Method for determining residual echo suppression coefficients and residual echo suppression method

Abstract

The application provides a residual echo suppression coefficient determination method, which is used for offline determination of the residual echo suppression coefficient, and comprises the following steps: determining a frequency response curve of a playback link of an electronic device based on equalization parameters of the playback link; selecting a plurality of to-be-processed frequency points based on resonance characteristics of the electronic device, and dividing the plurality of to-be-processed frequency points to form a plurality of to-be-processed frequency bands; and determining residual echo suppression coefficients corresponding to the plurality of to-be-processed frequency bands based on the frequency response curve. The residual echo suppression coefficient determination method and the residual echo suppression method provided by the application can efficiently and high-quality complete the suppression of residual echo.

Description

Technical Field

[0001] This invention relates to the field of audio processing technology, and in particular to a method for determining the residual echo suppression coefficient and a method for suppressing residual echo. Background Technology

[0002] In voice interaction scenarios such as hands-free calling, video conferencing, and smart speakers, the far-end signal played by the speaker is picked up again by the microphone, forming an acoustic echo. Acoustic Echo Cancellation (AEC) is the core technology to solve this problem. Its basic principle is to use an adaptive filter to estimate the acoustic echo path and subtract the estimated echo component from the microphone signal. A typical AEC system usually includes a linear echo canceller and a residual echo suppressor. The linear echo canceller usually uses an adaptive filter (such as NLMS, IPNLMS, etc.) to simulate the acoustic echo path and outputs an echo estimation signal; the residual echo suppressor (RES) further suppresses residual echoes based on the nonlinear and time-varying characteristics of the acoustic path.

[0003] In existing technologies, residual echo suppressors typically rely on the coherence between the reference signal and the error signal, echo loss enhancement, or deep learning models based on echo suppression. Regardless of the type of residual echo suppressor used, they all have a long delay and fail to provide effective residual echo suppression quickly when the call starts or when the echo path changes abruptly. Summary of the Invention

[0004] The method for determining the residual echo suppression coefficient and the residual echo suppression method provided by this invention can efficiently and effectively suppress residual echoes.

[0005] In a first aspect, the present invention provides a method for determining the residual echo suppression coefficient, used for offline determination of the residual echo suppression coefficient, the method comprising:

[0006] The frequency response curve of the playback link is determined based on the equalization parameters of the playback link of the electronic device. Based on the resonant characteristics of electronic devices, multiple frequency points to be processed are selected, and these multiple frequency points to be processed are divided into multiple frequency bands to be processed. Based on the frequency response curve, the residual echo suppression coefficients corresponding to multiple frequency bands to be processed are determined.

[0007] Optionally, determining the residual echo suppression coefficients corresponding to the multiple frequency bands to be processed based on the frequency response curve includes: Based on the frequency response curve, the average playback gain of the electronic device for each frequency band to be processed in the playback link is determined; The a priori gain is obtained by taking the smaller value of the average playback gain and 0, and then calculating the absolute value of the result. The prior gain is weighted and summed with the preset base gain to determine the residual echo suppression coefficient for each frequency band to be processed.

[0008] Optionally, determining the residual echo suppression coefficients corresponding to the multiple frequency bands to be processed based on the frequency response curve includes: Based on the calibration signal collected by the electronic device, the measured average gain of the electronic device in each of the frequency bands to be processed is determined; Based on the frequency response curve, the average playback gain of the electronic device for each frequency band to be processed in the playback link is determined; The a priori gain is obtained by taking the smaller value of the average playback gain and 0, and then calculating the absolute value of the result. The prior gain and the measured average gain are weighted and summed to determine the residual echo suppression coefficient for each frequency band to be processed.

[0009] Optionally, selecting multiple frequency points to be processed based on the resonant characteristics of the electronic device includes: Based on the resonant characteristics of electronic devices, frequency points with uniform response and far from the resonant peak are removed.

[0010] Optionally, dividing the multiple frequency points to be processed into multiple frequency bands to be processed includes: The number of frequency points contained in the multiple frequency bands to be processed is set to be positively correlated with the frequency corresponding to the frequency band to be processed, so that the bandwidth of the frequency band to be processed in the low frequency range is less than the bandwidth of the frequency band to be processed in the high frequency range.

[0011] Optionally, after determining the residual echo suppression coefficients corresponding to the plurality of frequency bands to be processed based on the frequency response curve, the method further includes: The residual echo suppression coefficients corresponding to the multiple frequency bands to be processed are mapped to Q8 linear suppression gains and stored.

[0012] Secondly, the present invention also provides a residual echo suppression method for online residual echo suppression, the method comprising: Based on the aforementioned method, the residual echo suppression coefficient is obtained; Obtain the online echo suppression coefficient of the electronic device; The online echo suppression coefficient and the residual echo suppression coefficient are fused to determine the fusion gain coefficient; Based on the fusion gain coefficient, gain control is performed on multiple frequency bands to be processed to achieve residual echo suppression.

[0013] Optionally, fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The fusion gain coefficient is determined by multiplying the online echo suppression coefficient by the residual echo suppression coefficient.

[0014] Optionally, fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The fusion gain coefficient is determined by taking the smaller of the online echo suppression coefficient and the residual echo suppression coefficient.

[0015] Optionally, fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The online echo suppression coefficient and the residual echo suppression coefficient are weighted and fused to determine the fusion gain coefficient; wherein, the weight of the online echo suppression parameter is positively correlated with the convergence degree of the online echo cancellation filter of the pickup link, and the weight of the residual echo suppression coefficient is negatively correlated with the convergence degree of the online echo cancellation filter of the pickup link.

[0016] Optionally, the gain control of the multiple frequency bands to be processed based on the fusion gain coefficient includes: Get the current probability of two lectures; The dual-talk probability is fused with the fusion gain coefficient to determine the gain control coefficient; wherein the gain control coefficient is positively correlated with the dual-talk probability.

[0017] Optionally, obtaining the residual echo suppression coefficient includes: Obtain the residual echo suppression coefficients of various devices in various scenarios, and select the corresponding residual echo suppression coefficient based on the current device and the current scenario.

[0018] In the technical solution provided by this invention, the frequency response curve of the playback link is obtained by using the equalization parameters of the playback link of the electronic device, and the residual echo suppression coefficient is determined based on the frequency response curve of the playback link. Since the frequency response curve of the playback link is prior knowledge of the acoustic path based on the known equalizer (EQ) configuration information of the electronic device, the deeper the frequency band of the equalizer in the playback link applies notch / negative gain, the stronger the cavity resonance / acoustic coupling identified by the designer is often. Therefore, based on this inherent static prior information of the electronic device, the frequency domain differential suppression strategy can be guided with almost zero computational overhead. This enables effective frequency domain suppression to be provided directly when the call starts or when the echo path changes abruptly, achieving high-efficiency, high-quality and low-overhead suppression of residual echo. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for determining the residual echo suppression coefficient according to an embodiment of the present invention; Figure 2 This is a flowchart of a residual echo suppression method according to another embodiment of the present invention; Figure 3 This is a spectrum diagram of the residual echo suppression method in each stage of audio processing according to another embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides a method for determining the residual echo suppression coefficient, used for offline determination of the residual echo suppression coefficient, such as... Figure 1 As shown, the method includes: The frequency response curve of the playback link is determined based on the equalization parameters of the playback link of the electronic device. In some embodiments, equalization parameters for the playback link are parsed from the audio profile (such as YAML format) of the electronic device. Each equalization segment consists of multiple cascaded bi-second-order filters, and the equalization parameters may include, for example, the parameters shown in the table below:

[0022] For a cascaded equalizer with N equalizer segments, the overall frequency response, i.e., the frequency response curve, is as follows: ; in, For the first The frequency response of the dual second-order filter is calculated according to the Audio EQ Cookbook formula.

[0023] Based on the resonant characteristics of electronic devices, multiple frequency points to be processed are selected, and these multiple frequency points to be processed are divided into multiple frequency bands to be processed. In some embodiments, in electronic devices, taking a typical embedded speech processing system as an example, the number of FFT points... Sampling rate Its frequency domain parameters are typically as follows: Nyquist frequency: ; Frequency resolution: ; Number of effective frequency points: One frequency point (frequency point 0~64); Since the device's cavity is primarily made of plastic, its acoustic resonance energy is mainly concentrated in the mid-frequency region (approximately 375~5625Hz). The DC component (frequency 0) and the high-frequency attenuation region (frequency 45+) do not require suppression parameters. This is because the cavity response is uniform and far from the resonance peak in the high-frequency range (>5625Hz), allowing for direct global suppression without the need for individual suppression of each frequency band. Therefore, the actual frequency range of interest is frequency 3~45, approximately 43 frequency points, covering the range of 375Hz~5625Hz.

[0024] Meanwhile, due to the high Q value of the plastic cavity resonance, the low-frequency resonance peak is narrow and steep, requiring fine frequency resolution; high-frequency resonance energy diffuses, so coarse resolution is sufficient; the very high frequency (>4kHz) cavity response tends to be flat, merging into a single broadband unified suppression. Therefore, when dividing the frequency band, a non-uniform frequency band division can be adopted—narrow low-frequency bandwidth and wide high-frequency bandwidth. For example, the frequency band division can be performed as follows:

[0025] Based on the frequency response curve, the residual echo suppression coefficients corresponding to multiple frequency bands to be processed are determined.

[0026] In some embodiments, since the frequency response curve of the playback link represents prior knowledge of the acoustic path implied by the known equalizer (EQ) configuration information of the electronic device, the residual echo suppression coefficients of multiple frequency bands to be processed can be determined with near-zero computational overhead based on the frequency response curve. Therefore, residual echoes can be suppressed directly at any time, without the suppression delay problem caused by the start of a call or scene changes.

[0027] In the technical solution provided by this invention, the frequency response curve of the playback link is obtained by using the equalization parameters of the playback link of the electronic device, and the residual echo suppression coefficient is determined based on the frequency response curve of the playback link. Since the frequency response curve of the playback link is prior knowledge of the acoustic path based on the known equalizer (EQ) configuration information of the electronic device, the deeper the frequency band of the equalizer in the playback link applies notch / negative gain, the stronger the cavity resonance / acoustic coupling identified by the designer is often. Therefore, based on this inherent static prior information of the electronic device, the frequency domain differential suppression strategy can be guided with almost zero computational overhead. This will enable effective frequency domain suppression to be provided directly when the call starts or when the echo path changes abruptly, achieving high-efficiency, high-quality and low-overhead suppression of residual echo.

[0028] As an optional implementation, determining the residual echo suppression coefficients corresponding to the multiple frequency bands to be processed based on the frequency response curve includes: Based on the frequency response curve, the average playback gain of the electronic device for each frequency band to be processed in the playback link is determined; The a priori gain is obtained by taking the smaller value of the average playback gain and 0, and then calculating the absolute value of the result. The prior gain is weighted and summed with the preset base gain to determine the residual echo suppression coefficient for each frequency band to be processed.

[0029] In some embodiments, the residual echo suppression coefficient can be derived directly from the cascaded frequency response curve of the playback link, completely omitting the calibration signal acquisition step. For example, it can be calculated using the following formula:

[0030] in: The equalizer cascade system for the playback link is in the first stage. The average gain (dB) of the frequency band is calculated offline using the frequency response formula of the dual second-order filter; This is the coupling amplification factor of the cavity structure of the electronic device (typical values ​​can be, for example, 1.5~2.0). Based on the prior suppression depth, it can provide guaranteed suppression for all frequency bands.

[0031] Because the deeper the frequency response curve of the equalizer in the playback link is suppressed in a certain frequency band (e.g., -6dB), the stronger the main resonance peak of the cavity structure of the electronic device, that is, the stronger the structural echo. For example, when the suppression depth of a certain frequency band is -6dB, with... For example, , Bonus, synergy The residual echo suppression coefficient is approximately 10 dB. When the suppression depth in a certain frequency band is 0 dB, it indicates that the cavity structure of the electronic device has no resonance, i.e., there is no structural echo. At this point, The residual echo suppression coefficient only contributes At this point, only a very weak bias is present. Furthermore, when the suppression depth in a certain frequency band is +4dB, it indicates that the cavity structure of the electronic device has a soft echo. The residual echo suppression coefficient only contributes At this point, it has only a very weak bias.

[0032] As an optional implementation, determining the residual echo suppression coefficients corresponding to the multiple frequency bands to be processed based on the frequency response curve includes: Based on the calibration signal collected by the electronic device, the measured average gain of the electronic device in each of the frequency bands to be processed is determined; Based on the frequency response curve, the average playback gain of the electronic device for each frequency band to be processed in the playback link is determined; The a priori gain is obtained by taking the smaller value of the average playback gain and 0, and then calculating the absolute value of the result. The prior gain and the measured average gain are weighted and summed to determine the residual echo suppression coefficient for each frequency band to be processed.

[0033] In some embodiments, besides deriving the residual echo suppression coefficient directly from the cascaded frequency response curve of the playback link as described above, the residual echo suppression coefficient can also be determined by combining the measured average gain and the playback average gain. For example, for the first... When the residual echo suppression coefficient for a frequency band is determined by both the measured average gain and the playback average gain, it can be calculated using the following formula:

[0034] in: The measured average gain (dB) of the acoustic path needs to be obtained by measuring using a calibration signal; The average gain (dB) for playback; This represents the coupling amplification factor of the cavity structure of the electronic device (typically 1.5 to 2.0). Specifically, it can be calculated using the following formula. :

[0035] in, and Let f be the power spectral density of the microphone and the far-end reference signal, respectively. By scanning the frequency, the power spectral density of a certain frequency band can be obtained. The specific scanning method can be implemented using any existing technology.

[0036] As an optional implementation, selecting multiple frequency points to be processed based on the resonant characteristics of the electronic device includes: Based on the resonant characteristics of electronic devices, frequency points with uniform response and far from the resonant peak are removed.

[0037] In some embodiments, the device, primarily made of plastic cavity, concentrates its acoustic resonance energy in the mid-frequency region (approximately 375~5625Hz). The DC component (frequency 0) and the high-frequency attenuation region (frequency 45+) do not require suppression parameters because the cavity response is uniform and far from the resonance peak in the high-frequency range (>5625Hz), allowing for direct global suppression. Therefore, the actual frequency range of interest is frequency 3~45, approximately 43 frequency points, covering 375~5625Hz.

[0038] As an optional implementation, dividing the plurality of frequency points to be processed into a plurality of frequency bands to be processed includes: The number of frequency points contained in the multiple frequency bands to be processed is set to be positively correlated with the frequency corresponding to the frequency band to be processed, so that the bandwidth of the frequency band to be processed in the low frequency range is less than the bandwidth of the frequency band to be processed in the high frequency range.

[0039] In some embodiments, the plastic cavity resonator has a high Q value, with a narrow and steep low-frequency resonant peak, requiring fine frequency resolution; while the high-frequency resonant energy diffuses, requiring only coarse resolution; the very high frequency (>4kHz) cavity response tends to be flat, merging into a single broadband unified suppression. Therefore, a non-uniform frequency band division method is used, making the bandwidth of the low-frequency band narrow and the bandwidth of the high-frequency band wide. For example, the following division method can be used:

[0040] As an optional implementation, after determining the residual echo suppression coefficients corresponding to the multiple frequency bands to be processed based on the frequency response curve, the method further includes: The residual echo suppression coefficients corresponding to the multiple frequency bands to be processed are mapped to Q8 linear suppression gains and stored.

[0041] In some embodiments, the residual echo suppression coefficient for each frequency band The linear suppression gain, directly converted to Q8 format in offline tools, is already an 8-bit Q8 value when written to the reserve field. Therefore, it can be used directly without looking up a table when performing residual echo suppression. In practical applications, if only the residual echo suppression coefficient is used to suppress residual echoes, then the residual echo suppression coefficient... When mapped to the Q8 linear suppression gain, it can be calculated using the following formula:

[0042] in, The Q8 linear suppression gain for frequency band m is... is the residual echo suppression coefficient for frequency band m.

[0043] Using the aforementioned calculation method, the dB value can be converted into Q8 linear suppression gain according to the following table:

[0044] After converting the residual echo suppression coefficient to Q8 linear suppression gain, the suppression result for a certain frequency band can be calculated using the following formula in practical applications:

[0045] in, For the output gain of a certain frequency band, For a certain frequency band, the input gain is... This represents the Q8 linear suppression gain for a certain frequency band.

[0046] This storage method allows for offline conversion from floating-point dB to Q8 linear suppression gain, avoiding floating-point exponential operations during online residual echo suppression and significantly reducing computational overhead. Specifically, during storage, the Q8 linear suppression gain can be carried within the SPEECH_ENHANCE_RT structure. In the encoding process, each `suppress_tbl[i]` field can be split into high 8 bits and low 8 bits, each carrying the Q8 linear suppression gain for two frequency bands (value range 1~255). Thus, only 5 int16 bytes are used to carry the Q8 linear suppression gain for 10 frequency bands, effectively saving storage space and transmission overhead.

[0047] This invention also provides a method for suppressing residual echoes online, such as... Figure 2 As shown, the method includes: Based on the aforementioned method, the residual echo suppression coefficient is obtained; In some embodiments, the residual echo suppression coefficient for each frequency band can be obtained by using the methods provided in the foregoing embodiments. This residual echo suppression coefficient is implemented offline and does not require consuming computing resources during operation.

[0048] Obtain the online echo suppression coefficient of the electronic device; In some embodiments, to more accurately suppress residual echoes, the online echo suppression coefficient of the electronic device can be obtained using a reference signal. Specifically, this can be achieved first by estimating the online echo likelihood based on coherence for the current frame. The For each frequency band, calculate the short-term coherence between the far-end reference signal and the error signal:

[0049] in: For the coherence between the reference signal and the error signal: This is the cross-power spectrum (smoothing estimate) of the reference signal and the error signal. As a reference signal power spectrum, The self-power spectrum of the error signal Then, the echo likelihood is derived from the coherence using the following formula:

[0050] Therefore, the online echo suppression coefficient can be calculated using the following formula:

[0051] in, The online echo suppression coefficient is... For echo similarity This is the minimum suppression coefficient of the online algorithm (e.g., 0.1, i.e., -20dB).

[0052] The online echo suppression coefficient and the residual echo suppression coefficient are fused to determine the fusion gain coefficient; In some embodiments, the online echo suppression coefficient and the residual echo suppression coefficient are fused to obtain a fused gain coefficient. The fused gain coefficient can effectively characterize the residual echo suppression coefficient calculated in the offline state and the online echo suppression coefficient obtained in the online state. Therefore, it can suppress residual echo more accurately.

[0053] Based on the fusion gain coefficient, gain control is performed on multiple frequency bands to be processed to achieve residual echo suppression.

[0054] In some embodiments, each frequency band to be processed has a corresponding fusion gain coefficient. By controlling the gain through the corresponding fusion gain coefficient, residual echo suppression can be effectively achieved.

[0055] As an optional implementation, fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The fusion gain coefficient is determined by multiplying the online echo suppression coefficient by the residual echo suppression coefficient.

[0056] In some embodiments, multiplication can be used to achieve dual confirmation of the online echo suppression coefficient and the residual echo suppression coefficient. Specifically, for example, the following formula can be used to calculate the fusion gain coefficient:

[0057] in, For frequency band The fusion gain coefficient, For frequency band The residual echo suppression coefficient (Q8 format, value range 16~256, the smaller the value, the stronger the suppression). This represents the online echo suppression coefficient.

[0058] By multiplying two coefficients, the residual echo suppression coefficient is strong ( Small) and strong online echo suppression coefficient ( When (small), the fusion gain coefficient This results in a very small value, thus achieving the strongest suppression; when the residual echo suppression coefficient is strong but the online echo suppression coefficient is weak (near-end speech), the fusion gain coefficient is... The residual echo suppression coefficient is moderate, thus protecting near-end speech; the residual echo suppression coefficient is weak. However, when the online echo suppression coefficient is strong, When the residual echo suppression coefficient is close to the online echo suppression coefficient value, residual echoes are suppressed normally; when both the residual echo suppression coefficient and the online echo suppression coefficient are weak, This means that almost no suppression of residual echo is required. This fusion method can avoid the "smoothing average" effect of dB-domain weighting. That is, in the cavity resonant frequency band (i.e. the frequency band with strong residual echo suppression coefficient), even with moderate linear coherence, multiplicative fusion can still provide strong suppression; conversely, if linear weighting is used in the dB domain, the directivity of the residual echo suppression coefficient is easily "averaged out".

[0059] As an optional implementation, fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The fusion gain coefficient is determined by taking the smaller of the online echo suppression coefficient and the residual echo suppression coefficient.

[0060] In some embodiments, the fusion gain coefficient can be calculated using, for example, the following formula:

[0061] The fusion gain coefficient is determined by taking the smaller value, and the more aggressive suppression between the online echo suppression coefficient and the residual echo suppression coefficient is selected to ensure that "strong suppression is executed whenever either dimension deems it necessary". This fusion method is suitable for scenarios with extremely low echo tolerance.

[0062] As an optional implementation, fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The online echo suppression coefficient and the residual echo suppression coefficient are weighted and fused to determine the fusion gain coefficient; wherein, the weight of the online echo suppression parameter is positively correlated with the convergence degree of the online echo cancellation filter of the pickup link, and the weight of the residual echo suppression coefficient is negatively correlated with the convergence degree of the online echo cancellation filter of the pickup link.

[0063] In some embodiments, confidence is introduced. To balance the online echo suppression coefficient and the residual echo suppression coefficient, a weighted summation method can be used, for example, as shown in the following formula:

[0064] Or equivalent geometric weighting:

[0065] in, The convergence of the online echo cancellation filter in the pickup link is dynamically adjusted, and when it does not converge... Approaching 1, meaning more trust After convergence, it makes Approaching the preset (Typical value 0.3~0.5). Specifically, it can be based on the convergence. Dynamically adjust confidence level :

[0066] in, The confidence level of the current frame k , These are preset values, which can be set based on empirical values. This represents the convergence level of the online echo cancellation filter in the picking link of the current frame k. In some preferred embodiments, The data can be stored in the Q15 format, for example, it can be stored together with the residual echo suppression coefficient stored in the aforementioned Q8 format.

[0067] Using the aforementioned fusion method, in When it approaches 0 (i.e., when it has not converged). Approaching 1, i.e., fully relying on the prior residual echo suppression coefficient; in When it approaches 1 (i.e., when it has converged). Then it tends to The prior residual echo suppression coefficient is used to assist the online echo suppression coefficient in residual echo suppression. Regarding convergence... For example, it can be evaluated using echo return attenuation enhancement, i.e., ELE, or any other existing technique.

[0068] As an optional implementation, the gain control of the multiple frequency bands to be processed based on the fusion gain coefficient includes: Get the current probability of two lectures; The dual-talk probability is fused with the fusion gain coefficient to determine the gain control coefficient; wherein the gain control coefficient is positively correlated with the dual-talk probability.

[0069] In some embodiments, dual-talk probability is employed. Modulating the fusion gain coefficients can effectively protect near-end speech:

[0070] in, This is the gain control coefficient. For dual-way protection strength. When When it approaches 1 (i.e. when it is certain that a double talk has occurred). It approaches 1 (i.e., the inhibition weakens).

[0071] As an optional implementation, obtaining the residual echo suppression coefficient includes: Obtain the residual echo suppression coefficients of various devices in various scenarios, and select the corresponding residual echo suppression coefficient based on the current device and the current scenario.

[0072] In some embodiments, different devices (phones / speakers / cameras) and different playback scenarios (speakers / earpieces / headphones) have different equalizer EQ configurations and acoustic paths, so it is necessary to maintain an independent residual echo suppression coefficient for each "device × scenario" combination; when the device switches playback scenarios (such as switching from speakers to headphones), the corresponding residual echo suppression coefficient is loaded immediately without the need for re-estimation.

[0073] In some preferred embodiments, if the electronic device supports online calibration, the calibration signal can be reacquired during idle periods to update the measured average gain.

[0074] in, For update rate, This is the updated measured average gain. This is the measured average gain before the update. This represents the measurement results during the current update process.

[0075] In some preferred embodiments, after residual echo suppression, time-domain smoothing is performed on each frequency band independently, and first-order IIR smoothing is applied to the fused suppression coefficients to avoid musical noise.

[0076] in, This represents the smoothing suppression coefficient of the current frame k in the m frequency band. , The time constant is used. Asymmetric smoothing is employed, meaning that a fast time constant is used when the gain decreases (suppressing enhancement). When the gain increases (suppression decreases), a slow time constant is used. This allows inhibition to take effect quickly, recovery to proceed slowly, and the respiratory effect to be eliminated.

[0077] In some preferred embodiments, after smoothing, comfort noise filling is also required. Specifically, comfort noise filling can be performed using, for example, the following formula:

[0078] in, The input audio for the current frame k in the m frequency band. For frequency domain comfortable noise, the variance is consistent with the near-end background noise estimate; This is the output audio data of the current frame k in the m frequency band.

[0079] In some preferred embodiments, for the foregoing embodiments, the application of the residual echo suppression coefficient can be controlled by an enable switch. When the enable switch is off, residual echo suppression can be performed using only the online echo suppression parameters. When the enable switch is on, residual echo suppression can be performed based on the result of fusing the residual echo suppression coefficient and the online echo suppression parameters, thereby achieving backward compatibility with existing equipment. Similarly, an enable switch can be set for the online echo suppression coefficient. When the enable switch is off, residual echo suppression can be performed using only the residual echo suppression coefficient. When the enable switch is on, residual echo suppression can be performed based on the result of fusing the residual echo suppression coefficient and the online echo suppression parameters.

[0080] In some preferred embodiments, the residual echo suppression coefficients obtained in the foregoing embodiments can also be used alone, that is, the device only stores the residual echo suppression coefficients, and does not need to have a calculation module for online echo suppression parameters.

[0081] like Figure 3 As shown, the spectrum diagrams of the residual echo suppression process in the aforementioned embodiments of the present invention are illustrated. The first row of the spectrum diagram is the reference signal, the second row is the microphone acquisition signal, and the third row is the result of linear echo cancellation. As can be seen from the graph in the third row, the residual echo within the frame is caused by strong nonlinearity of frequency band resonance. The fourth row is the result after residual echo cancellation using the aforementioned embodiments of the present invention. As can be seen from the graph in the fourth row, the residual echo within the frame is effectively eliminated.

[0082] In the technical solutions provided in the various embodiments of this invention, the residual echo suppression coefficient is determined by using the equalization parameters of the playback link as prior data. When the measured average gain is not used, a calibration signal is not required, effectively simplifying the calculation process of the residual echo suppression coefficient. When Q8 linear gain coefficients are used for storage, floating-point operations are not required for their application, effectively adapting to embedded electronic devices. Furthermore, since the residual echo suppression coefficient is primarily based on the equalization parameters of the playback link, it has good physical interpretability, providing a clear direction during debugging and optimization. In specific applications for residual echo suppression, the fusion of online echo suppression parameters and the residual echo suppression coefficient produces a dual confirmation effect, which is more frequency-domain directional than a simple linear weighting method. Moreover, since the residual echo suppression coefficient is obtained and stored offline, it can be applied at any time without any preconditions, thus providing effective frequency-domain prior suppression at the initial stage of a call or during scene switching. Because the aforementioned embodiments can maintain multiple independent residual echo suppression coefficients through device × scene combinations, they are naturally adaptable to multiple device platforms. Furthermore, the residual echo suppression coefficient in the aforementioned embodiments can be inserted as a post-processing module into any echo cancellation system, and the application of the residual echo suppression coefficient can be controlled by an enable switch, thereby achieving good compatibility with existing systems.

[0083] Those skilled in the art will understand that all or part of the processes in the above method embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0084] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included 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 method for determining the residual echo suppression coefficient, characterized in that, The method for determining the residual echo suppression coefficient offline includes: The frequency response curve of the playback link is determined based on the equalization parameters of the playback link of the electronic device. Based on the resonant characteristics of electronic devices, multiple frequency points to be processed are selected, and these multiple frequency points to be processed are divided into multiple frequency bands to be processed. Based on the frequency response curve, the residual echo suppression coefficients corresponding to multiple frequency bands to be processed are determined.

2. The method according to claim 1, characterized in that, The determination of residual echo suppression coefficients corresponding to multiple frequency bands to be processed based on the frequency response curve includes: Based on the frequency response curve, the average playback gain of the electronic device for each frequency band to be processed in the playback link is determined; The a priori gain is obtained by taking the smaller value of the average playback gain and 0, and then calculating the absolute value of the result. The prior gain is weighted and summed with the preset base gain to determine the residual echo suppression coefficient for each frequency band to be processed.

3. The method according to claim 1, characterized in that, The determination of residual echo suppression coefficients corresponding to multiple frequency bands to be processed based on the frequency response curve includes: Based on the calibration signal collected by the electronic device, the measured average gain of the electronic device in each of the frequency bands to be processed is determined; Based on the frequency response curve, the average playback gain of the electronic device for each frequency band to be processed in the playback link is determined; The a priori gain is obtained by taking the smaller value of the average playback gain and 0, and then calculating the absolute value of the result. The prior gain and the measured average gain are weighted and summed to determine the residual echo suppression coefficient for each frequency band to be processed.

4. The method according to claim 1, characterized in that, The selection of multiple frequency points to be processed based on the resonant characteristics of electronic devices includes: Based on the resonant characteristics of electronic devices, frequency points with uniform response and far from the resonant peak are removed.

5. The method according to claim 1, characterized in that, The step of dividing the multiple frequency points to be processed into multiple frequency bands to be processed includes: The number of frequency points contained in the multiple frequency bands to be processed is set to be positively correlated with the frequency corresponding to the frequency band to be processed, so that the bandwidth of the frequency band to be processed in the low frequency range is less than the bandwidth of the frequency band to be processed in the high frequency range.

6. The method according to claim 1, characterized in that, After determining the residual echo suppression coefficients corresponding to the multiple frequency bands to be processed based on the frequency response curve, the method further includes: The residual echo suppression coefficients corresponding to the multiple frequency bands to be processed are mapped to Q8 linear suppression gains and stored.

7. A method for suppressing residual echoes, characterized in that, The method for online residual echo suppression includes: Based on the method described in claims 1-6, the residual echo suppression coefficient is obtained; Obtain the online echo suppression coefficient of the electronic device; The online echo suppression coefficient and the residual echo suppression coefficient are fused to determine the fusion gain coefficient; Based on the fusion gain coefficient, gain control is performed on multiple frequency bands to be processed to achieve residual echo suppression.

8. The method according to claim 7, characterized in that, The step of fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The fusion gain coefficient is determined by multiplying the online echo suppression coefficient by the residual echo suppression coefficient.

9. The method according to claim 7, characterized in that, The step of fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The fusion gain coefficient is determined by taking the smaller of the online echo suppression coefficient and the residual echo suppression coefficient.

10. The method according to claim 7, characterized in that, The step of fusing the online echo suppression coefficient and the residual echo suppression coefficient to determine the fusion gain coefficient includes: The online echo suppression coefficient and the residual echo suppression coefficient are weighted and fused to determine the fusion gain coefficient; wherein, the weight of the online echo suppression parameter is positively correlated with the convergence degree of the online echo cancellation filter of the pickup link, and the weight of the residual echo suppression coefficient is negatively correlated with the convergence degree of the online echo cancellation filter of the pickup link.

Images