left channel output signal. Based on the intensity ratio of the right channel vocal signal to the accompaniment signal (R - vocal-accompaniment ratio), the vocal signal and accompaniment signal obtained above are mixed to obtain the right channel output signal.

[0128] Through the embodiments of this disclosure, the human voice signal and accompaniment in the audio signal can be separated, and the position of the human voice can be changed according to the preset intensity ratio of the left and right channels.

[0129] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0130] Based on the same inventive concept, this application also provides an audio signal processing apparatus for implementing the audio signal processing method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more audio signal processing apparatus embodiments provided below can be found in the limitations of the audio signal processing method described above, and will not be repeated here.

[0131] In one embodiment, Figure 5 Here is a structural block diagram of an audio signal processing device in one embodiment, such as... Figure 5 As shown, it includes:

[0132] The first acquisition module 501 is used to acquire the frequency domain signal corresponding to the time domain audio signal to be processed;

[0133] The separation module 503 is used to input the frequency domain signal into the audio signal separation model, and output the mask corresponding to the first sub-signal in the frequency domain signal through the audio signal separation model; wherein, the audio signal separation model is trained based on the correspondence between the time domain audio signal sample to be processed and the processed time domain audio signal;

[0134] The first determining module 505 is used to determine a first sub-signal and a second sub-signal in the frequency domain signal based on the frequency domain signal and the mask; wherein the first sub-signal and the second sub-signal have different timbres;

[0135] The second determining module 507 is used to determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and a preset intensity ratio between the first sub-signal and the second sub-signal.

[0136] In one possible implementation, the time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal, and the separation module includes:

[0137] The first input submodule is used to input the first frequency domain signal corresponding to the first time domain audio signal, the second frequency domain signal corresponding to the second time domain audio signal, and the differential signal between the first frequency domain signal and the second frequency domain signal into the audio signal separation model; wherein the first time domain audio signal and the second time domain audio signal have different channels;

[0138] The first output submodule is used to output, via the audio signal separation model, a mask corresponding to the first sub-signal in the first frequency domain signal and a mask corresponding to the first sub-signal in the second frequency domain signal.

[0139] In one possible implementation, the first determining module includes:

[0140] The first determining submodule is used to determine the first sub-signal and the second sub-signal in the first frequency domain signal based on the first frequency domain signal and the mask corresponding to the first sub-signal in the first frequency domain signal.

[0141] The second determining submodule is used to determine the first sub-signal and the second sub-signal in the second frequency domain signal based on the second frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal.

[0142] The second determining module includes:

[0143] The acquisition submodule is used to acquire the updated first preset intensity ratio and second preset intensity ratio;

[0144] The mixing submodule is used to mix the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio to obtain a first mixed audio domain signal; and to mix the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio to obtain a second mixed audio domain signal.

[0145] The first conversion submodule is used to perform frequency domain to time domain conversion processing on the first mixed audio domain signal and the second mixed audio domain signal respectively, to obtain the processed first time domain audio signal and second time domain audio signal.

[0146] In one possible implementation, the first acquisition module includes:

[0147] The acquisition submodule is used to acquire the time-domain audio signal to be processed;

[0148] The second conversion submodule is used to perform Z-transform on the time-domain audio signal to obtain a complex spectrum signal.

[0149] One possible implementation also includes:

[0150] The second acquisition module is used to acquire a set of time-domain audio signal samples to be processed. The set includes the time-domain audio signal samples to be processed, and the processed time-domain audio signal samples determined according to the first time-domain sub-signal sample, the second time-domain sub-signal sample and the preset intensity ratio in the time-domain audio signal samples.

[0151] The prediction module is used to input the frequency domain signal sample corresponding to the time domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal;

[0152] The third determining module is used to determine the first sub-signal and the second sub-signal in the frequency domain signal sample based on the frequency domain signal sample and the mask;

[0153] The fourth determining module is used to determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and the preset intensity ratio.

[0154] The generation module is used to iteratively adjust the training parameters based on the difference between the time-domain audio signal and the processed time-domain audio signal sample until the difference meets the preset requirements, thereby obtaining an audio signal separation model.

[0155] In one possible implementation, the initial audio signal separation model includes a first recurrent neural network, an attention mechanism network, and a second recurrent neural network, and the prediction module includes:

[0156] The second input submodule is used to input the frequency domain signal corresponding to the current frame in the time domain audio signal sample into the first recurrent neural network and output the feature signal.

[0157] The filtering submodule is used to filter the feature signals using the attention mechanism network to obtain filtered feature signals; wherein, the attention mechanism network caches frequency domain signals of a preset number of frames prior to the current frame;

[0158] The second output submodule is used to input the filtered feature signal into the second recurrent neural network and output the mask corresponding to the first sub-signal.

[0159] Each module in the aforementioned audio signal processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0160] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores audio signal processing data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements an audio signal processing method.

[0161] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements an audio signal processing method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0162] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0163] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0164] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An audio signal processing method, characterized in that, include: Obtain the frequency domain signal corresponding to the time-domain audio signal to be processed; The frequency domain signal is input into the audio signal separation model, and the audio signal separation model outputs the mask corresponding to the first sub-signal in the frequency domain signal; wherein, the audio signal separation model is trained based on the correspondence between the time domain audio signal sample to be processed and the processed time domain audio signal; Based on the frequency domain signal and the mask, a first sub-signal and a second sub-signal in the frequency domain signal are determined; wherein the first sub-signal and the second sub-signal have different timbres; The processed time-domain audio signal is determined based on the first sub-signal, the second sub-signal, and the preset intensity ratio between the first sub-signal and the second sub-signal. The audio signal separation model is obtained through the following methods: A set of time-domain audio signal samples to be processed is obtained, the set including the time-domain audio signal samples to be processed, and the processed time-domain audio signal samples determined according to the first time-domain sub-signal sample, the second time-domain sub-signal sample and the preset intensity ratio in the time-domain audio signal samples; The frequency domain signal sample corresponding to the time domain audio signal sample is input into the initial audio signal separation model to predict the mask corresponding to the first sub-signal; Based on the frequency domain signal sample and the mask, determine the first sub-signal and the second sub-signal in the frequency domain signal sample; The processed time-domain audio signal is determined based on the first sub-signal, the second sub-signal, and the preset intensity ratio. Based on the difference between the time-domain audio signal and the processed time-domain audio signal sample, the training parameters are iteratively adjusted until the difference meets the preset requirements, thus obtaining the audio signal separation model.

2. The method according to claim 1, characterized in that, The time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal. The step of inputting the frequency-domain signal into an audio signal separation model, and outputting a mask corresponding to the first sub-signal in the frequency-domain signal through the audio signal separation model, includes: The first frequency domain signal corresponding to the first time domain audio signal, the second frequency domain signal corresponding to the second time domain audio signal, and the difference signal between the first frequency domain signal and the second frequency domain signal are input into the audio signal separation model; wherein, the first time domain audio signal and the second time domain audio signal have different channels; The audio signal separation model outputs the mask corresponding to the first sub-signal in the first frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal.

3. The method according to claim 2, characterized in that, Determining the first sub-signal and the second sub-signal in the frequency domain signal based on the frequency domain signal and the mask includes: Based on the first frequency domain signal and the mask corresponding to the first sub-signal in the first frequency domain signal, determine the first sub-signal and the second sub-signal in the first frequency domain signal; Based on the second frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal, determine the first sub-signal and the second sub-signal in the second frequency domain signal; The step of obtaining the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and a preset intensity ratio between the first sub-signal and the second sub-signal includes: Obtain the updated first preset intensity ratio and second preset intensity ratio; A first mixed-audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio; and a second mixed-audio domain signal is obtained by mixing the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio. The first mixed audio domain signal and the second mixed audio domain signal are respectively converted from the frequency domain to the time domain to obtain the processed first time domain audio signal and second time domain audio signal.

4. The method according to claim 1, characterized in that, The frequency domain signal includes a complex spectrum signal, and the step of obtaining the frequency domain signal corresponding to the time-domain audio signal to be processed includes: Acquire the time-domain audio signal to be processed; The time-domain audio signal is subjected to Z-transform to obtain a complex spectrum signal.

5. The method according to claim 1, characterized in that, The initial audio signal separation model includes a first recurrent neural network, an attention mechanism network, and a second recurrent neural network. The step of inputting the frequency domain signal sample corresponding to the time-domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal includes: The frequency domain signal corresponding to the current frame in the time-domain audio signal sample is input into the first recurrent neural network, and the feature signal is output. The attention mechanism network is used to filter the feature signals to obtain filtered feature signals; wherein, the attention mechanism network caches frequency domain signals of a preset number of frames prior to the current frame; The filtered feature signals are input into the second recurrent neural network, which outputs the mask corresponding to the first sub-signal.

6. An audio signal processing device, characterized in that, The device includes: The first acquisition module is used to acquire the frequency domain signal corresponding to the time domain audio signal to be processed; A separation module is used to input the frequency domain signal into an audio signal separation model, and output a mask corresponding to the first sub-signal in the frequency domain signal through the audio signal separation model; wherein, the audio signal separation model is trained based on the correspondence between the time domain audio signal samples to be processed and the processed time domain audio signal; The first determining module is used to determine a first sub-signal and a second sub-signal in the frequency domain signal based on the frequency domain signal and the mask; wherein the first sub-signal and the second sub-signal have different timbres; The second determining module is used to determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and a preset intensity ratio between the first sub-signal and the second sub-signal. The second acquisition module is used to acquire a set of time-domain audio signal samples to be processed. The set includes the time-domain audio signal samples to be processed, and the processed time-domain audio signal samples determined according to the first time-domain sub-signal sample, the second time-domain sub-signal sample and the preset intensity ratio in the time-domain audio signal samples. The prediction module is used to input the frequency domain signal sample corresponding to the time domain audio signal sample into the initial audio signal separation model to predict the mask corresponding to the first sub-signal; The third determining module is used to determine the first sub-signal and the second sub-signal in the frequency domain signal sample based on the frequency domain signal sample and the mask; The fourth determining module is used to determine the processed time-domain audio signal based on the first sub-signal, the second sub-signal, and the preset intensity ratio. The generation module is used to iteratively adjust the training parameters based on the difference between the time-domain audio signal and the processed time-domain audio signal sample until the difference meets the preset requirements, thereby obtaining an audio signal separation model.

7. The apparatus according to claim 6, characterized in that, The time-domain audio signal to be processed includes a first time-domain audio signal and a second time-domain audio signal, and the separation module includes: The first input submodule is used to input the first frequency domain signal corresponding to the first time domain audio signal, the second frequency domain signal corresponding to the second time domain audio signal, and the differential signal between the first frequency domain signal and the second frequency domain signal into the audio signal separation model; wherein the first time domain audio signal and the second time domain audio signal have different channels; The first output submodule is used to output, via the audio signal separation model, a mask corresponding to the first sub-signal in the first frequency domain signal and a mask corresponding to the first sub-signal in the second frequency domain signal.

8. The apparatus according to claim 7, characterized in that, The first determining module includes: The first determining submodule is used to determine the first sub-signal and the second sub-signal in the first frequency domain signal based on the first frequency domain signal and the mask corresponding to the first sub-signal in the first frequency domain signal. The second determining submodule is used to determine the first sub-signal and the second sub-signal in the second frequency domain signal based on the second frequency domain signal and the mask corresponding to the first sub-signal in the second frequency domain signal. The second determining module includes: The acquisition submodule is used to acquire the updated first preset intensity ratio and second preset intensity ratio; The mixing submodule is used to mix the first sub-signal and the second sub-signal in the first frequency domain signal according to the first preset intensity ratio to obtain a first mixed audio domain signal; and to mix the first sub-signal and the second sub-signal in the second frequency domain signal according to the second preset intensity ratio to obtain a second mixed audio domain signal. The first conversion submodule is used to perform frequency domain to time domain conversion processing on the first mixed audio domain signal and the second mixed audio domain signal respectively, to obtain the processed first time domain audio signal and second time domain audio signal.

9. The apparatus according to claim 6, characterized in that, The first acquisition module includes: The acquisition submodule is used to acquire the time-domain audio signal to be processed; The second conversion submodule is used to perform Z-transform on the time-domain audio signal to obtain a complex spectrum signal.

10. The apparatus according to claim 6, characterized in that, The initial audio signal separation model includes a first recurrent neural network, an attention mechanism network, and a second recurrent neural network. The prediction module includes: The second input submodule is used to input the frequency domain signal corresponding to the current frame in the time domain audio signal sample into the first recurrent neural network and output the feature signal. The filtering submodule is used to filter the feature signals using the attention mechanism network to obtain filtered feature signals; wherein, the attention mechanism network caches frequency domain signals of a preset number of frames prior to the current frame; The second output submodule is used to input the filtered feature signal into the second recurrent neural network and output the mask corresponding to the first sub-signal.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the audio signal processing method according to any one of claims 1 to 5.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the audio signal processing method according to any one of claims 1 to 5.

13. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the audio signal processing method according to any one of claims 1 to 5.

Images