Audio signal processing method, device, computer equipment and storage medium

Abstract

The present application belongs to the field of audio processing technology, and relates to an audio signal processing method and related equipment, the method comprising: receiving an audio signal processing request sent by a user terminal; performing an adaptive echo cancellation operation on an initial audio signal to obtain an echo-cancelled audio signal; performing a time-frequency analysis operation on the echo-cancelled audio signal to obtain spectrum feature data; performing a mixed cancellation operation on the echo-cancelled audio signal according to the spectrum feature data to obtain a mixed cancellation audio signal; calling a trained deep learning model, and inputting the mixed cancellation audio signal into the deep learning model for noise cancellation to obtain a target audio signal; and outputting the target audio signal to the user terminal. The present application can comprehensively improve the quality of the audio signal, making the target audio signal outputted in the end clearer, purer, and more natural. In addition, the present application is applicable to a variety of voice communications and audio processing scenarios, and provides high-quality audio processing solutions for different application scenarios.

Description

Technical Field

[0001] The present application relates to the technical field of audio processing of artificial intelligence, and in particular to an audio signal processing method, apparatus, computer equipment and storage medium. Background Art

[0002] In remote conferencing systems, audio quality issues are becoming increasingly prominent due to the complex and changing environments of multiple participants.

[0003] There is an existing audio signal processing method, that is, by inverting the noise sound wave and superimposing it with the audio sound wave to be processed, the noise sound wave contained in the sound wave to be processed can be eliminated, thereby achieving a noise reduction effect.

[0004] However, the applicant found that this method can only eliminate a single noise in the surrounding environment. In practical applications, especially in high-noise and strong reverberation scenarios, traditional single echo cancellation or noise suppression methods are difficult to meet the needs of clear calls. How to achieve comprehensive optimization of audio signals in complex environments under limited computing resources has become a problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide an audio signal processing method, apparatus, computer equipment and storage medium to solve the problem of comprehensive optimization of audio signals in complex environments.

[0006] In order to solve the above technical problems, the present application provides an audio signal processing method, which adopts the following technical solution:

[0007] Receiving an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed;

[0008] Performing an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal;

[0009] Performing a time-frequency analysis operation on the echo cancellation audio signal to obtain frequency spectrum feature data;

[0010] Performing a mixed cancellation operation on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed cancellation audio signal;

[0011] Calling a trained deep learning model, inputting the mixed elimination audio signal into the deep learning model for noise elimination operation, and obtaining a target audio signal;

[0012] The target audio signal is output to the user terminal.

[0013] Furthermore, the step of performing an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal specifically includes the following steps:

[0014] Performing a filtering operation on the initial audio signal according to an adaptive filter to obtain a transfer function estimation value of the echo path;

[0015] An inverse filtering operation is performed on the initial audio signal according to the estimated value of the transfer function to obtain the echo-cancelled audio signal.

[0016] Furthermore, the step of performing a mixed cancellation operation on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed cancellation audio signal specifically includes the following steps:

[0017] Performing energy threshold segmentation processing on the frequency spectrum feature data to obtain mixed interference components;

[0018] Acquire a mixed power spectrum of the mixed interference component according to a minimum statistics method;

[0019] Subtracting the mixed power spectrum from the frequency spectrum feature data to obtain a pure speech signal power spectrum;

[0020] Calculating optimal filter coefficients of the clean speech signal power spectrum and the mixed power spectrum;

[0021] Performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain a Wiener filtered signal;

[0022] The Wiener filtered signal is used as the hybrid cancellation audio signal.

[0023] Furthermore, the step of performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain the Wiener filtered signal specifically includes the following steps:

[0024] Performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain an audio signal after Wiener filtering;

[0025] Performing an inverse time-frequency transform operation on the audio signal after the Wiener filter processing to obtain a time domain signal;

[0026] Performing a positioning operation on the time domain signal according to a voice activity detection algorithm to obtain a voice segment and a non-voice segment;

[0027] A speech enhancement operation is performed on the speech segment, and a mixed suppression process is performed on the non-speech segment to obtain the Wiener filter signal.

[0028] Furthermore, after the step of performing speech enhancement operation on the speech segment and performing mixed suppression processing on the non-speech segment to obtain the Wiener filter signal, the following step is also included:

[0029] Performing an azimuth positioning operation on the Wiener filter signal according to a sound source localization technology to obtain sound source azimuth information;

[0030] Performing a spatial filtering operation on the mixed-elimination audio signal according to the sound source orientation information to obtain a spatially filtered audio signal;

[0031] The step of using the Wiener filtered signal as the mixed audio signal specifically comprises the following steps:

[0032] The spatially filtered audio signal is used as the hybrid cancellation audio signal.

[0033] Furthermore, before the step of calling the trained deep learning model and inputting the mixed elimination audio signal into the deep learning model for noise reduction to obtain the target audio signal, the following steps are also included:

[0034] Acquire initial sample data, wherein the sample data includes a clean speech sample and a noise sample;

[0035] Performing sample preprocessing operations on the initial sample data to obtain a normalized training data set;

[0036] Performing an expansion operation on the training data set according to the data enhancement technology to obtain training sample data;

[0037] Construct an initial neural network model, input the training sample data into the initial neural network model for model training, and adjust the weight and bias parameters of the initial neural network model according to the forward propagation algorithm and the back propagation algorithm to obtain the trained deep learning model.

[0038] Furthermore, the step of calling the trained deep learning model, inputting the mixed elimination audio signal into the deep learning model for noise reduction operation, and obtaining the target audio signal specifically includes the following steps:

[0039] Identify key feature data for distinguishing speech and noise in the mixed-cancelled audio signal according to the trained deep learning model;

[0040] Constructing the speech feature vector and the noise feature vector according to the key feature data;

[0041] Performing clustering operations on the speech feature vector and the noise feature vector respectively according to a clustering algorithm to obtain a speech clustering center and a noise clustering center;

[0042] Calculating the speech center distance and the noise center distance between the mixed-cancelled audio signal and the speech cluster center and the noise cluster center respectively;

[0043] Classify the mixed elimination audio signal according to the speech center distance and the noise center distance to obtain a noise audio signal and a speech audio signal;

[0044] The noise audio signal is subjected to noise suppression processing to obtain the target audio signal.

[0045] In order to solve the above technical problems, the embodiment of the present application further provides an audio signal processing device, which adopts the following technical solution:

[0046] A request receiving module, configured to receive an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed;

[0047] An adaptive echo cancellation module, configured to perform an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal;

[0048] A time-frequency analysis module, used to perform a time-frequency analysis operation on the echo cancellation audio signal to obtain frequency spectrum feature data;

[0049] A mixed elimination module, configured to perform a mixed elimination operation on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed elimination audio signal;

[0050] A noise reduction module is used to call a trained deep learning model, input the mixed audio signal to the deep learning model for noise reduction, and obtain a target audio signal;

[0051] An output module is used to output the target audio signal to the user terminal.

[0052] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0053] The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the audio signal processing method as described above when executing the computer-readable instructions.

[0054] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0055] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the audio signal processing method described above are implemented.

[0056] The present application provides an audio signal processing method, including: receiving an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed; performing an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal; performing a time-frequency analysis operation on the echo-cancelled audio signal to obtain spectrum feature data; performing a mixed cancellation operation on the echo-cancelled audio signal according to the spectrum feature data to obtain a mixed cancellation audio signal; calling a trained deep learning model, inputting the mixed cancellation audio signal into the deep learning model for noise cancellation operation to obtain a target audio signal; and outputting the target audio signal to the user terminal. Compared with the prior art, the present application can comprehensively improve the quality of audio signals, making the final output target audio signal clearer, purer, and more natural. In addition, the present application is applicable to a variety of voice communication and audio processing scenarios, and provides high-quality audio processing solutions for different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0059] Figure 2 is a flowchart of an implementation of the audio signal processing method provided in Embodiment 1 of the present application;

[0060] Figure 3 is a structural diagram of an audio signal processing device provided in Embodiment 2 of the present application;

[0061] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0063] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0064] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0065] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0066] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0067] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0068] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0069] It should be noted that the audio signal processing method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the audio signal processing device is generally arranged in the server / terminal device.

[0070] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0071] Embodiment 1

[0072] Continue to refer Figure 2 , shows a flow chart of an embodiment of an audio signal processing method according to the present application. The audio signal processing method comprises: step S201, step S202, step S203, step S204, step S205 and step S206.

[0073] In step S201, an audio signal processing request sent by a user terminal is received, wherein the audio signal processing request includes an initial audio signal to be processed.

[0074] In the embodiments of the present application, the user terminal refers to a terminal device used to execute the image processing method for preventing document abuse provided by the present application. The user terminal can be a mobile terminal such as a mobile phone, a smart phone, a laptop computer, a digital broadcast receiver, a PDA (personal digital assistant), a PAD (tablet computer), a PMP (portable multimedia player), a navigation device, etc., as well as a fixed terminal such as a digital TV, a desktop computer, etc. It should be understood that the examples of user terminals here are only for convenience of understanding and are not used to limit the present application.

[0075] In the embodiments of the present application, the initial audio signal refers to the unprocessed original audio data, which may contain human voice, environmental noise, and echoes generated by the sound propagating in the space and reflecting back to the microphone (such as the sound emitted by the speaker being captured again by the microphone). These echoes will interfere with the original audio signal and reduce the clarity and intelligibility of the communication.

[0076] In an embodiment of the present application, after obtaining the initial audio signal to be processed, the initial audio signal can be preprocessed. The preprocessing operation can be operations such as removing DC components and normalizing the amplitude. Specifically, a high-pass filter can be used to remove the DC component of 0-20 Hz, and then the amplitude can be normalized to the range of [-1,1].

[0077] In step S202, an adaptive echo cancellation operation is performed on the initial audio signal to obtain an echo-cancelled audio signal.

[0078] In the embodiment of the present application, the adaptive echo cancellation operation (AEC) is mainly to automatically adjust parameters through an algorithm to eliminate or suppress echoes. The adaptive echo cancellation operation can be:

[0079] (1) Echo path estimation: Estimate the characteristics of the echo path from the speaker to the microphone, including the delay, attenuation, and frequency response of the path;

[0080] (2) Adaptive filter design: Based on the estimation of the echo path, an adaptive filter is designed to simulate the echo path. The coefficients of this filter are dynamically adjusted according to the difference between the actual received signal and the expected echo signal to minimize the impact of the echo;

[0081] (3) Echo cancellation: The analog echo signal generated by the adaptive filter is subtracted from the total signal received by the microphone to obtain a relatively pure original audio signal. This process needs to be performed in real time to adapt to the ever-changing communication environment and sound conditions.

[0082] In step S203, a time-frequency analysis operation is performed on the echo-cancelled audio signal to obtain frequency spectrum feature data.

[0083] In the embodiment of the present application, the operation of performing time-frequency analysis on the echo cancellation audio signal may be:

[0084] (1) Signal preprocessing: preprocess the echo cancellation audio signal, including denoising and filtering, to improve the signal quality;

[0085] (2) Time-frequency analysis: The preprocessed echo cancellation audio signal is divided into multiple short-time frames, and each frame is Fourier transformed to obtain the spectrum representation of the frame. This method can provide information about the signal in both time and frequency. Based on the results of the short-time Fourier transform (STFT), the time-frequency distribution can be further calculated, such as the Wigner-Ville distribution and the Choi-Williams distribution, to more intuitively display the time-frequency characteristics of the signal.

[0086] (3) Spectral feature extraction: Extract spectral feature data from the results of time-frequency analysis, such as spectral power, spectral peak, spectral bandwidth, etc.

[0087] In the embodiment of the present application, interference components such as white noise, wind noise, and reverberation in the audio can be identified by performing energy statistical analysis on the frequency spectrum.

[0088] In the embodiment of the present application, the frequency spectrum feature data is obtained through the time-frequency analysis operation, which can reveal the frequency components of the signal that change over time, which is particularly important for the analysis of speech signals.

[0089] In step S204, a mixed cancellation operation is performed on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed cancellation audio signal.

[0090] In the embodiment of the present application, the mixed cancellation operation may be a residual echo cancellation operation, a noise suppression operation, a sound enhancement operation, an audio mixing operation, etc. Specifically:

[0091] Residual echo cancellation: By analyzing the spectral feature data, the frequency characteristics and time patterns of these residual echoes can be identified, and more sophisticated filtering techniques (such as frequency-domain-based adaptive filtering) can be used to further eliminate them;

[0092] Noise suppression: In addition to echo, audio signals may contain other unwanted noise components. Spectral feature data can help identify the frequency range of these noises and apply noise suppression algorithms (such as spectral subtraction, Wiener filtering, etc.) to reduce their impact;

[0093] Sound enhancement: In some cases, it may be necessary to enhance specific sound components (such as human voice) in the audio signal while suppressing other components. By analyzing the spectral feature data, it is possible to determine which frequency components belong to the target sound and apply gain adjustment or filtering techniques to enhance them;

[0094] Audio mixing: If the audio signal contains multiple sound sources (such as different instruments or vocals in a multi-track recording), these sound sources may need to be mixed based on the spectral characteristic data to achieve the desired audio effect. This may include adjusting the volume, balance, reverberation and other parameters of the sound source.

[0095] In step S205, the trained deep learning model is called, and the mixed elimination audio signal is input into the deep learning model for noise reduction operation to obtain the target audio signal.

[0096] In an embodiment of the present application, a key feature for distinguishing speech and noise in a mixed-eliminated audio signal is identified through a trained deep learning model, and the noisy frequency of the mixed-eliminated audio signal is predicted based on the key feature, and the noise component is removed to obtain a clear speech signal.

[0097] In step S206, the target audio signal is output to the user terminal.

[0098] In an embodiment of the present application, an audio signal processing method is provided, including: receiving an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed; performing an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal; performing a time-frequency analysis operation on the echo-cancelled audio signal to obtain spectrum feature data; performing a mixed cancellation operation on the echo-cancelled audio signal according to the spectrum feature data to obtain a mixed cancellation audio signal; calling a trained deep learning model, inputting the mixed cancellation audio signal into the deep learning model for noise cancellation operation to obtain a target audio signal; and outputting the target audio signal to the user terminal. Compared with the prior art, the present application can comprehensively improve the quality of the audio signal, making the target audio signal outputted in the end clearer, purer, and more natural. In addition, the present application is applicable to a variety of voice communication and audio processing scenarios, and provides high-quality audio processing solutions for different application scenarios.

[0099] In some optional implementations of the embodiments of the present application, the above step S202 specifically includes the following steps:

[0100] Performing a filtering operation on the initial audio signal according to the adaptive filter to obtain a transfer function estimation value of the echo path;

[0101] The initial audio signal is subjected to an inverse filtering operation according to the estimated value of the transfer function to obtain an echo-cancelled audio signal.

[0102] In the embodiment of the present application, an adaptive filter is a filter that can automatically adjust its parameters according to the characteristics of the input signal. In the context of echo cancellation, an adaptive filter is used to simulate the echo path (also called the echo path) from the speaker to the microphone. This path may include various physical and electrical characteristics, such as delay, attenuation, frequency response changes, etc.

[0103] In the embodiment of the present application, the filtering operation refers to processing the initial audio signal through an adaptive filter to estimate the transfer function of the echo path. This transfer function describes the impact of the echo path on the audio signal, including the amplitude, phase and frequency response of the signal. By continuously adjusting the parameters of the adaptive filter, the output of the adaptive filter can be gradually approximated to the actual echo signal, thereby obtaining an estimated value of the transfer function of the echo path.

[0104] In the embodiment of the present application, the estimated value of the transfer function is an important result obtained by the adaptive filter after the filtering operation. It represents the overall impact of the echo path on the audio signal and is the basis for the subsequent reverse filtering operation. The accuracy of the estimated value of the transfer function directly affects the effect of echo cancellation. If the estimated value is inaccurate, the reverse filtering operation may not be able to completely eliminate the echo and may even introduce new distortion.

[0105] In the embodiment of the present application, the reverse filtering operation refers to processing the initial audio signal using the obtained transfer function estimate to eliminate the echo component therein. Specifically, the initial audio signal is processed through a filter (i.e., a reverse filter) that is opposite to the transfer function estimate. The function of this reverse filter is to invert the phase and adjust the amplitude of the echo signal so that it and the original echo signal cancel each other out in the time domain.

[0106] In the embodiment of the present application, the echo component in the initial audio signal can be effectively eliminated through the reverse filtering operation to obtain an echo-cancelled audio signal, which is purer and clearer, and is convenient for subsequent voice processing or communication transmission.

[0107] In the embodiment of the present application, the initial audio signal can be filtered according to the adaptive filter to obtain the estimated transfer function value of the echo path; the initial audio signal can be inversely filtered according to the estimated transfer function to obtain the echo-canceled audio signal.

[0108] In the embodiment of the present application, after the adaptive echo cancellation operation, the echo component in the original audio signal is significantly reduced or eliminated, thereby obtaining an echo-cancelled audio signal. This signal is clearer, reduces interference caused by the echo, and improves the comfort and efficiency of communication.

[0109] In some optional implementations of the embodiments of the present application, the above step S204 specifically includes the following steps:

[0110] Perform energy threshold segmentation processing on the spectrum feature data to obtain mixed interference components;

[0111] Obtain the mixed power spectrum of the mixed interference components according to the minimum statistics method;

[0112] Subtract the mixed power spectrum from the spectral feature data to obtain the pure speech signal power spectrum;

[0113] Calculate the optimal filter coefficients of the pure speech signal power spectrum and the mixed power spectrum;

[0114] Performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain a Wiener filtered signal;

[0115] The Wiener filtered signal is used as the hybrid cancellation audio signal.

[0116] In the embodiment of the present application, the spectrum feature data includes the energy distribution information of the audio signal in the frequency domain. By setting an energy threshold, these data can be segmented. Specifically, the frequency components in the spectrum with energy higher than the threshold are regarded as mixed components containing speech signals and possible interference (such as noise, residual echo, etc.), while the frequency components with energy lower than the threshold may be mainly background noise or very weak signals. This process helps to identify mixed interference components that need further processing.

[0117] In the embodiment of the present application, the minimum statistics method is a common method for estimating the noise power spectrum, but here, it is used to estimate the power spectrum of the mixed interference component. This method estimates the power spectrum of the mixed interference component in the frequency domain by analyzing the statistical characteristics (such as the statistics of energy distribution) in the spectrum feature data. This power spectrum reflects the energy distribution of the mixed interference component at different frequencies and is an important basis for subsequent signal processing.

[0118] In an embodiment of the present application, after obtaining a mixed power spectrum of a mixed interference component, the mixed power spectrum can be subtracted from the original spectrum feature data to try to remove the interference component. This step assumes that most of the energy in the spectrum feature data comes from the mixture of the speech signal and the interference, and the remaining part after subtracting the mixed power spectrum is closer to the pure speech signal power spectrum. However, it should be noted that this simple subtraction operation may not completely eliminate all interference, especially when the interference and the speech signal overlap more in the frequency domain.

[0119] In the embodiment of the present application, in order to further optimize the signal processing effect, the optimal filter coefficients between the power spectrum of the pure speech signal and the mixed power spectrum can be calculated. These coefficients are used to design a filter that can further optimize the mixed elimination audio signal in the frequency domain to maximize the retention of the speech signal and suppress interference. The calculation of the optimal filter coefficients usually involves some optimization algorithms, such as the design of the Wiener filter under the minimum mean square error (MMSE) criterion.

[0120] In the embodiment of the present application, the implementation method of calculating the optimal filter coefficient may be:

[0121] 1. Signal preprocessing:

[0122] Read signals: Use signal processing tools such as MATLAB to read pure speech signals and mixed speech signals.

[0123] Framing and windowing: The speech signal is divided into short time frames and a window function (such as a Hamming window) is applied to each frame to reduce spectral leakage.

[0124] 2. Spectrum analysis:

[0125] Fast Fourier Transform (FFT): Perform FFT on each frame of signal to convert the time domain signal into a frequency domain signal.

[0126] Power spectrum estimation: Calculate the power spectrum of each frame of the signal, usually by taking the square of the modulus of the FFT result.

[0127] 3. Compare power spectra:

[0128] Draw power spectrum: Draw the power spectrum of the pure speech signal and the power spectrum of the mixed speech signal respectively for intuitive comparison.

[0129] Analyze the difference: Identify the additional energy distribution caused by noise in the mixed power spectrum, which is usually more significant in the high-frequency part.

[0130] 4. Design filter:

[0131] Determine the filter type: Select the appropriate filter type (such as low-pass, band-pass, notch, etc.) based on the noise characteristics (such as frequency distribution, intensity, etc.).

[0132] Set filter parameters: including cutoff frequency, passband / stopband attenuation, etc. These parameters should be determined based on the difference in power spectrum between noise and clean speech signals.

[0133] 5. Calculate the filter coefficients:

[0134] (1) Filter design method:

[0135] ① FIR filter design based on window function: Design finite impulse response (FIR) filter using window function (such as Hanning window, Hamming window, etc.). According to the required cutoff frequency and filter length, select the appropriate window function and calculate the filter coefficients.

[0136] ②IIR filter design: For infinite impulse response (IIR) filters, you can use MATLAB's butter, cheby1, ellip and other functions to calculate the filter coefficients based on design indicators (such as passband and stopband cutoff frequency, attenuation, etc.).

[0137] (2) Optimization algorithm: In order to obtain the optimal filter coefficients, an optimization algorithm (such as gradient descent, genetic algorithm, etc.) can be used to minimize a cost function, which is usually related to the distortion and residual noise level of the filtered signal.

[0138] 6. Filter performance test:

[0139] Apply filter: Apply the designed filter to the mixed speech signal to obtain the filtered speech signal.

[0140] Evaluation effect: The effect of the filter is evaluated by comparing the power spectrum of the filtered speech signal with the power spectrum of the pure speech signal and by subjective listening tests.

[0141] 7. Iterative optimization:

[0142] If the filtering effect does not meet the requirements, you can adjust the filter parameters or change the filter design method, and then repeat the above steps for iterative optimization.

[0143] In the embodiment of the present application, the above steps can be used to compare the pure speech signal power spectrum and the mixed power spectrum, and calculate the optimal filter coefficient to achieve the purpose of suppressing noise while maintaining the quality of the speech signal as much as possible.

[0144] In the embodiment of the present application, after obtaining the optimal filter coefficients, the Wiener filtering algorithm can be used to filter the echo cancellation audio signal. Wiener filtering is a filtering method based on statistical optimality criteria, which can design the optimal filter to suppress noise and retain the signal when the statistical characteristics of the signal and noise are known. Here, the Wiener filtering algorithm uses the optimal filter coefficients calculated previously to process the echo cancellation audio signal to obtain a mixed cancellation audio signal. This signal further suppresses the interference components while retaining the speech signal.

[0145] In the embodiment of the present application, the mixed cancellation audio signal obtained by the Wiener filter processing is used as the final output. This signal has been optimized in terms of echo cancellation and noise suppression, so it is purer and clearer and suitable for subsequent voice processing or communication transmission.

[0146] In some optional implementations of the embodiments of the present application, the step of performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain the Wiener filtered signal specifically includes the following steps:

[0147] Performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain an audio signal after Wiener filtering;

[0148] Performing an inverse time-frequency transform operation on the audio signal after Wiener filtering to obtain a time domain signal;

[0149] Performing positioning operation on the time domain signal according to the voice activity detection algorithm to obtain the voice segment and the non-voice segment;

[0150] The speech segment is subjected to speech enhancement operation, and the non-speech segment is subjected to mixed suppression processing to obtain a Wiener filter signal.

[0151] In the embodiment of the present application, Wiener filtering is a filtering method based on statistical optimal criteria, which uses the statistical characteristics of the signal (such as power spectrum) to design the filter to minimize the mean square value of the estimation error. In audio processing, Wiener filtering is often used to suppress noise and interference while retaining the useful information of the original signal as much as possible. Performing Wiener filtering on the audio signal after echo cancellation can further reduce the residual noise and interference in the signal and improve the signal-to-noise ratio of the signal, thereby obtaining a clearer audio signal.

[0152] In the embodiment of the present application, time-frequency analysis can convert the audio signal from the time domain to the frequency domain for analysis. However, for the needs of subsequent processing (such as voice activity detection, voice enhancement, etc.), it is usually necessary to convert the processed frequency domain signal back to the time domain. The inverse time-frequency transform operation is the key step to achieve this conversion. By performing an inverse time-frequency transform on the audio signal after Wiener filtering, its representation in the time domain, that is, the time domain signal, can be obtained.

[0153] In an embodiment of the present application, Voice Activity Detection (VAD) is an algorithm for determining whether there is a voice component in an audio signal. After obtaining the time domain signal, the VAD algorithm can be used to locate the signal and divide the signal into a voice segment and a non-voice segment. The voice segment is the part containing human voice or other voice signals, while the non-voice segment is mainly noise, silence or other non-voice components. This step provides a basis for subsequent targeted processing of different parts.

[0154] In an embodiment of the present application, after the division of speech segments and non-speech segments is obtained, these two parts can be processed specifically. For speech segments, a speech enhancement algorithm can be used to further improve the clarity and intelligibility of speech signals. Speech enhancement algorithms usually include noise suppression, echo cancellation, spectrum repair and other technologies, aiming to improve the quality of speech signals. For non-speech segments, mixed suppression processing is mainly performed, that is, further suppressing the noise and interference components therein to reduce their impact on the overall audio quality. Through this step, a mixed elimination audio signal can be obtained, which effectively suppresses the noise and interference components while retaining the speech signal, thereby improving the overall quality of the audio.

[0155] In the embodiment of the present application, by performing Wiener filtering, inverse time-frequency transform, voice activity detection and targeted processing based on voice segments and non-voice segments on the audio signal after echo cancellation, a purer and clearer mixed cancellation audio signal can be obtained. This processing flow has wide application value in many fields such as voice communication, voice recognition, and audio editing.

[0156] In some optional implementations of the embodiments of the present application, after performing a speech enhancement operation on the speech segment and performing a mixed suppression process on the non-speech segment to obtain a Wiener filter signal, the following steps are also included:

[0157] Performing azimuth positioning operation on the mixed-elimination audio signal according to the sound source localization technology to obtain the sound source azimuth information;

[0158] Performing a spatial filtering operation on the mixed elimination audio signal according to the sound source orientation information to obtain a spatial filtered audio signal;

[0159] The step of using the Wiener filter signal as a mixed elimination audio signal specifically includes the following steps:

[0160] The spatially filtered audio signal is used as a mixed-cancellation audio signal.

[0161] In the embodiment of the present application, the time difference, phase difference, spectrum characteristics and other clues in the signal are further analyzed to determine the specific position or direction of the sound source in the three-dimensional space, wherein the result of the azimuth localization operation is the sound source azimuth information, which may include the direction (such as angle) of the sound source, the distance (in some cases) or a more complex description of the spatial position. This information is crucial for subsequent spatial filtering operations because it guides how to distinguish and process audio signals from different directions in space.

[0162] In the embodiment of the present application, after obtaining the sound source orientation information, a spatial filtering operation may be performed on the mixed elimination audio signal to improve the spatial selectivity of the signal.

[0163] In the embodiment of the present application, after completing the spatial filtering operation, the spatially filtered audio signal obtained will usually have a significant improvement in quality. It not only removes more noise and interference, but also enhances the directionality of the target sound source, making the entire audio scene clearer and more realistic. Therefore, in the later stage of the audio processing flow, this audio signal processed by spatial filtering can be output as the final mixed elimination audio signal.

[0164] In the embodiment of the present application, the spatial perception quality and signal-to-noise ratio of the mixed-cancelled audio signal can be effectively improved by combining the sound source localization technology with spatial filtering. This technology has broad application prospects in many fields such as audio conferencing, speech recognition, and soundscape reconstruction.

[0165] In some optional implementations of the embodiments of the present application, before the above step S205, the following steps are also included:

[0166] Acquire initial sample data, wherein the sample data includes a clean speech sample and a noise sample;

[0167] Perform sample preprocessing operations on the initial sample data to obtain a standardized training data set;

[0168] The training data set is expanded according to the data enhancement technology to obtain training sample data;

[0169] Construct an initial neural network model, input the training sample data into the initial neural network model for model training, and adjust the weight and bias parameters of the initial neural network model according to the forward propagation algorithm and the back propagation algorithm to obtain a trained deep learning model.

[0170] In an embodiment of the present application, a large number of clean speech samples and noise samples need to be prepared first, and the sample data is preprocessed, including amplitude normalization, silence removal and other operations, to obtain a standardized training data set. Specifically, the audio data can be sampled at 16kHz, quantized at 16 bits, and spectral features extracted by short-time Fourier transform. The logarithmic power spectrum of the spectrum is then calculated, and the MFCC features are obtained by DCT transformation. For data enhancement, the audio data can be processed at a speed of 9 times to 1 times, Gaussian white noise with a signal-to-noise ratio of -5dB to 20dB can be added, and the reverberation effect can be simulated by room impulse response.

[0171] In the embodiment of the present application, when constructing a convolutional neural network model, the classic LeNet-5 structure can be used, the input layer is a 32×32 grayscale image, the first convolution layer contains 6 5×5 convolution kernels, the second convolution layer contains 16 5×5 convolution kernels, and both convolution layers are followed by a maximum pooling layer, and finally the classification result is obtained through a fully connected layer and a Softmax output layer. For a recurrent neural network, an LSTM structure can be used, the input layer is a 39-dimensional MFCC feature vector sequence, the hidden layer contains 128 LSTM units, and the output layer is the probability distribution of the noise category.

[0172] In the embodiment of the present application, during the model training phase, the Adam optimizer can be used, the learning rate is set to 0.01, the batch size is 128, and 100 epochs are trained. At the same time, the early stopping method is used, and the training is stopped when the validation set loss does not decrease for 5 consecutive epochs.

[0173] In the embodiment of the present application, if the model effect is not ideal, you can try to increase the number of convolutional layers to 4, the number of LSTM hidden units to 256, and introduce the L2 regularization term, and set the regularization coefficient to 01.

[0174] In the embodiment of the present application, in the model evaluation stage, indicators such as PESQ and STOI can be used to evaluate the quality and intelligibility of the noise-reduced speech. If the average PESQ score reaches 5 or more, and the STOI reaches 8 or more, it can be considered that the model performance meets the practical requirements.

[0175] In practical applications, the trained model can be used to process the real-time audio stream at the frame level, extract features through a sliding window, send them to the model for noise prediction, and use spectral subtraction and other techniques to perform noise reduction, and finally output a clear speech signal.

[0176] In some optional implementations of the embodiments of the present application, the above step S205 specifically includes the following steps:

[0177] Identify key feature data for distinguishing speech and noise in the mixed-cancellation audio signal based on the trained deep learning model;

[0178] Constructing a speech feature vector and a noise feature vector according to key feature data;

[0179] According to the clustering algorithm, the speech feature vector and the noise feature vector are clustered to obtain the speech cluster center and the noise cluster center;

[0180] Calculate the speech center distance and the noise center distance of the mixed-elimination audio signal and the speech cluster center and the noise cluster center respectively;

[0181] Classify the mixed elimination audio signal according to the speech center distance and the noise center distance to obtain a noise audio signal and a speech audio signal;

[0182] The noise audio signal is subjected to noise suppression processing to obtain a target audio signal.

[0183] In the embodiment of the present application, the trained deep learning model learns and identifies the key features that distinguish speech and noise in the audio signal. These features may include spectral characteristics, time dynamics, phase information, energy distribution, etc. The model learns how to distinguish these features through a large amount of training data (including pure speech and various noise samples), and accurately identifies the speech and noise components in the mixed elimination audio signal in practical applications.

[0184] In the embodiment of the present application, once the key feature data is identified, feature vectors can be constructed based on these features. For speech and noise, speech feature vectors and noise feature vectors are constructed respectively. These vectors are mathematical representations of the feature data and are used for subsequent processing and analysis.

[0185] In the embodiment of the present application, a clustering algorithm (such as K-means, hierarchical clustering, etc.) is used to cluster the speech feature vector and the noise feature vector. The purpose of clustering is to group similar feature vectors so as to find the cluster centers of speech and noise. These cluster centers represent the typical features of their respective categories, which is helpful for subsequent classification operations.

[0186] In the embodiment of the present application, after obtaining the cluster centers of speech and noise, the distance between each sample in the mixed-elimination audio signal and the two cluster centers is calculated. These distances (speech center distance and noise center distance) are used to evaluate the similarity between the sample and the speech or noise category.

[0187] In the embodiment of the present application, each sample in the mixed audio signal can be classified based on the distance calculated above. Usually, the sample is classified into a category (speech or noise) that is closer to its cluster center. In this way, the mixed signal can be decomposed into two parts: a noise audio signal and a speech audio signal.

[0188] In the embodiment of the present application, noise suppression processing is performed on the noise audio signal obtained by classification. This can be achieved by various methods, such as using a noise suppression filter, spectral subtraction, Wiener filtering, etc. The purpose of noise suppression is to further reduce or eliminate the noise component, thereby obtaining a purer target audio signal (i.e., speech signal).

[0189] In the embodiment of the present application, by combining deep learning models, feature extraction, clustering algorithms and noise suppression technology, a pure speech signal can be effectively extracted from a mixed audio signal and noise components can be suppressed. This technology has broad application prospects in many fields such as audio conferencing, speech recognition, and soundscape reconstruction, and can significantly improve the quality and efficiency of audio processing.

[0190] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0191] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0192] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0193] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0194] Embodiment 2

[0195] Further references Figure 3 , as a response to the above Figure 2The present application provides an embodiment of an audio signal processing device, and the device embodiment is Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0196] like Figure 3 As shown, the audio signal processing device 200 of the embodiment of the present application includes:

[0197] A request receiving module 210, configured to receive an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed;

[0198] The adaptive echo cancellation module 220 is used to perform an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal;

[0199] The time-frequency analysis module 230 is used to perform time-frequency analysis on the echo cancellation audio signal to obtain frequency spectrum feature data;

[0200] A mixed elimination module 240 is used to perform a mixed elimination operation on the echo cancellation audio signal according to the spectrum feature data to obtain a mixed elimination audio signal;

[0201] The noise reduction module 250 is used to call the trained deep learning model, input the mixed audio signal to the deep learning model for noise reduction, and obtain the target audio signal;

[0202] The output module 260 is used to output the target audio signal to the user terminal.

[0203] In an embodiment of the present application, an audio signal processing device 200 is provided, including: a request receiving module 210, which is used to receive an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed; an adaptive echo cancellation module 220, which is used to perform an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal; a time-frequency analysis module 230, which is used to perform a time-frequency analysis operation on the echo-cancelled audio signal to obtain spectrum feature data; a mixed elimination module 240, which is used to perform a mixed elimination operation on the echo-cancelled audio signal according to the spectrum feature data to obtain a mixed elimination audio signal; a noise reduction module 250, which is used to call a trained deep learning model, input the mixed elimination audio signal into the deep learning model for noise reduction operation, and obtain a target audio signal; an output module 260, which is used to output the target audio signal to the user terminal. Compared with the prior art, the present application can comprehensively improve the quality of the audio signal, so that the target audio signal finally outputted is clearer, purer, and more natural. In addition, the present application is applicable to a variety of voice communication and audio processing scenarios, and provides high-quality audio processing solutions for different application scenarios.

[0204] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of a computer device according to an embodiment of the present application.

[0205] The computer device 300 includes a memory 310, a processor 320, and a network interface 330 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 300 having components 310-330, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0206] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0207] The memory 310 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 310 can be an internal storage unit of the computer device 300, such as a hard disk or memory of the computer device 300. In other embodiments, the memory 310 can also be an external storage device of the computer device 300, such as a plug-in hard disk equipped on the computer device 300, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Of course, the memory 310 can also include both the internal storage unit of the computer device 300 and its external storage device. In the embodiment of the present application, the memory 310 is generally used to store the operating system and various application software installed on the computer device 300, such as computer-readable instructions of the audio signal processing method, etc. In addition, the memory 310 can also be used to temporarily store various data that have been output or are to be output.

[0208] The processor 320 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 320 is generally used to control the overall operation of the computer device 300. In the embodiment of the present application, the processor 320 is used to run the computer-readable instructions stored in the memory 310 or process data, such as computer-readable instructions for running the audio signal processing method.

[0209] The network interface 330 may include a wireless network interface or a wired network interface. The network interface 330 is generally used to establish a communication connection between the computer device 300 and other electronic devices.

[0210] The computer device provided by this application can comprehensively improve the quality of audio signals, making the final output target audio signals clearer, purer and more natural. In addition, this application is applicable to a variety of voice communication and audio processing scenarios, and provides high-quality audio processing solutions for different application scenarios.

[0211] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the audio signal processing method as described above.

[0212] The computer-readable storage medium provided by the present application can comprehensively improve the quality of the audio signal, making the final output target audio signal clearer, purer and more natural. In addition, the present application is applicable to a variety of voice communication and audio processing scenarios, and provides high-quality audio processing solutions for different application scenarios.

[0213] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0214] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An audio signal processing method, characterized in that: The steps include: Receiving an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed; Performing an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal; Performing a time-frequency analysis operation on the echo cancellation audio signal to obtain frequency spectrum feature data; Performing a mixed cancellation operation on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed cancellation audio signal; Calling a trained deep learning model, inputting the mixed elimination audio signal into the deep learning model for noise elimination operation, and obtaining a target audio signal; Outputting the target audio signal to the user terminal; The step of performing a mixed cancellation operation on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed cancellation audio signal specifically comprises the following steps: Performing energy threshold segmentation processing on the frequency spectrum feature data to obtain mixed interference components; Acquire a mixed power spectrum of the mixed interference component according to a minimum statistics method; Subtracting the mixed power spectrum from the frequency spectrum feature data to obtain a pure speech signal power spectrum; Calculating optimal filter coefficients of the clean speech signal power spectrum and the mixed power spectrum; Performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain a Wiener filtered signal; Using the Wiener filtered signal as the mixed-cancellation audio signal; The step of calling the trained deep learning model, inputting the mixed elimination audio signal into the deep learning model for noise elimination operation, and obtaining the target audio signal specifically includes the following steps: Identify key feature data for distinguishing speech and noise in the mixed-cancelled audio signal according to the trained deep learning model; Constructing a speech feature vector and a noise feature vector according to the key feature data; Performing clustering operations on the speech feature vector and the noise feature vector respectively according to a clustering algorithm to obtain a speech clustering center and a noise clustering center; Calculating the speech center distance and the noise center distance between the mixed-cancelled audio signal and the speech cluster center and the noise cluster center respectively; Classify the mixed elimination audio signal according to the speech center distance and the noise center distance to obtain a noise audio signal and a speech audio signal; The noise audio signal is subjected to noise suppression processing to obtain the target audio signal.

2. The audio signal processing method according to claim 1, characterized in that: The step of performing an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal specifically comprises the following steps: Performing a filtering operation on the initial audio signal according to an adaptive filter to obtain a transfer function estimation value of the echo path; An inverse filtering operation is performed on the initial audio signal according to the estimated value of the transfer function to obtain the echo-cancelled audio signal.

3. The audio signal processing method according to claim 1, characterized in that: The step of performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain a Wiener filtered signal specifically includes the following steps: Performing Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain an audio signal after Wiener filtering; Performing an inverse time-frequency transform operation on the audio signal after the Wiener filter processing to obtain a time domain signal; Performing a positioning operation on the time domain signal according to a voice activity detection algorithm to obtain a voice segment and a non-voice segment; A speech enhancement operation is performed on the speech segment, and a mixed suppression process is performed on the non-speech segment to obtain the Wiener filter signal.

4. The audio signal processing method according to claim 3, characterized in that: After the step of performing speech enhancement operation on the speech segment and performing mixed suppression processing on the non-speech segment to obtain the Wiener filter signal, the following step is also included: Performing an azimuth positioning operation on the Wiener filter signal according to a sound source localization technology to obtain sound source azimuth information; Performing a spatial filtering operation on the mixed-elimination audio signal according to the sound source orientation information to obtain a spatially filtered audio signal; The step of using the Wiener filtered signal as the mixed audio signal specifically comprises the following steps: The spatially filtered audio signal is used as the hybrid cancellation audio signal.

5. The audio signal processing method according to claim 1, characterized in that: Before the step of calling the trained deep learning model and inputting the mixed elimination audio signal into the deep learning model for noise elimination to obtain the target audio signal, the following step is also included: Acquire initial sample data, wherein the sample data includes a clean speech sample and a noise sample; Performing sample preprocessing operations on the initial sample data to obtain a normalized training data set; Performing an expansion operation on the training data set according to the data enhancement technology to obtain training sample data; Construct an initial neural network model, input the training sample data into the initial neural network model for model training, and adjust the weight and bias parameters of the initial neural network model according to the forward propagation algorithm and the back propagation algorithm to obtain the trained deep learning model.

6. An audio signal processing device, characterized in that: include: A request receiving module, configured to receive an audio signal processing request sent by a user terminal, wherein the audio signal processing request includes an initial audio signal to be processed; An adaptive echo cancellation module, configured to perform an adaptive echo cancellation operation on the initial audio signal to obtain an echo-cancelled audio signal; A time-frequency analysis module, used to perform a time-frequency analysis operation on the echo cancellation audio signal to obtain frequency spectrum feature data; A mixed elimination module, configured to perform a mixed elimination operation on the echo cancellation audio signal according to the frequency spectrum feature data to obtain a mixed elimination audio signal; A noise reduction module is used to call a trained deep learning model, input the mixed audio signal to the deep learning model for noise reduction, and obtain a target audio signal; An output module, used to output the target audio signal to the user terminal; The hybrid elimination module includes: an energy threshold segmentation submodule, a hybrid power spectrum acquisition submodule, a hybrid power spectrum processing submodule, an optimal filter coefficient calculation submodule, a Wiener filtering submodule and a hybrid elimination audio signal confirmation submodule, wherein: The energy threshold segmentation submodule is used to perform energy threshold segmentation processing on the spectrum feature data to obtain mixed interference components; The mixed power spectrum acquisition submodule is used to obtain the mixed power spectrum of the mixed interference component according to the minimum statistics method; The mixed power spectrum processing submodule is used to subtract the mixed power spectrum from the frequency spectrum feature data to obtain a pure speech signal power spectrum; The optimal filter coefficient calculation submodule is used to calculate the optimal filter coefficients of the clean speech signal power spectrum and the mixed power spectrum; The Wiener filtering submodule is used to perform Wiener filtering on the echo cancellation audio signal according to the Wiener filtering algorithm and the optimal filter coefficient to obtain a Wiener filtered signal; The mixed-elimination audio signal confirmation submodule is used to use the Wiener filter signal as the mixed-elimination audio signal; The noise elimination module includes: a key feature recognition submodule, a feature vector construction submodule, a clustering submodule, a center distance calculation submodule, a classification submodule and a noise suppression submodule, wherein: The key feature identification submodule is used to identify key feature data for distinguishing speech and noise in the mixed elimination audio signal according to the trained deep learning model; The feature vector construction submodule is used to construct a speech feature vector and a noise feature vector according to the key feature data; The clustering submodule is used to perform clustering operations on the speech feature vector and the noise feature vector according to a clustering algorithm to obtain a speech clustering center and a noise clustering center; The center distance calculation submodule is used to calculate the speech center distance and the noise center distance between the mixed-cancelled audio signal and the speech cluster center and the noise cluster center respectively; The classification submodule is used to perform a classification operation on the mixed elimination audio signal according to the speech center distance and the noise center distance to obtain a noise audio signal and a speech audio signal; The noise suppression submodule is used to perform noise suppression processing on the noisy audio signal to obtain the target audio signal.

7. A computer device comprising a memory and a processor, characterized in that: The memory stores computer-readable instructions, and the processor implements the steps of the audio signal processing method according to any one of claims 1 to 5 when executing the computer-readable instructions.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the audio signal processing method according to any one of claims 1 to 5 are implemented.

Images