Residual echo suppression method and system based on expander

Abstract

The invention provides a residual echo suppression method and system based on an expander. The method comprises the following steps: converting a collected audio signal into a first frequency domain signal; calculating an inhibition factor of each sub-band; calculating a residual echo suppression loss ratio of the full frequency band, judging whether the loss ratio exceeds a preset threshold, if so, inputting the suppression factor of each sub-band into a downward expander for processing, and then processing by an upward expander to obtain a processed suppression factor; otherwise, not processing the suppression factor of each sub-band; processing the first frequency domain signal by adopting the obtained suppression factor of each sub-band to obtain a second frequency domain signal; and converting the second frequency domain signal into a time domain signal and outputting the time domain signal. According to the invention, a dual-mode residual echo suppression model is set for a duplex call scene, voice signals are reserved to the maximum extent, voice interruption is avoided, and call quality and user experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of audio signal processing, and in particular to a residual echo suppression method and system based on an expander. Background Art

[0002] In voice communication systems, due to the existence of hardware loops or acoustic loops, the far-end voice signal will be transmitted back to the near end, forming an unacceptable echo phenomenon. Therefore, modern communication systems often use echo cancellers to suppress echoes. However, in practical applications, due to factors such as nonlinear distortion and noise, echo cancellers cannot completely eliminate echoes, and there will often be some residual echoes. These residual echoes not only affect the call quality, but also introduce noise and distortion, reducing the intelligibility of the voice.

[0003] For example, the Chinese invention patent with publication number CN103718538A discloses a method of echo cancellation based on the frequency domain. In the nonlinear echo suppression part, after calculating the suppression factors of each sub-band of the current frame, the method proposes to arrange the suppression factors in ascending order and take the upper quartile as the upper limit threshold, so as to constrain the upper limit of the suppression factor and achieve the purpose of reducing the residual echo in the signal. However, when the near-end speech and the residual echo are mixed together, such as in a duplex call scenario, and the speech energy is weak, the method may suppress the near-end speech together, resulting in discontinuity of speech, thereby affecting the call quality. In the algorithm scheme disclosed by it, when speech and residual echo appear at the same time, the suppression factor of the sub-band where only speech exists is close to 1, the suppression factor of the sub-band where both speech and residual echo exist is between 0 and 1, and the suppression factor of the sub-band where only residual echo exists is close to 0. When the upper quartile appears in the speech sub-band, the speech will not be obviously suppressed, and the speech energy is stable and comfortable to listen to; when the upper quartile appears in the sub-band where speech and residual echo exist at the same time, the degree of speech suppression increases as the speech energy decreases, and the speech energy stability fluctuates; when the upper quartile appears in the sub-band where the residual echo is located, the above algorithm will obviously suppress the speech, and the speech will be obviously intermittent, affecting the call quality.

[0004] Although various post-processing methods are also used for the suppression factor in the prior art, a fixed quantile is usually used as an upper threshold to constrain the suppression factor value of the entire frequency band. This still causes problems such as speech distortion and amplitude attenuation. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a residual echo suppression method and system based on an expander, set a dual-mode residual echo suppression model for a duplex call scenario, retain the voice signal to the maximum extent, avoid voice interruption, and improve call quality and user experience.

[0006] In a first aspect, the present invention provides a residual echo suppression method based on an expander, the method comprising:

[0007] Step S1, converting the collected audio signal into a first frequency domain signal;

[0008] Step S2, calculating a suppression factor of each sub-band in the first frequency domain signal, wherein the suppression factor is calculated based on a correlation between the far-end signal and the near-end signal;

[0009] Step S3, executing a post-processing procedure, calculating the residual echo suppression loss ratio of the entire frequency band, and determining whether the residual echo suppression loss ratio value exceeds a preset threshold value, if so, proceeding to step S4, otherwise, proceeding to step S5;

[0010] Step S4, entering the echo suppression mode, inputting the suppression factor of each sub-band into a downward expander for processing, and then inputting the processed value into an upward expander for further processing, and obtaining the suppression factor of each sub-band as the output of the post-processing program;

[0011] Step S5, entering the speech mode, without processing the suppression factors of each sub-band, directly using them as the output of the post-processing program;

[0012] Step S6, using the suppression factor of each sub-frequency band output by the post-processing program to process the first frequency domain signal of each sub-frequency band to obtain a second frequency domain signal;

[0013] Step S7: convert the second frequency domain signal into a time domain signal for output.

[0014] Furthermore, the calculation formula of the inhibition factor in step S2 is as follows:

[0015]

[0016] Where d represents the near-end signal, x represents the far-end signal, and s k [n] represents the suppression factor of the nth frame at frequency subband k, P dk [n] is the near-end signal power spectrum at frequency subband k in the nth frame, P xk [n] is the far-end signal power spectrum at frequency subband k in the nth frame, P dxk [n] is the cross power spectrum of the near-end and far-end signals in the nth frame at frequency subband k, YesP dxk The transposed conjugate of [n];

[0017] The calculation methods of the near-end signal power spectrum, the far-end signal power spectrum and the near-end and far-end signal cross-power spectrum are as follows:

[0018]

[0019] Among them, D k [n] is the near-end frequency domain signal of the nth frame at frequency subband k, Yes D k The transposed conjugate of [n], λ = 0.92 is the smoothing factor, X k [n] is the far-end frequency domain signal of the nth frame at frequency subband k, Yes X k The transposed conjugate of [n].

[0020] Furthermore, the residual echo suppression loss ratio is calculated as follows:

[0021]

[0022] Where ERLE_total[n] is the residual echo suppression loss ratio of the nth frame, M is the number of subbands in the frequency domain, and s k [n] represents the suppression factor of the nth frame at frequency subband k.

[0023] Furthermore, both the downward expander and the upward expander use the following calculation formula:

[0024]

[0025] Among them, s k [n] represents the suppression factor of the nth frame input to the expander at frequency subband k, represents the suppression factor of the nth frame at the frequency subband k after processing by the expander, T is the threshold of the expander, [TW / 2,T+W / 2] is the soft knee interval, R is the compression ratio of the expander, and W is the soft knee of the expander;

[0026] The input of the upward expander is the suppression factor calculated in step S2, and the input of the downward expander is the output of the downward expander;

[0027] The parameters of the downward expander are set to T1=0.8, R1=1.6, W1=0.2, and the parameters of the upward expander are set to T2=0, R2=0, W2=-0.3.

[0028] Furthermore, the method further includes dynamically adjusting the trigger time and release time of the expander according to the historical data set, and before step S7, further includes smoothing the inhibition factor output by the post-processing program, and the specific process of the smoothing process is as follows:

[0029] when hour,

[0030]

[0031] when hour,

[0032]

[0033] Where S sk [n] is the suppression factor of the nth frame at frequency subband k after smoothing, is the suppression factor of the nth frame at frequency subband k obtained by post-processing, γ r and γ a is the smoothing coefficient, and the smoothing coefficient γ r and γ a The values ​​of are as follows:

[0034]

[0035] Among them, T a is the trigger time, T r is the release time, f s is the sampling rate.

[0036] Furthermore, the method also includes setting a gain compensation module to perform gain compensation processing on the second frequency domain signal.

[0037] In a second aspect, the present invention provides a residual echo suppression system based on an expander, the system comprising:

[0038] An input module, used for converting the collected audio signal into a first frequency domain signal;

[0039] A suppression factor calculation module, used to calculate a suppression factor of each sub-band in the first frequency domain signal, wherein the suppression factor is calculated based on a correlation between a far-end signal and a near-end signal;

[0040] A post-processing module, used to execute a post-processing program, calculate the residual echo suppression loss ratio of the full frequency band, and determine whether the residual echo suppression loss ratio value exceeds a preset threshold value. If so, the echo suppression mode module is executed, otherwise, the speech mode module is executed;

[0041] An echo suppression mode module, used to enter the echo suppression mode, input the suppression factor of each sub-band into a downward expander for processing, and then input the processed value into an upward expander for reprocessing, and obtain the suppression factor of each sub-band as the output of the post-processing program;

[0042] The speech mode module is used to enter the speech mode, and the suppression factors of each sub-band are not processed and are directly used as the output of the post-processing program;

[0043] A signal processing module, used to process the first frequency domain signals of the respective sub-frequency bands using the suppression factors of the respective sub-frequency bands output by the post-processing program to obtain second frequency domain signals;

[0044] The output module is used to convert the second frequency domain signal into a time domain signal for output.

[0045] Furthermore, the inhibition factor calculation formula in the inhibition factor calculation module is as follows:

[0046]

[0047] Where d represents the near-end signal, x represents the far-end signal, and s k [n] represents the suppression factor of the nth frame at frequency subband k, P dk [n] is the near-end signal power spectrum at frequency subband k in the nth frame, P xk [n] is the far-end signal power spectrum at frequency subband k in the nth frame, P dxk [n] is the cross power spectrum of the near-end and far-end signals in the nth frame at frequency subband k, YesP dxk The transposed conjugate of [n];

[0048] The calculation methods of the near-end signal power spectrum, the far-end signal power spectrum and the near-end and far-end signal cross-power spectrum are as follows:

[0049]

[0050] Among them, D k [n] is the near-end frequency domain signal of the nth frame at frequency subband k, Yes D k The transposed conjugate of [n], λ = 0.92 is the smoothing factor, X k [n] is the far-end frequency domain signal of the nth frame at frequency subband k, Yes X k The transposed conjugate of [n].

[0051] Furthermore, the residual echo suppression loss ratio is calculated as follows:

[0052]

[0053] Where ERLE_total[n] is the residual echo suppression loss ratio of the nth frame, M is the number of subbands in the frequency domain, and s k [n] represents the suppression factor of the nth frame at frequency subband k.

[0054] Furthermore, both the downward expander and the upward expander use the following calculation formula:

[0055]

[0056] Among them, s k [n] represents the suppression factor of the nth frame input to the expander at frequency subband k, represents the suppression factor of the nth frame at the frequency subband k after processing by the expander, T is the threshold of the expander, [TW / 2,T+W / 2] is the soft knee interval, R is the compression ratio of the expander, and W is the soft knee of the expander;

[0057] The input of the upward expander is the suppression factor calculated by the suppression factor calculation module, and the input of the downward expander is the output of the downward expander;

[0058] The parameters of the downward expander are set to T1=0.8, R1=1.6, W1=0.2, and the parameters of the upward expander are set to T2=0, R2=0, W2=-0.3.

[0059] Furthermore, the method further includes dynamically adjusting the trigger time and release time of the expander according to the historical data set, and also includes a smoothing processing module before the signal processing module, which is used to smooth the inhibition factor output by the post-processing program, and the specific process of the smoothing processing is as follows:

[0060] when hour,

[0061]

[0062] when hour,

[0063]

[0064] Where S sk [n] is the suppression factor of the nth frame at frequency subband k after smoothing, is the suppression factor of the nth frame at frequency subband k obtained by post-processing, γ r and γ a is the smoothing coefficient, and the smoothing coefficient γ r and γ a The values ​​of are as follows:

[0065]

[0066] Among them, T a is the trigger time, T r is the release time, f s is the sampling rate.

[0067] Furthermore, the system also includes setting a gain compensation module to perform gain compensation processing on the second frequency domain signal.

[0068] The present invention has the following advantages:

[0069] 1. Dynamically adjust the residual echo suppression strength according to the ERLE situation of the full frequency band to avoid excessive processing of speech components and ensure the fidelity of speech;

[0070] 2. Using a downward expander to further nonlinearly compress the suppression factor can effectively reduce the residual echo energy and reduce the damage to the speech energy; in a scenario where only speech exists, the speech will not be obviously distorted;

[0071] 3. In the scenario where speech and residual echo coexist, reduce the influence of the suppression factor of the sub-band where the residual echo is located on the suppression factor of the sub-band where the speech is located, and improve the problem of speech energy attenuation;

[0072] 4. The smooth transition of nonlinear mapping is achieved through the double-sided soft inflection points, avoiding the problems of sound mutation and distortion;

[0073] 5. Through multiple configurable parameters, nonlinear processing is made more flexible and controllable, and the aggressiveness of residual echo suppression can be customized to meet the adaptation requirements of devices with different distortion conditions;

[0074] 6. The present invention has a relatively small amount of calculation, is easy to implement real-time operation, and can be widely used in the field of voice communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0076] Figure 1 The present invention is an execution flow chart of an expander-based residual echo suppression method.

[0077] Figure 2 The present invention is a logic structure block diagram of a residual echo suppression system based on an expander.

[0078] Figure 3 It is a schematic diagram of the processing result of the suppression factor by the dual expander in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0079] The embodiment of the present application provides a residual echo suppression method and system based on an expander, and proposes a dual-mode residual echo suppression strategy for three scenarios, namely, voice only, residual echo only, and voice and residual echo mixture. The technical solution in the embodiment of the present application has the following overall idea: using ERLE to tailor different degrees of non-linear residual echo suppression for different frequency bands, further improving voice quality and call experience, while providing multiple configurable parameters to make nonlinear processing more flexible and controllable, reduce computational complexity, and also consider voice fidelity, avoiding sudden changes and distortion of sound through bilateral soft inflection point design.

[0080] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0081] Embodiment 1

[0082] like Figure 1 As shown, the present invention provides a residual echo suppression method based on an expander, the method comprising:

[0083] Step S1, converting the collected audio signal into a first frequency domain signal; this process can be obtained by using conventional Fourier transform;

[0084] Step S2, calculating a suppression factor of each sub-band in the first frequency domain signal, wherein the suppression factor is calculated based on a correlation between the far-end signal and the near-end signal;

[0085] Step S3, executing a post-processing program, calculating the residual echo suppression loss ratio of the entire frequency band, and determining whether the residual echo suppression loss ratio value exceeds a preset threshold value. If so, it indicates that the current scene is only residual echo or speech and residual echo coexist, and the process proceeds to step S4. Otherwise, it indicates that the current scene is only speech, and the process proceeds to step S5.

[0086] Step S4, entering the echo suppression mode, inputting the suppression factor of each sub-band into a downward expander for processing, and then inputting the processed value into an upward expander for further processing, to obtain the suppression factor of each sub-band as the output of the post-processing program; this mode requires further processing of the suppression factor corresponding to the frequency band with larger ERLE to reduce the energy of the residual echo; this mode ensures that nonlinear residual echo suppression is performed only on the frequency band with higher ERLE to avoid speech distortion caused by excessive processing;

[0087] Step S5, entering the speech mode, without processing the suppression factors of each sub-band, directly using them as the output of the post-processing program, so as to retain the speech components to the maximum extent;

[0088] Step S6, using the suppression factor of each sub-frequency band output by the post-processing program to process the first frequency domain signal of each sub-frequency band to obtain a second frequency domain signal;

[0089] Step S7: convert the second frequency domain signal into a time domain signal for output. The conversion process can convert the frequency domain signal into a time domain signal by inverse Fourier transform.

[0090] Preferably, the calculation formula of the inhibition factor in step S2 is as follows:

[0091]

[0092] Where d represents the near-end signal, x represents the far-end signal, and s k [n] represents the suppression factor of the nth frame at frequency subband k, P dk [n] is the near-end signal power spectrum at frequency subband k in the nth frame, P xk [n] is the far-end signal power spectrum at frequency subband k in the nth frame, P dxk [n] is the cross power spectrum of the near-end and far-end signals in the nth frame at frequency subband k, YesP dxk [n] is the transposed conjugate; the above k represents the frequency domain subband index, and n represents the frame index.

[0093] The calculation methods of the near-end signal power spectrum, the far-end signal power spectrum and the near-end and far-end signal cross-power spectrum are as follows:

[0094]

[0095] Among them, D k [n] is the near-end frequency domain signal of the nth frame at frequency subband k, Yes D k The transposed conjugate of [n], λ = 0.92 is the smoothing factor, X k [n] is the far-end frequency domain signal of the nth frame at frequency subband k, Yes X k The transposed conjugate of [n].

[0096] Preferably, the residual echo suppression loss ratio is calculated as follows:

[0097]

[0098] Where ERLE_total[n] is the residual echo suppression loss ratio of the nth frame, M is the number of subbands in the frequency domain, and s k [n] represents the suppression factor of the nth frame at frequency subband k.

[0099] Preferably, both the downward expander and the upward expander use the following calculation formula:

[0100]

[0101] Among them, s k [n] represents the suppression factor of the nth frame input to the expander at frequency subband k, represents the suppression factor of the nth frame at frequency subband k after processing by the expander, T is the threshold of the expander, [TW / 2,T+W / 2] is the soft knee interval, R is the compression ratio of the expander, which controls the downward expansion strength and corresponds to the suppression strength of the residual echo, and W is the soft knee of the expander, which allows a smooth transition of the suppression factor to avoid introducing distortion;

[0102] The input of the upward expander is the suppression factor calculated in step S2, and the input of the downward expander is the output of the downward expander;

[0103] The parameters of the downward expander are set to T1=0.8, R1=1.6, W1=0.2, and the parameters of the upward expander are set to T2=0, R2=0, W2=-0.3. The curve of the processed suppression factor output after the suppression factor is processed by the above two expanders is as follows: Figure 3 As shown, the two expanders are cascaded and are generally triggered when the echo suppression mode is triggered. It can be seen that there are two soft inflection points in the curve graph.

[0104] Preferably, the method further comprises dynamically adjusting the trigger time and release time of the expander according to the historical data set, respectively controlling the start and stop time lengths of the downward expander, and affecting the response sensitivity of the downward expander to the isolated residual echo.

[0105] Before step S7, the method further includes smoothing the inhibition factor output by the post-processing program. The specific process of the smoothing process is as follows:

[0106] when hour,

[0107]

[0108] when hour,

[0109]

[0110] Where S sk [n] is the suppression factor of the nth frame at frequency subband k after smoothing, is the suppression factor of the nth frame at frequency subband k obtained by post-processing, γ r and γa is the smoothing coefficient, and the smoothing coefficient γ r and γ a The values ​​of are as follows:

[0111]

[0112] Among them, T a is the trigger time, T r is the release time, f s is the sampling rate. Trigger time T a The value range is 0.1 to 0.2 seconds. Release time T r The value range is 0.5 to 1 second.

[0113] Preferably, the method further comprises setting a gain compensation module to perform gain compensation processing on the second frequency domain signal. The downward expander further attenuates the suppression factor coefficient of each sub-band. After the frequency domain signal is processed, the energy is reduced, so appropriate compensation can be performed to make up for the loudness loss.

[0114] Based on the same inventive concept, the present application also provides a system corresponding to the method in Example 1, see Example 2 for details.

[0115] Embodiment 2

[0116] In this embodiment, a residual echo suppression system based on an expander is provided. Figure 2 As shown, the system comprises:

[0117] An input module, used to convert the collected audio signal into a first frequency domain signal; this process can be obtained by using conventional Fourier transform;

[0118] A suppression factor calculation module, used to calculate a suppression factor of each sub-band in the first frequency domain signal, wherein the suppression factor is calculated based on a correlation between a far-end signal and a near-end signal;

[0119] A post-processing module, used to execute a post-processing program, calculate the residual echo suppression loss ratio of the full frequency band, and determine whether the residual echo suppression loss ratio value exceeds a preset threshold value. If so, it indicates that the current scene is only residual echo or voice and residual echo coexist, and the echo suppression mode module is executed; otherwise, it indicates that the current scene is only voice, and the voice mode module is executed;

[0120] The echo suppression mode module is used to enter the echo suppression mode, input the suppression factor of each sub-band into a downward expander for processing, and then input the processed value into an upward expander for further processing to obtain the suppression factor of each sub-band as the output of the post-processing program; this mode requires further processing of the suppression factor corresponding to the frequency band with larger ERLE to reduce the energy of the residual echo; this mode ensures that nonlinear residual echo suppression is performed only on the frequency band with higher ERLE to avoid excessive processing leading to speech distortion;

[0121] The speech mode module is used to enter the speech mode, and the suppression factors of each sub-band are not processed and directly used as the output of the post-processing program to maximize the retention of speech components;

[0122] A signal processing module, used to process the first frequency domain signals of the respective sub-frequency bands using the suppression factors of the respective sub-frequency bands output by the post-processing program to obtain second frequency domain signals;

[0123] The output module is used to convert the second frequency domain signal into a time domain signal for output. The conversion process can convert the frequency domain signal into a time domain signal by inverse Fourier transform.

[0124] Preferably, the inhibition factor calculation formula in the inhibition factor calculation module is as follows:

[0125]

[0126] Where d represents the near-end signal, x represents the far-end signal, and s k [n] represents the suppression factor of the nth frame at frequency subband k, P dk [n] is the near-end signal power spectrum at frequency subband k in the nth frame, P xk [n] is the far-end signal power spectrum at frequency subband k in the nth frame, P dxk [n] is the cross power spectrum of the near-end and far-end signals in the nth frame at frequency subband k, YesP dxk [n] is the transposed conjugate; the above k represents the frequency domain subband index, and n represents the frame index.

[0127] The calculation methods of the near-end signal power spectrum, the far-end signal power spectrum and the near-end and far-end signal cross-power spectrum are as follows:

[0128]

[0129] Among them, D k [n] is the near-end frequency domain signal of the nth frame at frequency subband k, Yes D k The transposed conjugate of [n], λ = 0.92 is the smoothing factor, Xk [n] is the far-end frequency domain signal of the nth frame at frequency subband k, Yes X k The transposed conjugate of [n].

[0130] Preferably, the residual echo suppression loss ratio is calculated as follows:

[0131]

[0132] Where ERLE_total[n] is the residual echo suppression loss ratio of the nth frame, M is the number of subbands in the frequency domain, and s k [n] represents the suppression factor of the nth frame at frequency subband k.

[0133] Preferably, both the downward expander and the upward expander use the following calculation formula:

[0134]

[0135] Among them, s k [n] represents the suppression factor of the nth frame input to the expander at frequency subband k, represents the suppression factor of the nth frame at frequency subband k after processing by the expander, T is the threshold of the expander, [TW / 2,T+W / 2] is the soft knee interval, R is the compression ratio of the expander, which controls the downward expansion strength and corresponds to the suppression strength of the residual echo, and W is the soft knee of the expander, which allows a smooth transition of the suppression factor to avoid introducing distortion;

[0136] The input of the upward expander is the suppression factor calculated by the suppression factor calculation module, and the input of the downward expander is the output of the downward expander;

[0137] The parameters of the downward expander are set to T1=0.8, R1=1.6, W1=0.2, and the parameters of the upward expander are set to T2=0, R2=0, W2=-0.3. The curve of the processed suppression factor output after the suppression factor is processed by the above two expanders is as follows: Figure 3 As shown, it can be seen that there are two soft inflection points in the curve graph.

[0138] Preferably, the method further comprises dynamically adjusting the trigger time and release time of the expander according to the historical data set, respectively controlling the start and stop time lengths of the downward expander, and affecting the response sensitivity of the downward expander to the isolated residual echo. A smoothing processing module is also included before the signal processing module, which is used to smooth the suppression factor output by the post-processing program, and the specific process of the smoothing processing is as follows:

[0139] when hour,

[0140]

[0141] when hour,

[0142]

[0143] Where S sk [n] is the suppression factor of the nth frame at frequency subband k after smoothing, is the suppression factor of the nth frame at frequency subband k obtained by post-processing, γ r and γ a is the smoothing coefficient, and the smoothing coefficient γ r and γ a The values ​​of are as follows:

[0144]

[0145]

[0146] Among them, T a is the trigger time, T r is the release time, f s is the sampling rate. Trigger time T a The value range is 0.1 to 0.2 seconds. Release time T r The value range is 0.5 to 1 second.

[0147] Preferably, the system further comprises a gain compensation module for performing gain compensation processing on the second frequency domain signal. The downward expander further attenuates the suppression factor coefficient of each sub-band, and the frequency domain signal has reduced energy after this processing, so appropriate compensation can be performed to make up for the loudness loss.

[0148] The present invention dynamically adjusts the residual echo suppression strength according to the residual echo suppression loss ratio of the whole frequency band, adopts a dual-mode residual echo suppression post-processing strategy, greatly improves the integrity of speech energy, and avoids excessive processing of speech components; adopts a downward expander to further nonlinearly compress the suppression factor, and uses two cascaded expanders to perform nonlinear expansion processing on the residual echo, while ensuring the suppression of the residual echo energy, while minimizing the damage of speech energy; in a scene where only speech exists, the speech will not be obviously distorted; in a scene where speech and residual echo coexist, the influence of the suppression factor of the sub-band where the residual echo is located on the suppression factor of the sub-band where the speech is located is reduced, and the problem of speech energy attenuation is improved; a smooth transition of nonlinear mapping is achieved through bilateral soft inflection points, avoiding the problem of sudden sound changes and distortion; through multiple configurable parameters, the nonlinear processing is more flexible and controllable, the aggressiveness of the residual echo suppression can be customized, and the equipment with different distortion degrees can be flexibly adapted; the gain compensation of the nonlinear distortion suppression makes up for the loudness distortion caused by the post-processing; the present invention has a relatively small amount of calculation, is easy to realize real-time operation, and can be widely used in the field of speech communication.

[0149] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A residual echo suppression method based on an expander, characterized in that: The method comprises: Step S1, converting the collected audio signal into a first frequency domain signal; Step S2, calculating a suppression factor of each sub-band in the first frequency domain signal, wherein the suppression factor is calculated based on a correlation between the far-end signal and the near-end signal; Step S3, executing a post-processing procedure, calculating the residual echo suppression loss ratio of the entire frequency band, and determining whether the residual echo suppression loss ratio value exceeds a preset threshold value, if so, proceeding to step S4, otherwise, proceeding to step S5; Step S4, entering the echo suppression mode, inputting the suppression factor of each sub-band into a downward expander for processing, and then inputting the processed value into an upward expander for further processing, and obtaining the suppression factor of each sub-band as the output of the post-processing program; Step S5, entering the speech mode, without processing the suppression factors of each sub-band, directly using them as the output of the post-processing program; Step S6, using the suppression factor of each sub-frequency band output by the post-processing program to process the first frequency domain signal of each sub-frequency band to obtain a second frequency domain signal; Step S7: convert the second frequency domain signal into a time domain signal for output.

2. The method for suppressing residual echo based on an expander according to claim 1, characterized in that: The calculation formula of the inhibition factor in step S2 is as follows: Where d represents the near-end signal, x represents the far-end signal, and s k [n] represents the suppression factor of the nth frame at frequency subband k, P dk [n] is the near-end signal power spectrum at frequency subband k in the nth frame, P xk [n] is the far-end signal power spectrum at frequency subband k in the nth frame, P dxk [n] is the cross power spectrum of the near-end and far-end signals in the nth frame at frequency subband k, YesP dxk The transposed conjugate of [n]; The calculation methods of the near-end signal power spectrum, the far-end signal power spectrum and the near-end and far-end signal cross-power spectrum are as follows: Among them, D k [n] is the near-end frequency domain signal of the nth frame at frequency subband k, Yes D k The transposed conjugate of [n], λ = 0.92 is the smoothing factor, X k [n] is the far-end frequency domain signal of the nth frame at frequency subband k, Yes X k The transposed conjugate of [n].

3. The method for suppressing residual echo based on an expander according to claim 1, characterized in that: The residual echo suppression loss ratio calculation formula is: Where ERLE_total[n] is the residual echo suppression loss ratio of the nth frame, M is the number of subbands in the frequency domain, and s k [n] represents the suppression factor of the nth frame at frequency subband k.

4. The method for suppressing residual echo based on an expander according to claim 1, characterized in that: The downward expander and the upward expander both use the following calculation formula: Among them, s k [n] represents the suppression factor of the nth frame input to the expander at frequency subband k, represents the suppression factor of the nth frame at the frequency subband k after processing by the expander, T is the threshold of the expander, [TW / 2,T+W / 2] is the soft knee interval, R is the compression ratio of the expander, and W is the soft knee of the expander; The input of the upward expander is the suppression factor calculated in step S2, and the input of the downward expander is the output of the downward expander; The parameters of the downward expander are set to T1=0.8, R1=1.6, W1=0.2, and the parameters of the upward expander are set to T2=0, R2=0, W2=-0.

3.

5. The method for suppressing residual echo based on an expander according to claim 1, characterized in that: The method further includes dynamically adjusting the trigger time and release time of the expander according to the historical data set, and before step S6, also includes smoothing the inhibition factor output by the post-processing program, and the specific process of the smoothing is as follows: when hour, when hour, Where S sk [n] is the suppression factor of the nth frame at frequency subband k after smoothing, is the suppression factor of the nth frame at frequency subband k obtained by post-processing, γ r and γ a is the smoothing coefficient, and the smoothing coefficient γ r and γ a The values ​​of are as follows: Among them, T a is the trigger time, T r is the release time, f s is the sampling rate.

6. A residual echo suppression system based on an expander, characterized in that: The system comprises: An input module, used for converting the collected audio signal into a first frequency domain signal; A suppression factor calculation module, used to calculate a suppression factor of each sub-band in the first frequency domain signal, wherein the suppression factor is calculated based on a correlation between a far-end signal and a near-end signal; A post-processing module, used to execute a post-processing program, calculate the residual echo suppression loss ratio of the full frequency band, and determine whether the residual echo suppression loss ratio value exceeds a preset threshold value. If so, the echo suppression mode module is executed, otherwise, the speech mode module is executed; An echo suppression mode module, used to enter the echo suppression mode, input the suppression factor of each sub-band into a downward expander for processing, and then input the processed value into an upward expander for reprocessing, and obtain the suppression factor of each sub-band as the output of the post-processing program; The speech mode module is used to enter the speech mode, and the suppression factors of each sub-band are not processed and are directly used as the output of the post-processing program; A signal processing module, used to process the first frequency domain signals of the respective sub-frequency bands using the suppression factors of the respective sub-frequency bands output by the post-processing program to obtain second frequency domain signals; The output module is used to convert the second frequency domain signal into a time domain signal for output.

7. The expander-based residual echo suppression system according to claim 6, characterized in that: The specific calculation formula of the inhibition factor in the inhibition factor calculation module is as follows: Where d represents the near-end signal, x represents the far-end signal, and s k [n] represents the suppression factor of the nth frame at frequency subband k, P dk [n] is the near-end signal power spectrum at frequency subband k in the nth frame, P xk [n] is the far-end signal power spectrum at frequency subband k in the nth frame, P dxk [n] is the cross power spectrum of the near-end and far-end signals in the nth frame at frequency subband k, YesP dxk The transposed conjugate of [n]; The calculation methods of the near-end signal power spectrum, the far-end signal power spectrum and the near-end and far-end signal cross-power spectrum are as follows: Among them, D k [n] is the near-end frequency domain signal of the nth frame at frequency subband k, Yes D k The transposed conjugate of [n], λ = 0.92 is the smoothing factor, X k [n] is the far-end frequency domain signal of the nth frame at frequency subband k, Yes X k The transposed conjugate of [n].

8. The expander-based residual echo suppression system according to claim 6, characterized in that: The residual echo suppression loss ratio calculation formula is: Where ERLE_total[n] is the residual echo suppression loss ratio of the nth frame, M is the number of subbands in the frequency domain, and s k [n] represents the suppression factor of the nth frame at frequency subband k.

9. The expander-based residual echo suppression system according to claim 6, characterized in that: The downward expander and the upward expander both use the following calculation formula: Among them, s k [n] represents the suppression factor of the nth frame input to the expander at frequency subband k, represents the suppression factor of the nth frame at the frequency subband k after processing by the expander, T is the threshold of the expander, [TW / 2,T+W / 2] is the soft knee interval, R is the compression ratio of the expander, and W is the soft knee of the expander; The input of the upward expander is the suppression factor calculated by the suppression factor calculation module, and the input of the downward expander is the output of the downward expander; The parameters of the downward expander are set to T1=0.8, R1=1.6, W1=0.2, and the parameters of the upward expander are set to T2=0, R2=0, W2=-0.

3.

10. The expander-based residual echo suppression system according to claim 6, characterized in that: The method further includes dynamically adjusting the trigger time and release time of the expander according to the historical data set, and also includes a smoothing processing module before the signal processing module, which is used to smooth the inhibition factor output by the post-processing program. The specific process of the smoothing processing is as follows: when hour, when hour, Where S sk [n] is the suppression factor of the nth frame at frequency subband k after smoothing, is the suppression factor of the nth frame at frequency subband k obtained by post-processing, γ r and γ a is the smoothing coefficient, and the smoothing coefficient γ r and γ a The values ​​of are as follows: Among them, T a is the trigger time, T r is the release time, f s is the sampling rate.

Images