An ANC system for headphones, noise reduction method, and storage medium

Abstract

The application relates to an ANC system, a noise reduction method and a storage medium for earphones. The ANC system comprises a system on chip, and the system on chip comprises a processor configured to: acquire a noise-containing signal collected by a feedforward microphone; filter the noise-containing signal by using a trained neural network filter comprising an ESN to output a filtered first signal, wherein the ESN is retrained under the condition that a first preset condition is met, and the first preset condition at least comprises that a wearing manner of the earphone changes or a leakage amount changes during use of the earphone; and generate a noise cancellation signal based on the first signal, wherein the noise cancellation signal is used for playing through a loudspeaker to cancel residual noise in the earphone. In this way, the noise reduction bandwidth and the noise reduction amount of the ANC system can be improved, so that a better noise reduction effect can be achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of active noise reduction, and more particularly, to an ANC system for earphones, a noise reduction method and a storage medium. BACKGROUND

[0002] With the development of technology, active noise reduction earphones are widely used. However, when the active noise reduction technology is applied in earphone products, an opposite phase cancellation signal is often generated to achieve the effect of noise reduction. In these scenarios, a linear filter is generally used to fit the main path system function of sound transmission from the outside to the ear canal, but the earphone, microphone and loudspeaker are not ideal linear acquisition and playback devices. Since the linear filter cannot fit the non-linear effects of physical devices, the noise reduction bandwidth and noise reduction amount are limited, affecting the user experience.

[0003] The existing method generally solves the above problems by optimizing the passive noise reduction curve of the earphone itself and the linearity of the loudspeaker and microphone. The main method is to reduce the nonlinearity of the system device and the mutation in the frequency spectrum and phase as much as possible through a large number of tests and experiments. However, this method puts high requirements on the earphone cavity design. Moreover, for the ANC system, there are still some problems with the method of optimizing the earphone cavity, such as a large amount of testing and modification, making it very difficult to optimize the nonlinearity of the cavity, and the optimized earphone cannot improve the music quality, such as being able to use only a single speaker and not being able to use a double speaker design to improve high-frequency sound quality. In addition, in the existing filter processing process using a filter containing an RNN, due to the large size of the RNN network, the resource occupation is high, the complexity is high, it is difficult to perform online training, and the data sampling rate during ANC operation is also relatively high. The power consumption of the RNN scheme is too large, which limits its application in earphones with high requirements for power consumption and resources. SUMMARY

[0004] The present application is provided to solve the above problems in the prior art. There is a need for an ANC system for earphones, a noise reduction method and a storage medium, which can quickly and efficiently perform adaptive filtering processing, and can avoid excessive workload of filtering processing to reduce the listening quality of users.

[0005] According to a first aspect of the present application, an ANC system for a headset is provided, the ANC system comprising a system on chip, the system on chip comprising a processor configured to: obtain a noise-containing signal collected by a feed-forward microphone; filter the noise-containing signal using a trained neural network filter comprising an echo state network (ESN) to output a filtered first signal, the ESN being retrained when a first preset condition is met, wherein the first preset condition at least comprises that a wearing manner or a leakage amount of the headset changes during use; and generate a noise cancellation signal based on the first signal, the noise cancellation signal being used to be played via a speaker to cancel residual noise in the headset.

[0006] According to a second aspect of the present application, a noise reduction method of an ANC system is provided, comprising: obtaining a noise-containing signal collected by a feed-forward microphone; filtering the noise-containing signal using a trained neural network filter comprising an echo state network (ESN) to output a filtered first signal, the ESN being retrained when a first preset condition is met, wherein the first preset condition at least comprises that a wearing manner or a leakage amount of the headset changes during use; and generating a noise cancellation signal based on the first signal, the noise cancellation signal being used to be played via a speaker to cancel residual noise in the headset.

[0007] According to a third aspect of the present application, a computer readable storage medium is provided, the computer readable storage medium storing computer program instructions, the computer program instructions causing a processor to execute the noise reduction method according to the embodiments of the present application when the computer program instructions are run by the processor.

[0008] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0009] The ANC system for a headset provided in the embodiments of the present application filters the noise-containing signal using a trained neural network filter comprising an echo state network (ESN) to output a filtered first signal, which can simultaneously fit linear components and nonlinear components in the ANC system, without optimizing the headset cavity, and can improve the noise reduction bandwidth and noise reduction amount. Moreover, the ESN uses simple linear regression to train output weights, and has much lower network size and complexity than RNN and LSTM, can be trained online, can achieve fast convergence, and improves the training efficiency. In addition, the ESN is retrained when a first preset condition is met, thereby realizing online training, enabling the neural network filter to automatically adjust network parameters according to the state of the headset, thereby improving the filtering effect, and improving the noise reduction bandwidth and noise reduction amount.

[0010] The above description is only a summary of the technical solutions of the present application. In order to enable one skilled in the art to better understand the technical means of the present application, the contents of the specification can be implemented, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more obvious and easy to understand, the specific embodiments of the present application are described below. BRIEF DESCRIPTION OF DRAWINGS

[0011] In the drawings, which are not necessarily drawn to scale, like numerals can describe similar components in different views. The drawings are intended to illustrate various embodiments by example, and are not intended to be limiting. The same reference numerals in different drawings can identify the same or similar elements. Such embodiments are illustrative rather than restrictive, and are not intended to be exhaustive or exclusive.

[0012] FIG. 1(a) shows a schematic diagram of an ANC system for headphones according to an embodiment of the present application.

[0013] FIG. 1(b) shows a schematic diagram of neural network training according to an embodiment of the present application.

[0014] FIG. 1(c) shows a schematic diagram of an ESN network structure according to an embodiment of the present application.

[0015] FIG. 1(d) shows a schematic diagram of ESN network training steps according to an embodiment of the present application.

[0016] FIG. 1(e) shows a schematic diagram of active noise reduction by an ANC system for headphones according to an embodiment of the present application.

[0017] FIG. 2(a) shows another schematic diagram of active noise reduction by an ANC system for headphones according to an embodiment of the present application.

[0018] FIG. 2(b) shows another schematic diagram of neural network training according to an embodiment of the present application.

[0019] Figure 3 FIG. 3 shows yet another schematic diagram of active noise reduction by an ANC system for headphones according to an embodiment of the present application.

[0020] Figure 4 FIG. 4 shows one of the schematic diagrams of active noise reduction by an ANC system for headphones according to an embodiment of the present application.

[0021] Figure 5 FIG. 5 shows a flowchart of a noise reduction method of an ANC system according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] For those skilled in the art to better understand the technical solutions of the present application, the present application will be described in detail below in combination with the drawings and specific embodiments. The embodiments of the present application will be described in further detail below in combination with the drawings and specific embodiments, but not as a limitation of the present application. The various steps described herein should not be considered as limiting the order described herein as an example, provided that the order can be adjusted by those skilled in the art, as long as the logic between them is not destroyed, resulting in the whole process cannot be implemented.

[0023] The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish. "Include" or "contain" and similar words mean that the elements before the word cover the elements listed after the word, and do not exclude the possibility of also covering other elements. In the present application, the arrows shown in the figure are only used as an example of the execution order, not as a limitation, and the technical solutions of the present application are not limited to the execution order described in the embodiments. The steps in the execution order can be combined, can be decomposed, can be exchanged in order, as long as the logical relationship of the execution content is not affected.

[0024] All terms used in the present application (including technical terms or scientific terms) have the same meaning as understood by those skilled in the art to which the present application belongs, unless otherwise specifically defined. It should also be understood that terms defined in, such as general dictionaries, should be interpreted to have meanings consistent with their meanings in the context of the relevant art, and should not be interpreted in an idealized or excessively formalized sense, unless otherwise defined herein. The technology, methods and devices known to those skilled in the relevant art can not be discussed in detail, but in appropriate cases, the technology, methods and devices should be considered as part of the specification.

[0025] Figure 1(a) shows a structural schematic diagram of an ANC system for headphones according to an embodiment of the present application. Figure 1(b) shows a schematic diagram of active noise reduction of an ANC system for headphones according to an embodiment of the present application.

[0026] According to some embodiments of the present application, an ANC system 100 for a headphone is provided, which includes a system on chip 101 including a processor 102. The processor 102 is configured to obtain a noise-containing signal collected by a feed-forward microphone 103, filter the noise-containing signal using a neural network filter 104 including an echo state network (ESN) to output a first filtered signal, the ESN being retrained when a first predetermined condition is met, wherein the first predetermined condition includes at least a change in a wearing manner or a leakage amount of the headphone during use. A noise cancellation signal is generated based on the first signal, which is used to be played via a speaker 105 to cancel residual noise in the ear of the headphone. Note that in the present application, various components, such as the processor 102 shown in FIG. 1(a), can be implemented by a SOC (system on chip), such as the system on chip 101. For example, various RISC (reduced instruction set computer) processor IPs purchased from ARM Corporation and the like can be used as the processor 102 of the SOC to perform corresponding functions, so that an embedded system can be implemented. Specifically, there are many modules on commercially available IPs, such as but not limited to memory, cache, and the like. In some embodiments, chip manufacturers can also develop customized versions of these modules on existing IPs. In addition, other components such as limiters, speakers, microphones, and the like can be externally connected to the IPs. Users can construct an ASIC (application specific integrated circuit) based on the purchased IPs or self-developed modules to construct the ANC system 100, so as to reduce power consumption and cost.

[0027] The processor 102 can be a processing device including one or more general- purpose processing devices, such as microprocessors, central processing units (CPUs), graphics processing units (GPUs), and the like. More particularly, the processor 102 can be a complex instruction set computing (CISC) microprocessor, reduced instruction set computing (RISC) microprocessor, very long instruction word (VLIW) microprocessor, processor running other instruction sets, or processors running a combination of instruction sets. The processor 102 can also be one or more special-purpose processing devices such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), and the like.

[0028] Specifically, the method of training the ESN is shown in Fig. 1(b). In the training stage, x(n) is the signal obtained by processing the noise-containing signal collected by the feed-forward microphone 103 through the analog-to-digital converter 108, the downsampling module 109, and the echo path filter 106, y(n) is the noise-containing signal collected by the feedback microphone 107 when the ANC function is not turned on. Through the training of the ESN, the absolute value error between the output of the neural network and y(n) is minimized, and the directions are opposite. The echo path filter 106 is an estimation of the path from the speaker 105 to the feedback microphone 107. The echo path filter 106 is a path estimation filter that describes the process of sound being broadcast by the speaker 105 and then being reflected by the ear canal to the feedback microphone 107. After filtering the feed-forward microphone 103 signal through this specific function (which is called an echo path function), the signal is simulated to pass through the process on the reflection path after being broadcast by the speaker 105.

[0029] As shown in Fig. 1(c), the network structure of the ESN is input layer, reservoir, and output layer in sequence. x(n), S(n), and y(n+L) represent the input, state (reservoir state), and output of the ESN network at time n, respectively. V, R, and W represent the input weight matrix, intermediate weight matrix, and output weight matrix, respectively. Here, x(n) is x(n) in Fig. 1(b), y(n+L) is y(n+L) at time n+L in Fig. 1(b), the state updating method of the reservoir, and the output of the network are as follows:

[0030] S(n) = tanh(RS(n-1) + Vx(n))

[0031] y(n+L) = WS(n)

[0032] Here, the network input data x(n) obtained by processing the feed-forward microphone 103 through the analog-to-digital converter 108, the downsampling module 109, and the echo path filter 106 at time n is used to predict the data y(n+L) collected by the feedback microphone 107 at time n+L after being processed by the downsampling module 109, which utilizes the prediction ability of the ESN network.

[0033] Figure 1(d) is the training step of the ESN network. First, the size of the reservoir is determined, i.e. the number of neurons. The more nodes, the stronger the fitting ability. Next, a random connection matrix is generated, which represents which neurons are connected, as well as the direction and weight of the connection, i.e. the matrix R in Figure 1(c). The next step is normalization, which can directly use a scaling factor. Due to the influence of the adopted activation function tanh, the weight is usually initialized to a value between 0 and 1. Finally, the input weight V and the output weight W are randomly generated. These two parameters will affect the length of the network's short-term memory. The smaller the input weight and the closer the spectral radius of the internal matrix to 1, the longer the network's short-term memory.

[0034] Next, the training is performed, the "idle" process in Figure 1(d), which is actually the initialization of the state of the reservoir. Because the internal connection of the reservoir is random, the noise of the reservoir state obtained by the initial input sequence will be large, so some data is used to initialize the state of the reservoir to reduce the influence of noise.

[0035] The last step of training is to determine the output weight W using linear regression. Here it is assumed that the 2-norm regularization is performed on the output weight W, and the regularization coefficient is λ. The network state matrix is S, and the output sequence matrix is Y. The optimization objective is as follows:

[0036]

[0037] Taking the derivative of the above equation and setting it to 0, based on the ridge regression least squares method, the solution is

[0038] W = YS T (SS T + λI) -1

[0039] From the training process, it can be seen that the ESN network determines the output weight W by linear regression least squares method. The training step is simple, and the computational complexity is much lower than that of RNN, LSTM and other networks, which is very suitable for online training and updating network parameters.

[0040] As shown in FIG. 1(e), the processor 102 acquires the noise signal collected by the feed-forward microphone 103 from the environment, and obtains the processed noise signal via the analog-to-digital converter 108 and the down-sampling module 109. The noise signal is filtered by the trained neural network filter 104 containing an echo state network (ESN) to fit the nonlinear component in the ANC system 100, so as to obtain a first signal. The prediction function of the ESN network can be effectively utilized to compensate for the delay in the feed-forward ANC path, so that the ANC path can have a wider noise reduction bandwidth, thereby improving the effect of active noise reduction. The first signal is processed by the up-sampling module 110 to generate a noise cancellation signal, and is played via the loudspeaker 105 to cancel the residual noise in the ear. The ESN neural network has high prediction capability, which can not only compensate for the system delay, but also realize online training, so that the filter parameters can be updated online. In this way, a better active noise reduction effect can be achieved without special cavity optimization of the earphone.

[0041] In some embodiments of the present application, the processor is further configured to, after acquiring the noise signal, fit the noise signal by using a preset IIR (Infinite Impulse Response) or FIR (Finite Impulse Response) linear filter, and then filter the fitted noise signal by using a trained neural network filter containing an echo state network (ESN) to output a filtered second signal, and generate a noise cancellation signal based on the second signal. As shown in FIG. 2(a), the feed-forward microphone 201 collects the noise signal in the environment, and the processed noise signal is obtained via the processing of the analog-to-digital converter 202 and the down-sampling module 203. Then, the processed noise signal is linearly fitted by using the preset IIR or FIR linear filter 204, which is conducive to reducing the size of the neural network. The noise signal after linear fitting is then subjected to nonlinear fitting by using the trained filter 205 containing an ESN, so that the filter 205 containing an ESN only fits the nonlinear part of the ANC system. By introducing the filter 205 containing an ESN to fit the nonlinear component in the ANC system on the basis of the preset IIR or FIR linear filter 204, a better ANC effect can be achieved. In this embodiment, the input x(n) of the neural network training 207 in the training phase of the neural network training is the noise signal collected by the feed-forward microphone 201 and processed by the preset IIR or FIR linear filter 204 and the echo path filter 206, and y(n) is the noise signal collected by the feedback microphone 208 when the ANC function is not turned on.

[0042] In some embodiments of this application, the processor is further configured to acquire a residual noise signal collected by a feedback microphone and a third signal output via an echo path filter, adaptively iteratively update the FIR filter coefficients using the first signal, the third signal, and the residual noise signal, and filter the first signal using the FIR filter to output a filtered fourth signal, and generate a noise cancellation signal based on the fourth signal. Specifically, as... Figure 3 As shown, the feedforward microphone 301 collects noisy signals from the environment. After processing by the analog-to-digital converter 302 and the downsampling module 303, the nonlinear components in the ANC system are first fitted using a filter 304 containing an ESN to output a first signal. The feedback microphone 307 collects residual noise signals from the headphones and transmits these signals to the FIR filter 305. Simultaneously, the third signal output by the echo path filter 306 is also transmitted to the FIR filter 305. In this case, the filter coefficients of the FIR filter 305 are adaptively iterated using the first signal, the third signal, and the residual noise signal. Considering the high computational complexity of online training, the filter 304 containing the ESN can be calculated offline or its coefficients can be updated periodically (the nonlinear effect is caused by electronic devices and generally does not change in real time). However, the linear part can be updated in real time using the adaptive FIR filter 305 to update the linear coefficients that are closely related to path changes based on the headphone wearing conditions. The coefficient update process of the adaptive FIR filter 305 is as follows:

[0043] Where w(n+1) represents the coefficients of FIR filter 305 at the next time step, and w(n) represents the coefficients of FIR filter 305 at the current time step, expressed as a vector. w(n)=[w0(n),w1(n),w2(n),...,w L-1 (n)] T Let R(n) be the length of the FIR filter 305. R(n) = [r(n), r(n-1), ..., r(n-L+1)] T Let r(n) be the noisy signal collected by the feedforward microphone 301 at the current moment, processed by the downsampling module 303, the filter 304 containing the ESN, and the signal after filtering by the echo path filter 306. Let e(n) be the residual noise signal collected by the feedback microphone 307 at the current moment. μ is the iteration step size coefficient, which is a constant. n is the current sampling time. In this way, not only can the noise reduction bandwidth and noise reduction amount be improved, but also linear fitting can be performed in real time according to the wearing situation of the headphones, further improving the noise reduction effect.

[0044] In some embodiments of the present application, the processor is further configured to perform adaptive iterative updating of FIR filter coefficients using the second signal, the third signal and the noise residual signal, and perform filtering processing on the second signal using the FIR filter to output a filtered fifth signal, and generate a noise cancellation signal based on the fifth signal, thereby further improving the noise reduction effect. Specifically, as shown in Figure 4 Similarly to the above, the feedforward microphone 401 collects the noise signal in the external environment, and after processing by the analog-to-digital converter 402 and the down-sampling module 403, the processed noise signal is filtered using the preset IIR linear filter 404, thereby facilitating reduction of the computational complexity of real-time adaptation. The filtered signal is input to the filter 405 containing the ESN for nonlinear fitting processing to output a filtered second signal, thereby realizing linear fitting of the linear component in the ANC system in advance. The feedback microphone 410 collects the noise residual signal in the earphone, and transmits the noise residual signal to the FIR filter 406. At the same time, the third signal output by the echo path filter 407 is also transmitted to the FIR filter 406. At the same time, the FIR filter 406 is subjected to adaptive iterative updating of filter coefficients using the second signal, the third signal and the noise residual signal, and the second signal is filtered using the FIR filter to output a filtered fifth signal. After processing by the up-sampling module 408, a noise cancellation signal is generated, and the cancellation signal is played via the loudspeaker 409 to cancel the residual noise in the earphone. The linear part adopts a combination scheme of the preset IIR filter 404 and the adaptive FIR filter 406, and the linear coefficient part closely related to the path change can be updated in real time according to the earphone wearing condition.

[0045] The prediction function of the ESN network can be used to compensate for the delay in the feedforward ANC path. The noise signal is input to the ANC system, first subjected to delay compensation and nonlinear compensation by the filter 405 containing the ESN, subjected to linear filtering by the preset IIR linear filter 404, and subjected to adaptive linear filtering by the adaptive FIR filter 406, thereby reducing the complexity of network calculation of the filter 405 containing the ESN. The ESN network for delay compensation can be trained offline, online or periodically updated. The combination of the linear filter and the filter containing the neural network improves the fitting accuracy of the ANC system, reduces the size of the neural network as much as possible, saves the computational complexity, and improves the noise reduction bandwidth and noise reduction amount of the ANC system. The prediction function of the neural network with the prediction function is used to predict the noise reduction data, which can also compensate for the system processing delay and further improve the noise reduction effect.

[0046] In some embodiments of the present application, the in-ear state of the earphone is obtained, and the ESN is retrained locally if the wearing manner changes or the leakage amount changes when the earphone is in the in-ear state. For example, when it is detected that the earphone has the action of being in-ear or has been in-ear, if the wearing manner changes or the leakage amount changes, the ESN can be retrained locally at this time to update the filtering parameters of the adaptive filter, so as to achieve better noise reduction effect. The method for detecting the wearing manner or the leakage amount after the earphone is in-ear is not limited, and the method disclosed in the prior art can be used to perform.

[0047] Figure 5 A flowchart of a noise reduction method of an ANC system according to an embodiment of the present application is shown. In step S501, a noise-containing signal collected by a feedforward microphone is obtained. In step S502, the noise-containing signal is filtered by using a trained neural network filter containing an ESN to output a filtered first signal, and the ESN is retrained under the condition that a first preset condition is met, wherein the first preset condition at least includes that the wearing manner changes or the leakage amount changes during use of the earphone. In step S503, a noise cancellation signal is generated based on the first signal, and the noise cancellation signal is used to play via a loudspeaker to cancel residual noise in the ear of the earphone.

[0048] In some embodiments of the present application, after the noise-containing signal is obtained, the noise-containing signal is fitted by using a preset IIR or FIR linear filter, and the fitted noise-containing signal is filtered by using a trained neural network filter containing an echo state network (ESN) to output a filtered second signal, and a noise cancellation signal is generated based on the second signal.

[0049] In some embodiments of the present application, a noise residual signal collected by a feedback microphone and a third signal output via an echo path filter are obtained; the FIR filter coefficients are adaptively and iteratively updated by using the first signal, the third signal and the noise residual signal, and the first signal is filtered by using the FIR filter to output a filtered fourth signal, and a noise cancellation signal is generated based on the fourth signal.

[0050] In some embodiments of the present application, the FIR filter coefficients are adaptively and iteratively updated by using the second signal, the third signal and the noise residual signal, and the second signal is filtered by using the FIR filter to output a filtered fifth signal, and a noise cancellation signal is generated based on the fifth signal.

[0051] In some embodiments of the present application, a computer readable storage medium is provided, having stored thereon computer program instructions, which, when executed by a processor, cause the processor to perform the method of noise reduction as described in various embodiments of the present application. Implementations of such methods can include software code, e.g., microcode, assembly language code, a higher-level languages code, etc. Various programs or program modules can be created using a variety of software programming techniques. For example, program portions or program modules can be designed in, or with the aid of, Java, Python, C, C++, assembly language, or any known programming language. One or more of such software portions or modules can be integrated into a computer system and / or computer readable medium. Such software code can include computer readable instructions for performing various methods. The software code can form part of a computer program product or a computer program module. Moreover, in examples, the software code can be tangibly stored on one or more volatile, non-transitory, or non-volatile tangible computer-readable media, such as during execution or at other times. Examples of these tangible computer-readable media can include, but are not limited to, hard disks, removable magnetic disks, removable optical disks (e.g., optical disks and Blu-ray disks), magnetic cassettes, memory cards or sticks, random access memories (RAMs), read only memories (ROMs), and the like.

[0052] Furthermore, although exemplary embodiments have been described herein, the scope of their protection is to be understood as including any and all embodiments which have equivalent elements, modifications, omissions, combinations (e.g., of

[0053] The above description is intended to be illustrative, and not restrictive. For example, the above-described examples (or one or more aspects thereof) can be used in combination with each other. Other examples can use, to one of ordinary skill in the art, upon reading the above description. In addition, in the above detailed description, various features can be grouped together in one or more embodiments for simplicity. This should not be interpreted as a requirement that the claimed subject matter must include such features. Rather, the subject matter described herein can be practiced without any of these specific details. In other instances, well-known structures have not been described in detail to avoid obscuring the subject matter. It is therefore intended that the scope of the application be determined by the following claims and their equivalents.

[0054] The above embodiments are only exemplary embodiments of the present application, and are not intended to limit the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application, and such modifications or equivalent replacements should also be considered to fall within the protection scope of the present application.

Claims

1. An ANC system for headphones, characterized in that, The ANC system includes a system-on-a-chip, which includes a processor. The processor is configured to: The noisy signal is acquired by the feedforward microphone. After the noisy signal is acquired, it is fitted using a preset IIR or FIR linear filter. The noisy signal is fitted using a trained neural network filter containing an echo state network (ESN) to output a filtered second signal. The ESN is retrained under the condition that a first preset condition is met to achieve online training. The first preset condition includes at least the following: the wearing method or leakage changes during the use of the headphones. A noise cancellation signal is generated based on the second signal, and the noise cancellation signal is played via a speaker to cancel residual noise in the ear of the headphones.

2. The ANC system according to claim 1, characterized in that, The processor is further configured to: acquire a residual noise signal collected by a feedback microphone and a third signal output via an echo path filter; adaptively iterate and update the FIR filter coefficients using the second signal, the third signal and the residual noise signal, and filter the second signal using the FIR filter to output a filtered fifth signal, and generate a noise cancellation signal based on the fifth signal.

3. The ANC system according to claim 1, characterized in that, The system obtains the in-ear and out-of-ear status of the earphones. When the earphones are in the in-ear state and the wearing method changes or the leakage changes, the ESN is retrained locally.

4. A noise reduction method for an ANC system, characterized in that, include: The noisy signal is acquired by the feedforward microphone. After the noisy signal is acquired, it is fitted using a preset IIR or FIR linear filter. The noisy signal is fitted using a trained neural network filter containing an echo state network (ESN) to output a filtered second signal. The ESN is retrained under the condition that a first preset condition is met to achieve online training. The first preset condition includes at least the following: the wearing method or leakage changes during the use of the headphones. A noise cancellation signal is generated based on the second signal, and the noise cancellation signal is used to play through a speaker to cancel the residual noise in the earphone.

5. The noise reduction method according to claim 4, characterized in that, Acquire the residual noise signal collected by the feedback microphone and the third signal output through the echo path filter; The FIR filter coefficients are adaptively iteratively updated using the second signal, the third signal, and the residual noise signal. The second signal is then filtered using the FIR filter to output the filtered fifth signal. A noise cancellation signal is generated based on the fifth signal.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, cause the processor to perform the noise reduction method as described in claim 4 or 5.

Images