Audio noise reduction method, apparatus, device, storage medium and program product
By decomposing the entire frequency band into subbands and processing them in parallel, the noise suppression capability and stability issues of active noise reduction systems are solved, achieving efficient and stable noise reduction for broadband noise while reducing hardware and computational complexity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU ANYKA MICROELECTRONICS CO LTD
- Filing Date
- 2026-06-10
- Publication Date
- 2026-07-31
AI Technical Summary
Existing active noise reduction systems have shortcomings in terms of noise suppression capability, flexibility, stability and hardware implementation. In particular, adaptive ANC systems have slow convergence speed or divergent algorithms, and fixed filters cannot track noise frequency changes in real time.
The full frequency band is decomposed into N sub-bands by a frequency divider filter. In the offline identification stage, an independent processing path is built for each sub-band. In the online noise reduction stage, the signals of each sub-band are denoised in parallel. Filters with fixed parameters and second-order section filters are cascaded and combined with dynamic weighting coefficients of a neural network model to control the signal.
It improves the convergence speed and stability of the algorithm, can flexibly track noise changes in different frequency bands, achieves efficient and stable noise reduction for broadband noise, and reduces hardware and computational complexity.
Smart Images
Figure CN122493818A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of audio processing technology, and in particular to an audio noise reduction method, apparatus, device, storage medium, and program product. Background Technology
[0002] Noise pollution can negatively impact residents' physical and mental health. Active noise control (ANC) methods effectively suppress low-frequency noise and offer significant advantages in terms of size, weight, volume, and cost, leading to rapid development in recent years. Open-ended noise reduction systems require adding a target noise estimation path to the traditional ANC pathway. However, while most adaptive ANC systems can effectively control noise, their convergence speed can be affected by noise levels. Slow convergence or algorithm divergence can prevent effective noise suppression, resulting in poor robustness. Using fixed filters cannot track noise frequency changes in real time, limiting noise reduction capabilities and failing to balance flexibility, stability, and ease of hardware implementation. Summary of the Invention
[0003] Therefore, it is necessary to provide an audio noise reduction method, device, equipment, storage medium, and program product that can ensure noise reduction capability while taking into account flexibility, stability, and ease of hardware implementation, in order to address the above-mentioned technical problems.
[0004] In a first aspect, this application provides an audio noise reduction method, the method comprising:
[0005] During the offline identification phase, the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function of the noise reduction system are measured.
[0006] The full-band described by the first primary path transfer function is decomposed using a frequency divider filter to obtain N sub-bands;
[0007] For each sub-band, based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, a sub-band processing path corresponding to the sub-band is constructed by fitting, wherein the sub-band processing path is composed of filters with fixed parameters;
[0008] During the online noise reduction stage, the original audio signal to be denoised is acquired through the reference microphone in the noise reduction system, and the original audio signal is decomposed into N sub-band signals using the frequency division filter.
[0009] For each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal;
[0010] Based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal.
[0011] In one embodiment, the step of constructing the sub-band processing path corresponding to each sub-band by fitting based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function includes:
[0012] For each sub-band, the system transfer function of the noise reduction system is determined based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function.
[0013] Construct a target cost function based on the distance between the system transfer function and the target transfer function preset for the noise reduction system;
[0014] Using the second primary path transfer function, the second secondary path transfer function, and the second feedback path transfer function under multiple different scenarios as inputs, and minimizing the target cost function as the optimization objective, a set of filter coefficients consisting of L cascaded second-order section filters is solved to construct the subband processing path corresponding to the subband; wherein, the subband processing path is composed of at least the L cascaded second-order section filters.
[0015] In one embodiment, constructing the target cost function based on the distance between the system transfer function and a preset target transfer function for the noise reduction system includes:
[0016] Multiple sampling points are obtained by sampling the frequency domain of both the system transfer function and the target transfer function preset for the noise reduction system.
[0017] The weighting factor corresponding to each of the plurality of sampling points is determined based on the target transfer function; wherein, the sampling points obtained by sampling in the frequency band with more peaks and valleys in the target transfer function have a higher weighting factor.
[0018] For each sampling point, an initial cost function corresponding to the sampling point is constructed based on the distance between the system transfer function and the target transfer function;
[0019] Based on the weight factors corresponding to each of the multiple sampling points, the initial cost functions corresponding to each of the multiple sampling points are weighted and summed to obtain the target cost function.
[0020] In one embodiment, sampling the frequency domain of both the system transfer function and a target transfer function preset for the noise reduction system to obtain multiple sampling points includes:
[0021] A sampling strategy is determined based on the target transfer function; wherein the sampling strategy is used to ensure that there are more sampling points in the frequency bands where the target transfer function has many peaks and valleys;
[0022] Based on the sampling strategy, the frequency domain of both the system transfer function and the target transfer function preset for the noise reduction system is sampled to obtain multiple sampling points.
[0023] In one embodiment, the step of taking multiple sets of second primary path transfer functions, second secondary path transfer functions, and second feedback path transfer functions under different scenarios as inputs, and minimizing the target cost function as the optimization objective, to solve for a set of filter coefficients composed of L cascaded second-order section filters, in order to construct the sub-band processing path corresponding to the sub-band, includes:
[0024] Using the second primary path transfer function, the second secondary path transfer function, and the second feedback path transfer function under multiple different scenarios as inputs, and minimizing the target cost function as the optimization objective, a set of filter coefficients consisting of L cascaded second-order section filters is solved to construct the second-order section filter bank corresponding to the sub-band.
[0025] An all-pass filter for delay adjustment will be introduced into the second-order section filter bank to construct the sub-band processing path corresponding to the sub-band; wherein, the sub-band processing path is composed of the all-pass filter and the L second-order section filters cascaded together.
[0026] In one embodiment, the step of inputting the sub-band signal into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal includes:
[0027] For each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the initial sub-band control signal corresponding to the sub-band signal;
[0028] Based on the frequency response characteristics of the loudspeaker in the noise reduction system, the initial sub-band control signal is subjected to frequency response equalization processing to obtain the sub-band control signal corresponding to the sub-band signal.
[0029] In one embodiment, the step of combining the sub-band control signals corresponding to the N sub-band signals according to the signal characteristics of the original audio signal using dynamic weighting coefficients to obtain the target control signal includes:
[0030] The original audio signal is input into a trained neural network model to extract the signal features of the original audio signal through the neural network model, and the weighting coefficients and combination strategies corresponding to the N sub-band signals are determined based on the signal features.
[0031] Based on the weighting coefficients and combination strategies corresponding to the N sub-band signals, the sub-band control signals corresponding to the N sub-band signals are combined to obtain the target control signal.
[0032] In one embodiment, the N sub-band signals are input in parallel to the sub-band processing paths corresponding to each of the N sub-band signals to perform noise reduction processing in parallel.
[0033] Secondly, this application provides an audio noise reduction device, the device comprising:
[0034] The measurement module is used to measure the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function of the noise reduction system during the offline identification phase.
[0035] The decomposition module is used to decompose the full frequency band described by the first primary path transfer function using a frequency division filter to obtain N sub-bands;
[0036] The construction module is used to construct a sub-band processing path corresponding to each sub-band by fitting based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, wherein the sub-band processing path is composed of filters with fixed parameters.
[0037] The acquisition module is used to acquire the original audio signal to be denoised through the reference microphone in the noise reduction system during the online noise reduction stage.
[0038] The decomposition module is also used to decompose the original audio signal into N sub-band signals using the frequency division filter;
[0039] The noise reduction module is used to input the sub-band signal into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal; based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal.
[0040] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the various method embodiments of this application.
[0041] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the various method embodiments of this application.
[0042] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the various method embodiments of this application.
[0043] In the aforementioned audio noise reduction method, apparatus, device, storage medium, and program product, during the offline identification phase, the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function of the noise reduction system are measured; a frequency-division filter is used to decompose the full frequency band described by the first primary path transfer function to obtain N sub-bands; for each sub-band, based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, a sub-band processing path corresponding to the sub-band is constructed by fitting, wherein the sub-band processing path is composed of filters with fixed parameters; during the online noise reduction phase, the original audio signal to be denoised is acquired through a reference microphone in the noise reduction system, and the original audio signal is decomposed into N sub-band signals using a frequency-division filter; for each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal; based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal. Compared to traditional audio noise reduction methods, this application decomposes the entire frequency band into N sub-bands using a frequency-division filter. In the offline identification stage, an independent processing path is constructed for each sub-band, and in the online noise reduction stage, the signals of each sub-band are processed in parallel. On the one hand, sub-band decomposition reduces the order and computational complexity of each path, facilitating hardware implementation and improving the algorithm's convergence speed and stability, avoiding the problems of easy divergence or slow convergence in full-band adaptive algorithms. On the other hand, the combination of fixed filters and sub-band structures allows for flexible tracking of noise changes in different frequency bands while maintaining system robustness, achieving efficient and stable noise reduction for broadband noise. Attached Figure Description
[0044] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0045] Figure 1 This is a flowchart illustrating an audio noise reduction method in one embodiment;
[0046] Figure 2 This is a schematic diagram of the offline identification stage system setup in one embodiment;
[0047] Figure 3 This is a schematic diagram of the online noise reduction stage system setup in one embodiment;
[0048] Figure 4 This is a schematic diagram of the system architecture after integration with chip hardware in one embodiment;
[0049] Figure 5 This is a schematic diagram comparing the sub-band optimized splicing and broadband optimized results in one embodiment;
[0050] Figure 6 This is a schematic diagram comparing the noise reduction effects of a sub-band combination scheme and an adaptive scheme in one embodiment.
[0051] Figure 7 This is a structural block diagram of an audio noise reduction device in one embodiment;
[0052] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0053] 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.
[0054] In one embodiment, such as Figure 1 As shown, an audio noise reduction method is provided, which can be applied to computer devices, and includes the following steps:
[0055] Step 102: In the offline identification stage, measure the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function of the noise reduction system.
[0056] Among them, such as Figure 2 As shown, in the offline identification stage, the noise reduction system includes a reference microphone, a target position error microphone, and a speaker (i.e., a secondary sound source) that outputs control signals. The target position error microphone refers to the error microphone placed at the target noise reduction location during the offline identification stage. This application requires placing an error microphone at the target noise reduction location during the offline identification stage to estimate the transfer function. For example... Figure 3 As shown, in the online noise reduction stage, the noise reduction system includes a reference microphone and a speaker (i.e., a secondary sound source) that outputs control signals, but does not include an error microphone.
[0057] It is understandable that, compared with traditional adaptive open noise reduction systems, this application reduces complexity and computational load while improving stability. Since open noise reduction systems cannot directly obtain the error signal at the target noise reduction location, they generally need to introduce a real error microphone (where a real error microphone refers to the error microphone placed at the target noise reduction location during the online noise reduction stage) and estimate the error at the target noise reduction location. However, in this scheme, during the online noise reduction stage, it is not necessary to place an error microphone at the target noise reduction location, nor is it necessary to estimate the direct noise signal at the target noise reduction location, thus reducing hardware system complexity. By directly combining the results after filtering with fixed parameters, the potential divergence problem of adaptive filtering systems is avoided, resulting in increased system stability.
[0058] The primary path transfer function is the transfer function of the primary path from the reference microphone to the target position error microphone; the secondary path transfer function is the transfer function of the secondary path from the secondary sound source (used to play noise control signals) to the target position error microphone; and the feedback path transfer function is the transfer function of the feedback path from the secondary sound source to the reference microphone.
[0059] Step 104: Use a frequency divider filter to decompose the full frequency band described by the first primary path transfer function to obtain N sub-bands.
[0060] Where N is a positive integer greater than 1.
[0061] In one embodiment, N frequency-divided filters A(z) are designed to divide the first primary path transfer function into N sub-bands (i.e., narrowbands). It is necessary to ensure that the time-domain response of the superimposed frequency-divided filters is as close as possible to the pulse signal, that is, to ensure that the signals of each sub-band after frequency division are... The superimposed signal is equivalent to the original signal x(n) in the time domain, meaning it must satisfy the following formula:
[0062] ;
[0063] Step 106: For each sub-band, based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, a sub-band processing path corresponding to the sub-band is constructed by fitting. The sub-band processing path is composed of filters with fixed parameters.
[0064] Step 108: In the online noise reduction stage, the original audio signal to be denoised is acquired through the reference microphone in the noise reduction system, and the original audio signal is decomposed into N sub-band signals using a frequency division filter.
[0065] Step 110: For each sub-band signal, input the sub-band signal into the corresponding sub-band processing path for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal.
[0066] Step 112: Based on the signal characteristics of the original audio signal, combine the sub-band control signals corresponding to each of the N sub-band signals according to the dynamic weighting coefficients to obtain the target control signal.
[0067] In one embodiment, a computer device can, based on the signal characteristics of the original audio signal, weight and combine the sub-band control signals corresponding to each of the N sub-band signals according to dynamic weighting coefficients to obtain a target control signal. It can be understood that, based on the original audio signal collected by the reference microphone, the target control signal is output after processing by the audio noise reduction method of this application. This target control signal drives the speaker (i.e., the secondary sound source) to play anti-noise, which is then superimposed on the original noise at the target noise reduction location to achieve noise reduction.
[0068] In the aforementioned audio noise reduction method, during the offline identification stage, the transfer function of the first primary path, the transfer function of the first intermediate path, and the transfer function of the first feedback path of the noise reduction system are measured. A frequency-division filter is used to decompose the entire frequency band described by the first primary path transfer function, resulting in N sub-bands. For each sub-band, a sub-band processing path is constructed by fitting based on the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function. The sub-band processing path consists of filters with fixed parameters. During the online noise reduction stage, the original audio signal to be denoised is acquired through a reference microphone in the noise reduction system, and the original audio signal is decomposed into N sub-band signals using a frequency-division filter. For each sub-band signal, the sub-band signal is input into the corresponding sub-band processing path for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal. Based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal. Compared to traditional audio noise reduction methods, this application decomposes the entire frequency band into N sub-bands using a frequency-division filter. In the offline identification stage, an independent processing path is constructed for each sub-band, and in the online noise reduction stage, the signals of each sub-band are processed in parallel. On the one hand, sub-band decomposition reduces the order and computational complexity of each path, facilitating hardware implementation and improving the algorithm's convergence speed and stability, avoiding the problems of easy divergence or slow convergence in full-band adaptive algorithms. On the other hand, the combination of fixed filters and sub-band structures allows for flexible tracking of noise changes in different frequency bands while maintaining system robustness, achieving efficient and stable noise reduction for broadband noise.
[0069] In one embodiment, for each subband, a subband processing path is constructed by fitting based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function. This includes: for each subband, determining the system transfer function of the noise reduction system based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function; constructing a target cost function based on the distance between the system transfer function and a preset target transfer function for the noise reduction system; and solving for a set of filter coefficients consisting of L cascaded second-order section filters, using multiple sets of second primary path transfer functions, second secondary path transfer functions, and second feedback path transfer functions under different scenarios as inputs, with minimizing the target cost function as the optimization objective, to construct the subband processing path corresponding to the subband. The subband processing path consists of at least L cascaded second-order section filters.
[0070] Where L is a positive integer greater than 1. Different scenarios include at least different wearing angles and fits of the device.
[0071] In one embodiment, the system transfer function It can be represented as:
[0072] ;
[0073] in, This represents the transfer function of the first primary path; This represents the first-order path transfer function; Represents the transfer function of the first feedback path; This represents the total frequency response of the fitted filter bank; Normalized digital angular frequency This indicates sampling in the frequency domain; This represents the coefficients of a cascaded SOS (second-order section filter).
[0074] In the above embodiments, each subband takes the transfer function under multiple scenarios as input, and solves for a set of subband processing paths consisting of L cascaded second-order section filters by minimizing the distance between the system transfer function and the objective function. The cascaded second-order section structure has high numerical stability, is insensitive to coefficient quantization, and is easy to implement in hardware. Multi-scenario optimization enables the filter to adapt to different noise environments, balancing flexibility and robustness, and avoiding the problem of limited noise reduction capability of a single fixed filter.
[0075] In one embodiment, constructing a target cost function based on the distance between the system transfer function and a preset target transfer function for the noise reduction system includes: sampling the frequency domain that simultaneously serves the system transfer function and the preset target transfer function for the noise reduction system to obtain multiple sampling points; determining the weighting factors corresponding to each of the multiple sampling points based on the target transfer function; wherein sampling points obtained in frequency bands with more peaks and valleys in the target transfer function have higher weighting factors; for each sampling point, constructing an initial cost function corresponding to the sampling point based on the distance between the system transfer function and the target transfer function; and weighting and summing the initial cost functions corresponding to the multiple sampling points based on the weighting factors corresponding to each of the multiple sampling points to obtain the target cost function.
[0076] In one embodiment, the optimization objective is to adjust The value of should be chosen to minimize the system transfer function. Transfer function with target The difference between them can be used to construct a cost function. Assess this gap. The primary and secondary pathways may change depending on the usage scenario, so it's necessary to uniformly consider the fitting results under different scenarios (e.g., different wearing angles and fits of the device). Assuming there are M different scenarios, corresponding to M different combinations of the second primary pathway transfer function, the second secondary pathway transfer function, and the second feedback pathway transfer function, the overall objective cost function should consider the errors of all scenarios together. Sampling is performed in the frequency domain, totaling N points. Summation is used instead of integration. Unlike traditional solutions that consider all frequency points together, this solution introduces a weighting factor. The importance of different frequency points is adjusted to avoid the final fitting effect being affected by excessively drastic local changes or partial frequency band measurement errors in the second primary path transfer function, the second secondary path transfer function, and the second feedback path transfer function. The final objective cost function... It can be represented as:
[0077] ;
[0078] in, Represents the initial cost function; represents the weighting factor; M represents the M scenarios; N represents the N sampling points.
[0079] In the above embodiments, denser sampling points and higher weighting factors are set in frequency bands with more peaks and valleys in the target transfer function, so that the cost function pays more attention to the noise reduction accuracy of the volatile frequency bands. The filter design is guided by a weighted approach, which prioritizes the noise reduction performance of key frequency bands, improves subjective listening experience and the overall noise reduction effect of the system, and avoids local performance deficiencies caused by equal optimization across the entire frequency band.
[0080] In one embodiment, sampling the frequency domain that simultaneously serves the system transfer function and the target transfer function preset for the noise reduction system to obtain multiple sampling points includes: determining a sampling strategy based on the target transfer function; wherein the sampling strategy is used to ensure that there are more sampling points in the frequency band with more peaks and valleys in the target transfer function; and sampling the frequency domain that simultaneously serves the system transfer function and the target transfer function preset for the noise reduction system based on the sampling strategy to obtain multiple sampling points.
[0081] In the above embodiments, the sampling strategy is adaptively determined based on the spectral characteristics of the target transfer function, and the sampling points are automatically densified in frequency bands with more peaks and valleys. This focuses on key frequency bands with fewer computational resources, improving the fitting accuracy between the system transfer function and the target function, avoiding resource waste or local underfitting caused by uniform sampling, and achieving efficient and accurate noise reduction performance.
[0082] In one embodiment, using multiple sets of second primary path transfer functions, second secondary path transfer functions, and second feedback path transfer functions under different scenarios as inputs, and minimizing the objective cost function as the optimization objective, a set of filter coefficients consisting of L cascaded second-order section filters is solved to construct the subband processing path corresponding to the subband. This includes: using multiple sets of second primary path transfer functions, second secondary path transfer functions, and second feedback path transfer functions under different scenarios as inputs, and minimizing the objective cost function as the optimization objective, solving for a set of filter coefficients consisting of L cascaded second-order section filters to construct the second-order section filter bank corresponding to the subband; introducing an all-pass filter for delay adjustment into the second-order section filter bank to construct the subband processing path corresponding to the subband; wherein, the subband processing path consists of an all-pass filter and L cascaded second-order section filters.
[0083] In the above embodiments, an all-pass filter is introduced on the basis of the second-order filter bank for delay adjustment. The all-pass filter can independently adjust the phase delay without changing the amplitude-frequency characteristics, making up for the shortcomings of pure second-order cascade in phase matching, so that the sub-band processing path can more accurately approximate the target transfer function, thereby improving the stability and noise suppression effect of the noise reduction system.
[0084] In one embodiment, for each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal. This includes: for each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the initial sub-band control signal corresponding to the sub-band signal; based on the frequency response characteristics of the loudspeaker in the noise reduction system, the initial sub-band control signal is subjected to frequency response equalization processing to obtain the sub-band control signal corresponding to the sub-band signal.
[0085] In the above embodiments, after obtaining the initial sub-band control signal, frequency response equalization processing is further performed based on the speaker's frequency response characteristics. This compensates for the speaker's own nonlinear response in different frequency bands, eliminates amplitude-frequency distortion introduced by the hardware, and enables each sub-band noise reduction signal to more accurately reproduce the target noise reduction amount, thereby improving the overall noise reduction system's output sound quality and noise reduction consistency.
[0086] In one embodiment, based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain a target control signal. This includes: inputting the original audio signal into a trained neural network model to extract the signal characteristics of the original audio signal through the neural network model, and determining the weighting coefficients and combination strategies corresponding to each of the N sub-band signals based on the signal characteristics; and combining the sub-band control signals corresponding to each of the N sub-band signals based on the weighting coefficients and combination strategies to obtain the target control signal.
[0087] In the above embodiments, a trained neural network model is used to extract features from the original audio signal, and the weighting coefficients and combination strategies of each sub-band signal are adaptively determined. Compared with a fixed combination method, the contribution of sub-bands under different noise scenarios can be dynamically matched, improving the fusion accuracy and robustness, avoiding phase mismatch or energy distortion introduced by sub-band splicing, and enhancing the intelligence and adaptability of the noise reduction system.
[0088] In one embodiment, N sub-band signals are input in parallel to the sub-band processing paths corresponding to the N sub-band signals to perform noise reduction processing in parallel.
[0089] In the above embodiments, N sub-band signals are input in parallel to their respective sub-band processing paths, and noise reduction is performed simultaneously. This significantly reduces the overall computational latency and meets the requirements for real-time noise reduction. Each sub-band is calculated independently in parallel, avoiding the accumulation of errors in serial processing, improving system response speed and processing efficiency, and facilitating hardware implementation.
[0090] In one embodiment, such as Figure 4As shown, N frequency-division filters A(z) are designed to divide the first primary path transfer function into N sub-bands (i.e., narrowbands). During the offline identification stage, the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function of the noise reduction system are measured. The full-band described by the first primary path transfer function is decomposed using the frequency-division filters to obtain N sub-bands. For each sub-band, a sub-band processing path is constructed based on the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function. The sub-band processing path consists of a full-pass filter AP(z) for delay adjustment and L second-order segmented filters (SOS) cascaded together. In the online noise reduction stage, the original audio signal to be denoised is acquired through a reference microphone in the noise reduction system and processed by an ADC (analog-to-digital converter). A frequency-division filter is used to decompose the original audio signal into N sub-band signals. For each sub-band signal, the sub-band signal is input into the corresponding sub-band processing path for noise reduction processing to obtain the corresponding sub-band control signal. The original audio signal is then input into a trained neural network model to extract the signal features of the original audio signal. Based on the signal features, the weighting coefficients g and combination strategies corresponding to the N sub-band signals are determined. Based on the weighting coefficients and combination strategies corresponding to the N sub-band signals, the sub-band control signals corresponding to the N sub-band signals are combined to obtain the target control signal, which is then processed by a DAC (digital-to-analog converter) and output to the speaker for playback. Since the filter output includes feedback, directly applying the weighting coefficients may cause discontinuities in the output target control signal, resulting in unnecessary noise. A smoothing algorithm can be used when switching weighting coefficients to ensure that the output target control signal remains continuous.
[0091] In one embodiment, Figure 5 This is a schematic diagram comparing the subband optimized splicing results and the traditional broadband optimization results of this application. Figure 6 This diagram illustrates a comparison between the noise reduction performance of the subband combination scheme in this application and that of the traditional adaptive scheme. Compared to traditional audio noise reduction methods, this application decomposes the entire frequency band into N subbands using a frequency-division filter. In the offline identification stage, an independent processing path is constructed for each subband, and in the online noise reduction stage, the signals of each subband are processed in parallel. On the one hand, subband decomposition reduces the order and computational complexity of each path, facilitating hardware implementation and improving the algorithm's convergence speed and stability, avoiding the problems of easy divergence or slow convergence in full-band adaptive algorithms. On the other hand, the combination of fixed filters and subband structures allows for flexible tracking of noise changes in different frequency bands while maintaining system robustness, achieving efficient and stable noise reduction for broadband noise.
[0092] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially, these steps are not necessarily executed in that order. Unless otherwise expressly 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 of the steps in the above embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.
[0093] Based on the same inventive concept, this application also provides an audio noise reduction apparatus for implementing the audio noise reduction 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 noise reduction apparatus embodiments provided below can be found in the limitations of the audio noise reduction method described above, and will not be repeated here.
[0094] In one embodiment, such as Figure 7 As shown, an audio noise reduction device 700 is provided, which specifically includes:
[0095] Measurement module 702 is used to measure the first primary path transfer function, the first intermediate path transfer function and the first feedback path transfer function of the noise reduction system during the offline identification stage.
[0096] The decomposition module 704 is used to decompose the full frequency band described by the first primary path transfer function using a frequency division filter to obtain N sub-bands;
[0097] The construction module 706 is used to construct the sub-band processing path corresponding to each sub-band by fitting based on the first primary path transfer function, the first secondary path transfer function and the first feedback path transfer function. The sub-band processing path is composed of filters with fixed parameters.
[0098] The acquisition module 708 is used to acquire the original audio signal to be denoised through the reference microphone in the noise reduction system during the online noise reduction stage.
[0099] The decomposition module 704 is also used to decompose the original audio signal into N sub-band signals using a frequency division filter;
[0100] The noise reduction module 710 is used to input the sub-band signal into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal; based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal.
[0101] In one embodiment, the construction module 706 is further configured to, for each subband, determine the system transfer function of the noise reduction system based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function; construct a target cost function based on the distance between the system transfer function and the target transfer function preset for the noise reduction system; and, using multiple sets of second primary path transfer functions, second secondary path transfer functions, and second feedback path transfer functions under different scenarios as inputs, and minimizing the target cost function as the optimization objective, solve for a set of filter coefficients composed of L cascaded second-order section filters to construct the subband processing path corresponding to the subband; wherein, the subband processing path is composed of at least L cascaded second-order section filters.
[0102] In one embodiment, the construction module 706 is further configured to sample the frequency domain that simultaneously serves the system transfer function and the target transfer function preset for the noise reduction system to obtain multiple sampling points; determine the weighting factor corresponding to each of the multiple sampling points based on the target transfer function; wherein, sampling points obtained in frequency bands with more peaks and valleys in the target transfer function have higher weighting factors; for each sampling point, construct an initial cost function corresponding to the sampling point according to the distance between the system transfer function and the target transfer function; and perform a weighted summation of the initial cost functions corresponding to the multiple sampling points based on the weighting factors corresponding to each of the multiple sampling points to obtain the target cost function.
[0103] In one embodiment, the construction module 706 is further configured to determine a sampling strategy based on the target transfer function; wherein the sampling strategy is configured to ensure that there are more sampling points in the frequency bands with more peaks and valleys in the target transfer function; based on the sampling strategy, the frequency domain that simultaneously serves the system transfer function and the target transfer function preset for the noise reduction system is sampled to obtain multiple sampling points.
[0104] In one embodiment, the construction module 706 is further configured to take multiple sets of second primary path transfer functions, second secondary path transfer functions, and second feedback path transfer functions under different scenarios as inputs, and minimize the target cost function as the optimization objective, to solve for a set of filter coefficients composed of L cascaded second-order section filters, so as to construct a second-order section filter bank corresponding to the sub-band; and to introduce an all-pass filter for delay adjustment into the second-order section filter bank to construct a sub-band processing path corresponding to the sub-band; wherein, the sub-band processing path is composed of an all-pass filter and L cascaded second-order section filters.
[0105] In one embodiment, the noise reduction module 710 is further configured to, for each sub-band signal, input the sub-band signal into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the initial sub-band control signal corresponding to the sub-band signal; and, based on the frequency response characteristics of the loudspeaker in the noise reduction system, perform frequency response equalization processing on the initial sub-band control signal to obtain the sub-band control signal corresponding to the sub-band signal.
[0106] In one embodiment, the noise reduction module 710 is further configured to input the original audio signal into a trained neural network model to extract the signal features of the original audio signal through the neural network model, and determine the weighting coefficients and combination strategies corresponding to each of the N sub-band signals based on the signal features; and combine the sub-band control signals corresponding to each of the N sub-band signals based on the weighting coefficients and combination strategies to obtain the target control signal.
[0107] In one embodiment, N sub-band signals are input in parallel to the sub-band processing paths corresponding to the N sub-band signals to perform noise reduction processing in parallel.
[0108] In the offline identification stage, the aforementioned audio noise reduction device measures the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function of the noise reduction system. A frequency-division filter is used to decompose the entire frequency band described by the first primary path transfer function, resulting in N sub-bands. For each sub-band, a sub-band processing path is constructed by fitting based on the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function. The sub-band processing path consists of filters with fixed parameters. In the online noise reduction stage, the original audio signal to be denoised is acquired through a reference microphone in the noise reduction system, and a frequency-division filter is used to decompose the original audio signal into N sub-band signals. For each sub-band signal, the sub-band signal is input into the corresponding sub-band processing path for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal. Based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal. Compared to traditional audio noise reduction methods, this application decomposes the entire frequency band into N sub-bands using a frequency-division filter. In the offline identification stage, an independent processing path is constructed for each sub-band, and in the online noise reduction stage, the signals of each sub-band are processed in parallel. On the one hand, sub-band decomposition reduces the order and computational complexity of each path, facilitating hardware implementation and improving the algorithm's convergence speed and stability, avoiding the problems of easy divergence or slow convergence in full-band adaptive algorithms. On the other hand, the combination of fixed filters and sub-band structures allows for flexible tracking of noise changes in different frequency bands while maintaining system robustness, achieving efficient and stable noise reduction for broadband noise.
[0109] Each module in the aforementioned audio noise reduction 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 corresponding operations of each module.
[0110] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. 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 input / output interfaces are used for exchanging information between the processor and external devices. 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 the computer program is executed by the processor, it implements an audio noise reduction method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0111] Those skilled in the art will understand that Figure 8 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.
[0112] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.
[0113] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.
[0114] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.
[0115] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0116] 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 memory 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, artificial intelligence (AI) processors, etc., and are not limited to these.
[0117] 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 application.
[0118] 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 noise reduction method, characterized by, The method includes: During the offline identification phase, the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function of the noise reduction system are measured. The full-band described by the first primary path transfer function is decomposed using a frequency divider filter to obtain N sub-bands; For each sub-band, based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, a sub-band processing path corresponding to the sub-band is constructed by fitting, wherein the sub-band processing path is composed of filters with fixed parameters; During the online noise reduction stage, the original audio signal to be denoised is acquired through the reference microphone in the noise reduction system, and the original audio signal is decomposed into N sub-band signals using the frequency division filter. For each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal; Based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to each of the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal.
2. The method of claim 1, wherein, For each sub-band, based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, the sub-band processing path corresponding to the sub-band is constructed by fitting, including: For each sub-band, the system transfer function of the noise reduction system is determined based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function. Construct a target cost function based on the distance between the system transfer function and the target transfer function preset for the noise reduction system; Using the second primary path transfer function, the second secondary path transfer function, and the second feedback path transfer function under multiple different scenarios as inputs, and minimizing the target cost function as the optimization objective, a set of filter coefficients consisting of L cascaded second-order section filters is solved to construct the subband processing path corresponding to the subband; wherein, the subband processing path is composed of at least the L cascaded second-order section filters.
3. The method according to claim 2, characterized in that, The step of constructing a target cost function based on the distance between the system transfer function and a preset target transfer function for the noise reduction system includes: Multiple sampling points are obtained by sampling the frequency domain of both the system transfer function and the target transfer function preset for the noise reduction system. The weighting factor corresponding to each of the plurality of sampling points is determined based on the target transfer function; wherein, the sampling points obtained by sampling in the frequency band with more peaks and valleys in the target transfer function have a higher weighting factor. For each sampling point, an initial cost function corresponding to the sampling point is constructed based on the distance between the system transfer function and the target transfer function; Based on the weight factors corresponding to each of the multiple sampling points, the initial cost functions corresponding to each of the multiple sampling points are weighted and summed to obtain the target cost function.
4. The method according to claim 3, characterized in that, The sampling of the frequency domain, which simultaneously serves the system transfer function and the target transfer function preset for the noise reduction system, yields multiple sampling points, including: A sampling strategy is determined based on the target transfer function; wherein the sampling strategy is used to ensure that there are more sampling points in the frequency bands where the target transfer function has many peaks and valleys; Based on the sampling strategy, the frequency domain of both the system transfer function and the target transfer function preset for the noise reduction system is sampled to obtain multiple sampling points.
5. The method according to claim 2, characterized in that, The process involves taking the transfer functions of the second primary path, the second secondary path, and the second feedback path under multiple different scenarios as inputs, and minimizing the target cost function as the optimization objective. A set of filter coefficients, consisting of L cascaded second-order section filters, is then obtained to construct the sub-band processing path corresponding to the sub-band. This includes: Using the second primary path transfer function, the second secondary path transfer function, and the second feedback path transfer function under multiple different scenarios as inputs, and minimizing the target cost function as the optimization objective, a set of filter coefficients consisting of L cascaded second-order section filters is solved to construct the second-order section filter bank corresponding to the sub-band. An all-pass filter for delay adjustment will be introduced into the second-order section filter bank to construct the sub-band processing path corresponding to the sub-band; wherein, the sub-band processing path is composed of the all-pass filter and the L second-order section filters cascaded together.
6. The method according to claim 1, characterized in that, For each sub-band signal, the sub-band signal is input into the corresponding sub-band processing path for noise reduction processing to obtain the corresponding sub-band control signal, including: For each sub-band signal, the sub-band signal is input into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the initial sub-band control signal corresponding to the sub-band signal; Based on the frequency response characteristics of the loudspeaker in the noise reduction system, the initial sub-band control signal is subjected to frequency response equalization processing to obtain the sub-band control signal corresponding to the sub-band signal.
7. The method according to claim 1, characterized in that, The method of combining the sub-band control signals corresponding to the N sub-band signals according to the signal characteristics of the original audio signal, using dynamic weighting coefficients, to obtain the target control signal includes: The original audio signal is input into a trained neural network model to extract the signal features of the original audio signal through the neural network model, and the weighting coefficients and combination strategies corresponding to the N sub-band signals are determined based on the signal features. Based on the weighting coefficients and combination strategies corresponding to the N sub-band signals, the sub-band control signals corresponding to the N sub-band signals are combined to obtain the target control signal.
8. The method according to claim 1, characterized in that, The N sub-band signals are input in parallel to their respective sub-band processing paths for parallel noise reduction processing.
9. An audio noise reduction device, characterized in that, The device includes: The measurement module is used to measure the first primary path transfer function, the first intermediate path transfer function, and the first feedback path transfer function of the noise reduction system during the offline identification phase. The decomposition module is used to decompose the full frequency band described by the first primary path transfer function using a frequency division filter to obtain N sub-bands; The construction module is used to construct a sub-band processing path corresponding to each sub-band by fitting based on the first primary path transfer function, the first secondary path transfer function, and the first feedback path transfer function, wherein the sub-band processing path is composed of filters with fixed parameters. The acquisition module is used to acquire the original audio signal to be denoised through the reference microphone in the noise reduction system during the online noise reduction stage. The decomposition module is also used to decompose the original audio signal into N sub-band signals using the frequency division filter; The noise reduction module is used to input the sub-band signal into the sub-band processing path corresponding to the sub-band signal for noise reduction processing to obtain the sub-band control signal corresponding to the sub-band signal; based on the signal characteristics of the original audio signal, the sub-band control signals corresponding to the N sub-band signals are combined according to dynamic weighting coefficients to obtain the target control signal.
10. 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 method according to any one of claims 1 to 8.
11. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.
12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.