A low-resource multi-channel echo suppression method, apparatus, electronic device and medium
The low-resource multi-channel echo suppression method using linear filtering and reference signal merging simplifies the computation and solves the problems of high computational cost and poor flexibility of existing algorithms, achieving effective echo suppression on low-resource devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNISOUND INFORMATION TECH CO LTD
- Filing Date
- 2023-07-26
- Publication Date
- 2026-07-21
AI Technical Summary
Existing deep learning-based multi-channel echo suppression algorithms are computationally intensive, lack flexibility, are difficult to adapt to changes in the number of channels, and have limited computing power on low-resource devices.
A low-resource, multi-channel echo suppression method is adopted. By using linear filtering and reference signal merging, the spectral gain coefficient is calculated using the processing result of one channel and reused in other channels, simplifying the computational load of the model.
It reduces computational load, meets the needs of low-resource devices, and maintains good echo suppression performance.
Smart Images
Figure CN116935877B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of echo suppression technology, and in particular to a low-resource multi-channel echo suppression method, apparatus, electronic device, and medium. Background Technology
[0002] Echo cancellation is a core issue in speech signal processing and a key processing module in voice communication and human-computer interaction. It's a technique for removing spontaneous echoes from devices. Because it's difficult to make the cavity of a pickup device or product perfectly linear, especially inexpensive products with significant non-linear components, a large amount of echo interference remains even after linear filtering. Furthermore, some sound effects settings use frequency division and equalization, which also introduce non-linear components, requiring further echo suppression. Currently, deep learning-based residual echo suppression has achieved good results.
[0003] Currently, most deep learning-based residual echo suppression algorithms fall into two categories. One is to model the model according to the number of microphone channels and the number of echo channels. This approach has the highest computational cost and is the least flexible, requiring retraining the model when the number of channels changes. The other approach is to build a single-microphone channel model, modeling each microphone channel separately. This is more flexible than multi-microphone channel modeling, but the computational cost is still very high because each microphone channel also requires the entire model to be calculated. Summary of the Invention
[0004] To address the aforementioned problems, this invention provides a low-resource multi-channel echo suppression method, apparatus, electronic device, and medium.
[0005] In a first aspect, embodiments of the present invention provide a low-resource multi-channel echo suppression method, comprising:
[0006] Acquire multi-channel microphone signals and multi-channel reference signals;
[0007] The multi-channel microphone signal and the multi-channel reference signal are fed into a linear filter to obtain the linearly filtered signal for each channel in the multi-channel array.
[0008] The multi-channel reference signal is fed into the Combine module to obtain the reference signal;
[0009] The linearly filtered signal of the first channel is combined with the reference signal and then fed into the model to obtain the estimated clean signal of the first channel.
[0010] The spectral gain coefficient is determined based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel.
[0011] The spectral gain coefficients are multiplexed into the other channels of the multi-channel system, excluding the first channel, to determine the clean signals of the other channels.
[0012] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the linear filter is one of NLMS, RLS, or Kalman filtering.
[0013] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the multi-channel reference signal is sent to the Combine module to obtain the reference signal by averaging the multi-channel reference signals Ref1 to RefN through the Combine module.
[0014] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the spectral gain coefficient is determined based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel using the following formula:
[0015]
[0016] Where S1(t,f) is the estimated clean signal s1 of the first channel after frame division and windowing Fourier transform, Y1(t,f) is the linearly filtered signal y1 of the first channel after frame division and windowing Fourier transform, EPS is the system minimum value to prevent division by zero, floor is the preset minimum value to avoid over-suppression, and the value is 0.02. M(t,f) is less than or equal to 1.
[0017] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, multiplexing the spectral gain coefficients to other channels besides the first channel to determine the clean signals of the other channels besides the first channel includes:
[0018] The linearly filtered signal yi of the other channels in the multi-channel system, excluding the first channel, is subjected to frame division and windowed Fourier transform to obtain Yi(t,f);
[0019] The product of Yi(t,f) and the spectral gain coefficient M(t,f) yields Si(t,f) = Yi(t,f) × M(t,f);
[0020] The clean signals si of the other channels in the multi-channel array, excluding the first channel, are obtained by performing an inverse Fourier transform on Si(t,f) and then adding them together.
[0021] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, obtaining the estimated clean signal of the first channel also includes:
[0022] The linearly filtered signal of the first channel, the microphone signal of the first channel, and the reference signal are combined and then fed into the residual echo suppression neural network model to obtain the estimated clean signal of the first channel.
[0023] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the model is a residual echo suppression neural network model.
[0024] Secondly, embodiments of the present invention also provide a low-resource multi-channel echo suppression device, comprising:
[0025] Acquisition module: used to acquire multi-channel microphone signals and multi-channel reference signals;
[0026] First input module: used to input the multi-channel microphone signal and the multi-channel reference signal into the linear filter to obtain the linearly filtered signal of each channel in the multi-channel;
[0027] The second input module is used to input the multi-channel reference signal into the Combine module to obtain the reference signal.
[0028] The third input module is used to combine the linearly filtered signal of the first channel with the reference signal and then input them into the model to obtain the estimated clean signal of the first channel.
[0029] First determining module: used to determine the spectral gain coefficient based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel;
[0030] The second determining module is used to multiplex the spectral gain coefficients to other channels in the multi-channel array besides the first channel to determine the clean signals of the other channels in the multi-channel array besides the first channel.
[0031] Thirdly, embodiments of the present invention also provide an electronic device, including: a processor and a memory;
[0032] The processor executes any of the aforementioned low-resource multi-channel echo suppression methods by calling programs or instructions stored in the memory.
[0033] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a program or instructions that cause a computer to execute any of the above-described low-resource multi-channel echo suppression methods.
[0034] The advantages of this invention are as follows: This invention relates to a low-resource multi-channel echo suppression method, apparatus, electronic device, and medium. The method includes: acquiring multi-channel microphone signals and multi-channel reference signals; feeding the multi-channel microphone signals and multi-channel reference signals into a linear filter to obtain a linearly filtered signal for each channel in the multi-channel system; feeding the multi-channel reference signals into a Combine module to obtain a reference signal; merging the linearly filtered signal of the first channel with the reference signal and feeding it into a model to obtain an estimated clean signal for the first channel; determining a spectral gain coefficient based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel; and multiplexing the spectral gain coefficient to other channels in the multi-channel system other than the first channel to determine the clean signals for the other channels. This invention reduces the computational load by merging echo channels, calculating the processing result of one microphone channel, and then calculating a spectral mask for multiplexing to other channels, thus meeting the application requirements of low-resource chips. Verification has shown that the echo suppression effect of this invention is still very good. Attached Figure Description
[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0036] Figure 1 This is a schematic diagram of a low-resource multi-channel echo suppression method provided in an embodiment of the present invention;
[0037] Figure 2 This is a schematic diagram of a low-resource multi-channel echo suppression device provided in an embodiment of the present invention;
[0038] Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0039] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0041] Figure 1 This is a schematic diagram of a low-resource multi-channel echo suppression method provided in an embodiment of the present invention.
[0042] In a first aspect, embodiments of the present invention provide a low-resource multi-channel echo suppression method, combined with Figure 1 It includes six steps, S101 to S106:
[0043] S101: Acquire multi-channel microphone signals and multi-channel reference signals.
[0044] Specifically, in this embodiment of the invention, multi-channel microphone signals Mic1 to MicM and multi-channel reference signals Ref1 to RefN are acquired. In the multi-channel microphone signals Mic1 to MicM, 1 to M represent the first channel to the Mth channel in the multi-channel signal, and 1 to N in the multi-channel reference signals Ref1 to RefN represent the first channel to the Nth channel in the multi-channel signal.
[0045] S102: The multi-channel microphone signal and the multi-channel reference signal are fed into a linear filter to obtain the linearly filtered signal of each channel in the multi-channel array.
[0046] Specifically, in this embodiment of the invention, the multi-channel microphone signals Mic1~MicM and the multi-channel reference signals Ref1~RefN are fed into a linear filter to obtain the linearly filtered signals y1~yM for each channel in the multi-channel array.
[0047] S103: Send the multi-channel reference signal into the Combine module to obtain the reference signal.
[0048] Specifically, in this embodiment of the invention, the reference signal is obtained by averaging the multiple reference signals Ref1 to RefN in the Combine module.
[0049] S104: The linearly filtered signal of the first channel is combined with the reference signal and then fed into the model to obtain the estimated clean signal of the first channel.
[0050] Specifically, in this embodiment of the invention, the linearly filtered signal y1 of the first channel is combined with the reference signal and then fed into the residual echo suppression neural network model to obtain the estimated clean signal s1 of the first channel.。
[0051] It should be understood that the clean signal s1 of the first channel is the signal after removing the echo from the first channel.
[0052] It should also be understood that the first channel here refers to the first channel in a multi-channel system, or it can be any channel in the multi-channel system. This invention uses the first channel as an example for illustration, but does not limit the scope of protection of this invention.
[0053] S105: Determine the spectral gain coefficient based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel.
[0054] Specifically, in this embodiment of the invention, the method for determining the spectral gain coefficient M(t,f) based on the estimated clean signal s1 of the first channel and the linearly filtered signal y1 of the first channel is described in detail below.
[0055] S106: Multiplex the spectral gain coefficients into the channels other than the first channel to determine the clean signals of the channels other than the first channel in the multi-channel.
[0056] Specifically, in this embodiment of the invention, the method of multiplexing the spectral gain coefficient M(t,f) to other channels in the multi-channel system other than the first channel to determine the clean signal of other channels in the multi-channel system other than the first channel is described in detail below.
[0057] It should be understood that this invention significantly simplifies modeling complexity by merging reference channels. Furthermore, the spectral gain is calculated using the processing result of one channel and then reused in other channels, further reducing the computational load to approximately 1 / N of the original scheme. By simplifying the computation, this invention meets the application requirements of low-resource chips, and verification shows that the echo suppression effect remains excellent.
[0058] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the linear filter is one of NLMS, RLS, or Kalman filtering.
[0059] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the multi-channel reference signal is sent to the Combine module to obtain the reference signal by averaging the multi-channel reference signals Ref1 to RefN through the Combine module.
[0060] Specifically, in this embodiment of the invention, averaging the multi-channel reference signals Ref1 to RefN involves summing the multi-channel reference signals Ref1 to RefN and then dividing by the number of channels.
[0061] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the spectral gain coefficient is determined based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel using the following formula:
[0062]
[0063] Where S1(t,f) is the estimated clean signal s1 of the first channel after frame division and windowing Fourier transform, Y1(t,f) is the linearly filtered signal y1 of the first channel after frame division and windowing Fourier transform, EPS is the system minimum value to prevent division by zero, floor is the preset minimum value to avoid over-suppression, and the value is 0.02. M(t,f) is less than or equal to 1.
[0064] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, multiplexing the spectral gain coefficients to other channels besides the first channel to determine the clean signals of the other channels besides the first channel includes:
[0065] The linearly filtered signal yi of the other channels in the multi-channel system, excluding the first channel, is subjected to frame division and windowed Fourier transform to obtain Yi(t,f);
[0066] The product of Yi(t,f) and the spectral gain coefficient M(t,f) yields Si(t,f) = Yi(t,f) × M(t,f);
[0067] The clean signals si of the other channels in the multi-channel array, excluding the first channel, are obtained by performing an inverse Fourier transform on Si(t,f) and then adding them together.
[0068] It should be understood that in this embodiment of the invention, by multiplexing the spectral gain coefficient M(t,f) to other channels in the multi-channel system other than the first channel to obtain the clean signal si of the other channels in the multi-channel system other than the first channel, the amount of calculation is simplified and the echo suppression efficiency is improved.
[0069] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, obtaining the estimated clean signal of the first channel also includes:
[0070] The linearly filtered signal of the first channel, the microphone signal of the first channel, and the reference signal are combined and then fed into the residual echo suppression neural network model to obtain the estimated clean signal of the first channel.
[0071] Specifically, in this embodiment of the invention, in S104 above, the linearly filtered signal y1 of the first channel is combined with the reference signal and then fed into the residual echo suppression neural network model to obtain the estimated clean signal s1 of the first channel. 。
[0072] In some embodiments, the linearly filtered signal y1 of the first channel, the microphone signal Mic1 of the first channel, and the reference signal can be merged and fed into the residual echo suppression neural network model to obtain the estimated clean signal s1 of the first channel. It has been verified that merging the microphone signal Mic1 of the first channel improves performance.
[0073] Furthermore, in the aforementioned low-resource multi-channel echo suppression method, the model is a residual echo suppression neural network model.
[0074] Figure 2 This is a schematic diagram of a low-resource multi-channel echo suppression device provided in an embodiment of the present invention.
[0075] Secondly, embodiments of the present invention also provide a low-resource multi-channel echo suppression device, combined with Figure 2 ,include:
[0076] Acquisition module 201: used to acquire multi-channel microphone signals and multi-channel reference signals.
[0077] Specifically, in this embodiment of the invention, the acquisition module 201 acquires multi-channel microphone signals Mic1~MicM and multi-channel reference signals Ref1~RefN, wherein 1~M in the multi-channel microphone signals Mic1~MicM represent the first channel to the Mth channel in the multi-channel signal, and 1~N in the multi-channel reference signals Ref1~RefN represent the first channel to the Nth channel in the multi-channel signal.
[0078] First input module 202: Used to input the multi-channel microphone signal and the multi-channel reference signal into a linear filter to obtain the linearly filtered signal of each channel in the multi-channel.
[0079] Specifically, in this embodiment of the invention, the first input module 202 inputs the multi-channel microphone signals Mic1~MicM and the multi-channel reference signals Ref1~RefN into a linear filter to obtain the linearly filtered signals y1~yM for each channel in the multi-channel.
[0080] The second input module 203 is used to input the multi-channel reference signal into the Combine module to obtain the reference signal.
[0081] Specifically, in this embodiment of the invention, the second input module 203 sends the multi-channel reference signals Ref1 to RefN into the Combine module to obtain the reference signal by averaging the multiple reference signals Ref1 to RefN.
[0082] The third input module 204 is used to combine the linearly filtered signal of the first channel with the reference signal and then input them into the model to obtain the estimated clean signal of the first channel.
[0083] Specifically, in this embodiment of the invention, the third input module 204 combines the linearly filtered signal y1 of the first channel with the reference signal and then inputs it into the residual echo suppression neural network model to obtain the estimated clean signal s1 of the first channel. 。
[0084] It should be understood that the clean signal s1 of the first channel is the signal after removing the echo from the first channel.
[0085] It should also be understood that the first channel here refers to the first channel in a multi-channel system, or it can be any channel in the multi-channel system. This invention uses the first channel as an example for illustration, but does not limit the scope of protection of this invention.
[0086] First determining module 205: used to determine the spectral gain coefficient based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel.
[0087] Specifically, in this embodiment of the invention, the method by which the first determining module 205 determines the spectral gain coefficient M(t,f) based on the estimated clean signal s1 of the first channel and the linearly filtered signal y1 of the first channel has been described in detail above.
[0088] The second determining module 206 is used to multiplex the spectral gain coefficients to other channels in the multi-channel system other than the first channel to determine the clean signals of other channels in the multi-channel system other than the first channel.
[0089] Specifically, in this embodiment of the invention, the method by which the second determining module 206 multiplexes the spectral gain coefficient M(t,f) into the other channels of the multi-channel system to determine the clean signals of the other channels besides the first channel has been described in detail above.
[0090] Thirdly, embodiments of the present invention also provide an electronic device, including: a processor and a memory;
[0091] The processor executes any of the aforementioned low-resource multi-channel echo suppression methods by calling programs or instructions stored in the memory.
[0092] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a program or instructions that cause a computer to execute any of the above-described low-resource multi-channel echo suppression methods.
[0093] Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this disclosure.
[0094] like Figure 3As shown, the electronic device includes at least one processor 301, at least one memory 302, and at least one communication interface 303. The various components of the electronic device are coupled together via a bus system 304. The communication interface 303 is used for information transmission with external devices. It is understood that the bus system 304 is used to implement communication between these components. In addition to a data bus, the bus system 304 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 3 The general designated all buses as Bus System 304.
[0095] It is understood that the memory 302 in this embodiment may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.
[0096] In some implementations, memory 302 stores elements such as executable units or data structures, or subsets thereof, or extended sets thereof: operating systems and applications.
[0097] The operating system, comprising various system programs such as the framework layer, core library layer, and driver layer, is used to implement various basic business functions and handle hardware-based tasks. The application programs, including media players and browsers, are used to implement various application functions. A program implementing any method in the low-resource multi-channel echo suppression method provided in this embodiment of the invention can be included in the application programs.
[0098] In this embodiment of the invention, the processor 301 executes the steps of various embodiments of the low-resource multi-channel echo suppression method provided in this embodiment of the invention by calling the program or instructions stored in the memory 302, specifically, the program or instructions stored in the application program.
[0099] Acquire multi-channel microphone signals and multi-channel reference signals;
[0100] The multi-channel microphone signal and the multi-channel reference signal are fed into a linear filter to obtain the linearly filtered signal of each channel in the multi-channel array.
[0101] The multi-channel reference signal is fed into the Combine module to obtain the reference signal;
[0102] The linearly filtered signal of the first channel is combined with the reference signal and then fed into the model to obtain the estimated clean signal of the first channel.
[0103] The spectral gain coefficient is determined based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel.
[0104] The spectral gain coefficients are multiplexed into the other channels of the multi-channel system, excluding the first channel, to determine the clean signals of the other channels.
[0105] Any of the methods in the low-resource multi-channel echo suppression method provided in this embodiment of the invention can be applied to, or implemented by, the processor 301. The processor 301 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 301 or by instructions in software form. The processor 301 can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, etc.
[0106] The steps of any method in the low-resource multi-channel echo suppression method provided in this embodiment of the invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software units in the decoding processor. The software units can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 302. The processor 301 reads the information in memory 302 and, in conjunction with its hardware, completes the steps of the low-resource multi-channel echo suppression method.
[0107] Those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.
[0108] Those skilled in the art will understand that the descriptions of the various embodiments have different focuses, and for parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0109] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A low-resource, multi-channel echo suppression method, characterized in that, include: Acquire multi-channel microphone signals and multi-channel reference signals; The multi-channel microphone signal and the multi-channel reference signal are fed into a linear filter to obtain the linearly filtered signal for each channel in the multi-channel system. The multi-channel reference signal is sent to the Combine module to obtain the reference signal; The linearly filtered signal of the first channel is combined with the reference signal and then fed into the model to obtain the estimated clean signal of the first channel. The spectral gain coefficient is determined based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel; The spectral gain coefficient is multiplexed into the other channels of the multi-channel system, excluding the first channel, to determine the clean signals of the other channels.
2. The low-resource multi-channel echo suppression method according to claim 1, characterized in that, Linear filters are one of the following: NLMS, RLS, and Kalman filtering.
3. The low-resource multi-channel echo suppression method according to claim 1, characterized in that, The multi-channel reference signal is sent to the Combine module to obtain the reference signal by averaging the multi-channel reference signals Ref1 to RefN through the Combine module.
4. The low-resource multi-channel echo suppression method according to claim 1, characterized in that, The spectral gain coefficient is determined based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel using the following formula: Where S1(t,f) is the estimated clean signal s1 of the first channel after frame division and windowing Fourier transform, Y1(t,f) is the linearly filtered signal y1 of the first channel after frame division and windowing Fourier transform, EPS is the system minimum value to prevent division by zero, floor is the preset minimum value to avoid over-suppression, and the value is 0.
02. M(t,f) is less than or equal to 1.
5. The low-resource multi-channel echo suppression method according to claim 1, characterized in that, The step of multiplexing the spectral gain coefficient to other channels in the multi-channel array, excluding the first channel, to determine the clean signals of the other channels, excluding the first channel, includes: The linearly filtered signal yi of the other channels in the multi-channel system, excluding the first channel, is subjected to frame division and windowed Fourier transform to obtain Yi(t,f); Based on the product of Yi(t,f) and the spectral gain coefficient M(t,f), we obtain Si(t,f) = Yi(t,f) × M(t,f); The clean signals si of the other channels in the multi-channel array, excluding the first channel, are obtained by performing an inverse Fourier transform on Si(t,f) and then overlapping and adding the signals.
6. The low-resource multi-channel echo suppression method according to claim 1, characterized in that, The estimated clean signal of the first channel further includes: The linearly filtered signal of the first channel, the microphone signal of the first channel, and the reference signal are combined and then fed into the residual echo suppression neural network model to obtain the estimated clean signal of the first channel.
7. The low-resource multi-channel echo suppression method according to claim 1, characterized in that, The model described is a residual echo suppression neural network model.
8. A low-resource multi-channel echo suppression device, characterized in that, include: Acquisition module: used to acquire multi-channel microphone signals and multi-channel reference signals; First input module: used to input the multi-channel microphone signal and the multi-channel reference signal into a linear filter to obtain the linearly filtered signal of each channel in the multi-channel; The second input module is used to input the multi-channel reference signal into the Combine module to obtain the reference signal; The third input module is used to combine the linearly filtered signal of the first channel with the reference signal and then input the result into the model to obtain the estimated clean signal of the first channel. First determining module: used to determine the spectral gain coefficient based on the estimated clean signal of the first channel and the linearly filtered signal of the first channel; The second determining module is used to multiplex the spectral gain coefficients to other channels in the multi-channel system besides the first channel to determine the clean signals of the other channels in the multi-channel system besides the first channel.
9. An electronic device, characterized in that, include: Processor and memory; The processor executes a low-resource multi-channel echo suppression method as described in any one of claims 1 to 7 by calling programs or instructions stored in the memory.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program or instructions that cause a computer to perform a low-resource multi-channel echo suppression method as described in any one of claims 1 to 7.