Howling suppression method and howling suppression device

Abstract

The invention provides a howling suppression method and a howling suppression device. The howling suppression method comprises the following steps: executing a frequency domain transformation operation on a microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal; dividing the frequency domain microphone signal into a plurality of sub-bands; detecting a first candidate howling frequency point where howling occurs by performing rough howling detection on the plurality of sub-bands; detecting a second candidate howling frequency point generating howling from a candidate howling band by executing fine howling detection on the candidate howling band comprising the first candidate howling frequency point; and performing howling suppression operation on the second candidate howling frequency point.

Description

Technical Field

[0001] The following description relates to the field of signal processing, and more specifically to a howling suppression method and howling suppression device. Background Technology

[0002] When the sound emitted by the device's speaker is transmitted directly or indirectly through the air and is recaptured and retransmitted or amplified by the device's microphone, a closed loop is formed between the microphone and the speaker, causing a feedback phenomenon.

[0003] Feedback can degrade audio quality and negatively impact user experience. Therefore, research is underway on how to address feedback. Summary of the Invention

[0004] According to one or more embodiments of the present disclosure, a howling suppression method is provided, the howling suppression method comprising: performing a frequency domain transformation operation on a microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal; dividing the frequency domain microphone signal into a plurality of sub-bands; detecting a first candidate howling frequency point from which howling occurs by performing coarse howling detection on the plurality of sub-bands; detecting a second candidate howling frequency point from the candidate howling band by performing fine howling detection on the candidate howling band including the first candidate howling frequency point; and performing a howling suppression operation on the second candidate howling frequency point.

[0005] In some embodiments, coarse howling detection may include howling detection based on a first howling feature, and fine howling detection may include howling detection based on a second howling feature, wherein the computational complexity of the first howling feature is lower than that of the second howling feature.

[0006] In some embodiments, the first howling feature may include peak-to-average power ratio and inter-frame peak duration, and the second howling feature may include peak-to-neighbor power ratio, peak-to-threshold power ratio, and inter-frame peak duration.

[0007] In some embodiments, the feedback suppression method may further include: acquiring a first speaker source signal and a second speaker source signal; before performing a frequency domain transformation operation, removing linear echoes corresponding to the first speaker source signal from the microphone signal by performing a first dual filter bank processing on the microphone signal based on a first dual filter bank and based on the first speaker source signal to generate a first microphone signal; and removing linear echoes corresponding to the second speaker source signal from the first microphone signal by performing a second dual filter bank processing on the first microphone signal based on a second dual filter bank and based on the second speaker source signal to generate a second microphone signal, wherein the operation of performing a frequency domain transformation operation on the microphone signal includes: performing a frequency domain transformation operation on the second microphone signal to generate a frequency domain microphone signal.

[0008] In some embodiments, one of the first speaker source signal and the second speaker source signal may correspond to the target speaker source signal, the dual filter group corresponding to the target speaker source signal in the first dual filter group and the second dual filter group may correspond to the target dual filter group, and the dual filter group processing corresponding to the target speaker source signal in the first dual filter group processing and the second dual filter group processing may correspond to the target dual filter group processing, wherein the target dual filter group includes a first adaptive filter and a second adaptive filter, wherein the target dual filter group processing includes a first-level adaptive filter processing based on the first adaptive filter and a second-level adaptive filter processing based on the second adaptive filter, wherein in the first-level adaptive filter processing, the filter coefficients of the first adaptive filter are dynamically updated through the filter coefficients of the second adaptive filter, and the linear echo corresponding to the target speaker source signal includes the signal output from the first adaptive filter that receives the target speaker source signal, and wherein in the second-level adaptive filter processing, the filter coefficients of the second adaptive filter are updated based on the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter, wherein the estimated feedback signal of the second adaptive filter includes the signal output from the second adaptive filter that receives the target speaker source signal, and the actual feedback signal of the second adaptive filter includes the microphone signal.

[0009] In some embodiments, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter decreases to a threshold error level, the filter coefficients of the second adaptive filter are passed to the filter coefficients of the first adaptive filter.

[0010] In some embodiments, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter is greater than a threshold, the filter coefficients of the first adaptive filter can be passed to the filter coefficients of the second adaptive filter.

[0011] In some embodiments, the step of performing a howling suppression operation on a second candidate howling frequency point may include: applying a first gain width to the second candidate howling frequency point within a first frequency range based on the second candidate howling frequency point; and applying a second gain width to the second candidate howling frequency point within a second frequency range based on the second candidate howling frequency point, wherein the frequency in the first frequency range is less than the frequency in the second frequency range, and wherein the first gain width is less than the second gain width.

[0012] In some embodiments, a candidate howling band may include a first candidate howling subband and a subband adjacent to the first candidate howling subband, wherein the first candidate howling subband includes a first candidate howling frequency point.

[0013] According to another aspect of one or more embodiments of the present disclosure, a howling suppression device is provided, the howling suppression device comprising: at least one processor and at least one memory storing one or more instructions, the one or more instructions being configured, when executed by the at least one processor, to cause the at least one processor to: perform a frequency domain transformation operation on a microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal; divide the frequency domain microphone signal into a plurality of sub-bands; detect a first candidate howling frequency point from which howling occurs by performing coarse howling detection on the plurality of sub-bands; detect a second candidate howling frequency point from the candidate howling band by performing fine howling detection on the candidate howling band including the first candidate howling frequency point; and perform a howling suppression operation on the second candidate howling frequency point.

[0014] In some embodiments, coarse howling detection may include howling detection based on a first howling feature, and fine howling detection may include howling detection based on a second howling feature, wherein the computational complexity of the first howling feature is lower than that of the second howling feature.

[0015] In some embodiments, the first howling feature may include peak-to-average power ratio and inter-frame peak duration, and the second howling feature may include peak-to-neighbor power ratio, peak-to-threshold power ratio, and inter-frame peak duration.

[0016] In some embodiments, the at least one processor may further be configured to: acquire a first speaker source signal and a second speaker source signal; before performing a frequency domain transformation operation, remove linear echoes corresponding to the first speaker source signal from the microphone signal by performing a first dual filter bank processing on the microphone signal based on a first dual filter bank and based on the first speaker source signal to generate a first microphone signal; and remove linear echoes corresponding to the second speaker source signal from the first microphone signal by performing a second dual filter bank processing on the first microphone signal based on a second dual filter bank and based on the second speaker source signal to generate a second microphone signal, wherein the at least one processor is configured to: perform a frequency domain transformation operation on the second microphone signal to generate a frequency domain microphone signal.

[0017] In some embodiments, one of the first speaker source signal and the second speaker source signal may correspond to the target speaker source signal, the dual filter group corresponding to the target speaker source signal in the first dual filter group and the second dual filter group may correspond to the target dual filter group, and the dual filter group processing corresponding to the target speaker source signal in the first dual filter group processing and the second dual filter group processing may correspond to the target dual filter group processing, wherein the target dual filter group includes a first adaptive filter and a second adaptive filter, wherein the target dual filter group processing includes a first-level adaptive filter processing based on the first adaptive filter and a second-level adaptive filter processing based on the second adaptive filter, wherein in the first-level adaptive filter processing, the filter coefficients of the first adaptive filter are dynamically updated through the filter coefficients of the second adaptive filter, and the linear echo corresponding to the target speaker source signal includes the signal output from the first adaptive filter that receives the target speaker source signal, wherein in the second-level adaptive filter processing, the filter coefficients of the second adaptive filter are updated based on the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter, wherein the estimated feedback signal of the second adaptive filter includes the signal output from the second adaptive filter that receives the target speaker source signal, and the actual feedback signal of the second adaptive filter includes a microphone signal.

[0018] In some embodiments, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter decreases to a threshold error level, the filter coefficients of the second adaptive filter can be passed to the filter coefficients of the first adaptive filter.

[0019] In some embodiments, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter is greater than a threshold, the filter coefficients of the first adaptive filter can be passed to the filter coefficients of the second adaptive filter.

[0020] In some embodiments, the at least one processor may be configured to: apply a first gain width to a second candidate howling frequency point within a first frequency range based on the second candidate howling frequency point; and apply a second gain width to the second candidate howling frequency point within a second frequency range based on the second candidate howling frequency point, wherein the frequency in the first frequency range is less than the frequency in the second frequency range, and the first gain width is less than the second gain width.

[0021] In some embodiments, a candidate howling band may include a first candidate howling subband and a subband adjacent to the first candidate howling subband, wherein the first candidate howling subband includes a first candidate howling frequency point.

[0022] In some embodiments, the at least one processor may be configured to detect a first candidate howling frequency by detecting the peak value of the power spectrum of a frequency domain microphone signal.

[0023] According to another aspect of one or more embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform any of the howling suppression methods described above. Attached Figure Description

[0024] Figure 1 A flowchart illustrating a howling suppression method according to an example embodiment of the present disclosure is shown.

[0025] Figure 2 A flowchart illustrating a howling suppression method according to an example embodiment of the present disclosure is shown.

[0026] Figure 3 A flowchart illustrating dual filter bank processing according to an exemplary embodiment of this disclosure is shown.

[0027] Figure 4 A flowchart illustrating the target dual filter bank processing according to an exemplary embodiment of the present disclosure is shown.

[0028] Figure 5 A flowchart illustrating a method for spectrum analysis according to an example embodiment of the present disclosure is shown.

[0029] Figure 6 A flowchart illustrating a method for detecting howling frequencies according to an example embodiment of the present disclosure is shown.

[0030] Figure 7 A schematic diagram illustrating the division of a spectrum into multiple subbands according to an exemplary embodiment of the present disclosure is shown.

[0031] Figure 8 A schematic diagram illustrating the extended detection spectral range according to an example embodiment of the present disclosure.

[0032] Figure 9 A flowchart illustrating howling suppression operation according to an example embodiment of the present disclosure is shown.

[0033] Figure 10 A schematic diagram illustrating a candidate howling gain according to an example embodiment of the present disclosure is shown.

[0034] Figure 11 A schematic diagram of a howling suppression method according to an example embodiment of the present disclosure is shown.

[0035] Figure 12 A block diagram of a howling suppression device according to an example embodiment of the present disclosure is shown. Detailed Implementation

[0036] The invention will be described more fully below with reference to the accompanying drawings, in which exemplary embodiments of the invention are illustrated. However, the invention may be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0037] It should be understood that although the terms first, second, third, etc., may be used herein to describe different elements, components, regions, layers, and / or portions, these elements, components, regions, layers, and / or portions should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer, or portion from another. Therefore, without departing from the teachings of the invention, the “first” element, component, region, layer, or portion discussed below may be referred to as a “second” element, component, region, layer, or portion. As used herein, the term “and / or” includes any and all combinations of one or more of the associated listed items.

[0038] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that when the terms “comprising” and / or “including” are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof.

[0039] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It will also be understood that, unless expressly defined herein, terms (such as those defined in a general dictionary) shall be interpreted as having the same meaning as they have in the context of the relevant field, and shall not be interpreted in an ideal or overly formal sense.

[0040] Figure 1 A flowchart illustrating a howling suppression method according to an example embodiment of the present disclosure is shown.

[0041] Reference Figure 1 In operation S110, a frequency domain transformation operation can be performed on the microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal.

[0042] A microphone signal can be a signal captured by a microphone in an electronic device. The electronic device can be any electronic device with a microphone (e.g., a mobile phone, computer, hearing aid, etc.).

[0043] Various frequency domain transformation methods can be used to perform frequency domain transformation operations on microphone signals. These methods include, but are not limited to, Fourier Transform (FT), Short-Time Fourier Transform (STFT), Wavelet Transform (WT), Hilbert Transform (HT), and Hilbert-Huang Transform (HHT).

[0044] In operation of S120, the frequency domain microphone signal can be divided into multiple sub-bands.

[0045] Various methods can be used to divide a frequency domain microphone signal into multiple sub-bands. In one example, the frequency domain microphone signal can be uniformly divided into a certain number of sub-bands. The number of sub-bands can be a predetermined number. In another example, the frequency domain microphone signal can be non-uniformly divided into a certain number of sub-bands. However, the above examples are merely illustrative, and this disclosure is not limited thereto, and the method for dividing the frequency domain microphone signal can be set according to other division criteria.

[0046] In operation S130, the first candidate howling frequency point from multiple subbands can be detected by performing coarse howling detection on multiple subbands.

[0047] In this disclosure, a "candidate howling frequency" can be a frequency that may experience howling, but is not necessarily a frequency that actually experiences howling. A first candidate howling frequency may include one or more candidate howling frequencies. The first candidate howling frequency can be used to determine candidate howling subbands where howling occurs (or may occur).

[0048] Performing coarse howling detection over multiple subbands means detecting, using relatively coarse detection criteria, which subbands among multiple subbands might have one or more first candidate howling frequencies where howling is likely to occur. In other words, coarse howling detection can be used to quickly locate the spectral range where howling occurs or is likely to occur (e.g., candidate howling bands). For example, coarse howling detection can be performed over multiple subbands by detecting the peak value of the power spectrum of a frequency-domain microphone signal to detect first candidate howling frequencies.

[0049] In operation S140, fine-grained howling detection can be performed on a candidate howling band that includes a first candidate howling frequency point to detect a second candidate howling frequency point that is howling. The candidate howling band may include a first candidate howling sub-band, and the first candidate howling sub-band includes the first candidate howling frequency point. For example, in some embodiments, the candidate howling band may include one or more sub-bands, and the one or more sub-bands may include the first candidate howling sub-band. The first candidate howling frequency point is included in the first candidate howling sub-band among multiple sub-bands.

[0050] Since the second candidate howling frequency can be detected by fine howling detection within a narrowed computational range (i.e., including the candidate howling band of the first candidate howling subband corresponding to the first candidate howling frequency determined in operation S130, rather than all of the multiple subbands), the computational load is reduced and the accuracy of the detection of the second candidate howling frequency is guaranteed during the detection of the final second candidate howling frequency.

[0051] Performing fine-grained howling detection on candidate howling bands means using relatively refined detection criteria to detect potential second candidate howling frequencies within the candidate howling bands. The accuracy of fine-grained howling detection in detecting the second howling frequency can be greater than the accuracy of coarse howling detection in detecting the first howling frequency. In one embodiment, the computational complexity or computational cost of coarse howling detection in detecting the first howling frequency can be lower or less than the computational complexity or computational cost of fine-grained howling detection in detecting the second howling frequency.

[0052] According to at least one example embodiment, coarse howling detection may include howling detection based on a first howling feature, and fine howling detection may include howling detection based on a second howling feature. The computational complexity of the first howling feature may be lower than that of the second howling feature. Because the first howling feature with relatively low computational complexity is used in coarse howling detection, and the second howling feature is used in fine howling detection, the computational cost of coarse howling detection can be reduced while ensuring the accuracy of fine howling detection.

[0053] The first and second howling features can be characteristics that characterize howling and can be used to detect howling. The first and second howling features can be selected from various howling features. These various howling features may include, but are not limited to, Peak-to-Threshold Power Ratio (PTPR), Peak-to-Average Power Ratio (PAPR), Peak-to-Neighboring Power Ratio (PNPR), Peak-to-Harmonics Power Ratio (PHPR), Interframe Peak Magnitude Persistence (IPMP), and Interframe Magnitude Slope Deviation (IMSD).

[0054] According to at least one example embodiment, the first howling feature may include peak-to-average power ratio and inter-frame peak durability, and the second howling feature includes peak-to-neighbor power ratio, peak-to-threshold power ratio, and inter-frame peak durability. For example, in one embodiment, the first howling feature may include fewer howling features than the second howling feature. The combination of the first and second howling features according to the above at least one example embodiment can effectively reduce the computational cost of coarse howling detection and ensure the accuracy of fine howling detection, thereby achieving less computation and higher accuracy in the overall howling detection.

[0055] According to at least one example embodiment, a candidate howling band may include a first candidate howling sub-band and a sub-band adjacent to the first candidate howling sub-band. Since the candidate howling band to be detected may include the first candidate howling sub-band and a sub-band adjacent to the first candidate howling sub-band, inaccurate howling detection caused by the howling frequency point being located at the sub-band boundary can be avoided. For example, in one embodiment, a candidate howling band may include a first candidate howling sub-band and a sub-band immediately adjacent to the first candidate howling sub-band.

[0056] In operation S150, a howling suppression operation can be performed on the second candidate howling frequency point.

[0057] Howling suppression can be performed on the second candidate howling frequency using various methods. These methods may include, but are not limited to, phase modulation, gain control, and adaptive feedback cancellation.

[0058] According to at least one example embodiment, howling suppression can be performed within a first frequency range based on a second candidate howling frequency point by applying a first gain width to the second candidate howling frequency point, and howling suppression can be performed within a second frequency range based on the second candidate howling frequency point by applying a second gain width to the second candidate howling frequency point. The frequencies in the first frequency range can be lower than the frequencies in the second frequency range, and the first gain width can be lower than the second gain width. Since different gain widths can be set according to the frequency range of the howling, damage to the desired signal can be minimized, and effective howling suppression can be achieved.

[0059] Figure 2 A flowchart illustrating a howling suppression method according to an example embodiment of the present disclosure is shown.

[0060] In operation S210, the first speaker source signal and the second speaker source signal can be acquired.

[0061] The first speaker source signal and the second speaker source signal can be the source signal corresponding to the first speaker and the source signal corresponding to the second speaker of the electronic device, respectively. Since the electronic device can include two speakers (i.e., the first speaker and the second speaker), it can be a stereo speaker or a stereo speaker system. However, the number of speakers in the electronic device can be more than two. Stereo speaker devices or stereo speaker systems are generally susceptible to feedback. In particular, closed-loop feedback can occur severely when multiple stereo speaker devices (e.g., mobile devices) in a room simultaneously play audio, such as during live karaoke streaming or gaming communications.

[0062] In operation S220, the linear echo corresponding to the first speaker source signal in the microphone signal can be removed by performing a first dual-filter bank processing based on the first dual-filter bank on the microphone signal, thereby generating the first microphone signal. For example, the first dual-filter bank processing can apply the first dual-filter bank to the microphone signal.

[0063] According to at least one example embodiment, the first dual filter bank may include two adaptive filters (ADFs) (e.g., corresponding first adaptive filter and second adaptive filter). The adaptive filter may minimize the mean square error between the desired signal and the output signal by continuously adjusting the filter parameters, based on the statistical properties of the signal and the minimum mean square error criterion, assuming that the input signal consists of a desired signal and noise.

[0064] The first dual filter bank processing may include a first-level adaptive filter processing based on a first adaptive filter in the first dual filter bank and a second-level adaptive filter processing based on a second adaptive filter in the first dual filter bank.

[0065] In the first-stage adaptive filter processing, the filter coefficients of the first adaptive filter can be dynamically updated using the filter coefficients of the second adaptive filter, and the linear echo corresponding to the first speaker source signal can include the signal output from the first adaptive filter that receives the first speaker source signal. In the second-stage adaptive filter processing, the filter coefficients of the second adaptive filter are updated based on the error between the estimated feedback signal and the actual feedback signal of the second adaptive filter. The estimated feedback signal of the second adaptive filter includes the signal output from the second adaptive filter that receives the first speaker source signal, and the actual feedback signal of the second adaptive filter includes the microphone signal.

[0066] According to at least one example embodiment, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter decreases to a threshold error level, the filter coefficients of the second adaptive filter are passed to the filter coefficients of the first adaptive filter. The threshold error level can be preset. Since the filter coefficients of the second adaptive filter when the error decreases can be passed to or copied to the filter coefficients of the first adaptive filter, a good suppression effect on the linear echo corresponding to the first loudspeaker can be achieved by using the first adaptive filter having the filter coefficients of the second adaptive filter when the error decreases.

[0067] According to at least one example embodiment, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter is greater than a preset threshold, the filter coefficients of the first adaptive filter are transferred to the filter coefficients of the second adaptive filter. Since the filter coefficients of the first adaptive filter can be transferred or copied to the filter coefficients of the second adaptive filter when the error of the filter coefficients of the second adaptive filter is greater than the preset threshold, the stability of the second adaptive filter can be maintained even when the error of the filter coefficients of the second adaptive filter is too large, thereby improving the robustness of the algorithm.

[0068] In operation S230, the linear echo corresponding to the second speaker source signal in the first microphone signal can be removed by performing a second dual-filter bank processing based on the second dual-filter bank on the first microphone signal, thereby generating the second microphone signal. For example, the second dual-filter bank processing can apply the second dual-filter bank to the first microphone signal.

[0069] For example, the linear echo in the first microphone signal corresponding to the first speaker source signal and the linear echo in the second speaker source signal can constitute the howling linear component in the echo path of stereo sound.

[0070] According to at least one example embodiment, the second dual filter bank may include two adaptive filters (ADFs) (e.g., corresponding first adaptive filter and second adaptive filter).

[0071] The second dual filter bank processing may include first-level adaptive filter processing based on the first adaptive filter in the second dual filter bank and second-level adaptive filter processing based on the second adaptive filter in the second dual filter bank.

[0072] In the first-stage adaptive filter processing, the filter coefficients of the first adaptive filter can be dynamically updated using the filter coefficients of the second adaptive filter, and the linear echo corresponding to the first speaker source signal can include the signal output from the first adaptive filter that receives the first speaker source signal. In the second-stage adaptive filter processing, the filter coefficients of the second adaptive filter are updated based on the error between the estimated feedback signal and the actual feedback signal of the second adaptive filter. The estimated feedback signal of the second adaptive filter includes the signal output from the second adaptive filter that receives the first speaker source signal, and the actual feedback signal of the second adaptive filter includes the microphone signal.

[0073] According to at least one example embodiment, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter decreases, the filter coefficients of the second adaptive filter are passed to the filter coefficients of the first adaptive filter. Since the filter coefficients of the second adaptive filter when the error decreases can be passed to or copied to the filter coefficients of the first adaptive filter, a good suppression effect on the linear echo corresponding to the first loudspeaker can be achieved by using the first adaptive filter having the filter coefficients of the second adaptive filter when the error decreases.

[0074] According to at least one example embodiment, when the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter is greater than a preset threshold, the filter coefficients of the first adaptive filter are transferred to the filter coefficients of the second adaptive filter. Since the filter coefficients of the first adaptive filter can be transferred or copied to the filter coefficients of the second adaptive filter when the error of the filter coefficients of the second adaptive filter is greater than the preset threshold, the stability of the second adaptive filter can be maintained even when the error of the filter coefficients of the second adaptive filter is too large, thereby improving the robustness of the algorithm.

[0075] As described above, the first dual filter bank processing in operation S220 and the second dual filter bank processing in operation S230 are essentially the same.

[0076] exist Figure 2 In the case of the embodiments, Figure 1 Operation S110 may include: performing a frequency domain transformation operation on the second microphone signal to generate a frequency domain microphone signal. That is, in one embodiment, Figure 2 The operations shown can be performed in Figure 1 The operation performed before the operation.

[0077] According to this disclosure Figure 1 and Figure 2 The feedback suppression method shown in the example embodiment can effectively remove the linear echo corresponding to the speaker source signal in the microphone signal by performing dual filter bank processing on each speaker source signal when dealing with a stereo speaker device or stereo speaker system.

[0078] Figure 3 A flowchart illustrating dual filter bank processing according to an exemplary embodiment of this disclosure is shown.

[0079] Reference Figure 3 The microphone signal is shown. First speaker source signal Second speaker source signal .

[0080] like Figure 3 As shown, microphone signal and the first speaker source signal It can be input to a first dual filter bank. According to at least one example embodiment, the first dual filter bank may include two adaptive filters.

[0081] The signal processed by the first dual filter bank and the second speaker source signal It can be input to a second dual filter bank. According to at least one example embodiment, the second dual filter bank may include two adaptive filters. A desired signal can be output from the second dual filter bank. Expected signal It can correspond to Figure 2 The second microphone signal.

[0082] The following will combine Figure 4 The target dual filter bank processing corresponding to one of the first and second dual filter banks is described in more detail.

[0083] Figure 4A flowchart illustrating the target dual filter bank processing according to an exemplary embodiment of the present disclosure is shown.

[0084] Assuming microphone signal The expected signal is already included. and the speaker signal in the feedback loop Speaker signal It can be the response of the speaker to the externally played original audio signal (i.e., the speaker source signal). The output signal. For example, in the case of two speakers, the speaker signal... It may include the response of the first loudspeaker to the first loudspeaker source signal. The output of the first speaker signal and the response of the second speaker to the second speaker source signal The output of the second speaker signal Microphone signal Through equations To determine.

[0085] Reference Figure 4 It can transmit the speaker source signal (Also known as a stereo remote signal source) According to the cascade structure and microphone signal (Also known as near-end microphone signal) ) Perform target dual filter bank processing.

[0086] First, the signal from the first speaker can be... Execute the processing. This processing may include first-level ADF processing corresponding to the first-level ADF and second-level ADF processing corresponding to the second-level ADF.

[0087] The first-stage ADF does not perform adaptive processing independently, but the filter coefficients of the first-stage ADF can be dynamically updated in real time by the second-stage ADF, and the error signal of the first-stage ADF can be transmitted to subsequent modules as the system output.

[0088] The second-level ADF can perform coefficient updates in real time, for example, but the error of the second-level ADF is... It is not output by the system internally. It can first be done via a system based on... Apply convolution processing To obtain the estimated feedback signal . This represents an estimate of the actual feedback path. Then, it can be expressed using the equation... Calculate and estimate the feedback signal and actual feedback signals Error between During the adaptive process, this error... The coefficients will gradually approach zero. The least mean square error algorithm can be used to update the ADF coefficients here. For a fixed step size coefficient ρ, the larger the value, the faster the coefficient update; the smaller the value, the slower the convergence speed. With the equation... As the number of iterations increases, The value will approach the actual feedback path w infinitely.

[0089] Reference Figure 4 In operation S410, it can be determined whether the second-level ADF is diverging. When the second-level ADF diverges, the energy of the error signal of the second-level ADF increases sharply. According to at least one example embodiment, in order to determine the stationarity of the error signal output by the second-level ADF, it can be determined whether the error signal is greater than a threshold. If it is determined in operation S410 that the second-level ADF is diverging (S410, Yes), the stability coefficients stored in the first-level ADF can be copied or transferred back to maintain the stability of the second-level ADF, thereby improving the robustness of the algorithm.

[0090] If it is determined in operation S410 that the second-level ADF is not diverging (S410, No), it can be determined whether to update the filter coefficients of the first-level ADF. According to at least one example embodiment, the feedback suppression capabilities of the first-level ADF and the second-level ADF can be compared in each iteration of the algorithm. The filter coefficients of the second-level ADF are updated to match those of the first-level ADF only when the suppression effect of the second-level ADF is better than that of the first-level ADF, i.e., the error decreases (e.g., decreases to the threshold error level). When the suppression effect of the second-level ADF is better than that of the first-level ADF, i.e., the error decreases (e.g., S420, Yes), the filter coefficients of the second-level ADF are updated to match those of the first-level ADF. Otherwise (S420, No), the filter coefficients are not updated. Therefore, the second-level ADF can be configured to have better suppression performance.

[0091] In response to the first speaker signal After the processing is completed, it can be done through the equation. To process the second speaker signal in series Final error This serves as the final output of the second dual filter bank for the next stage.

[0092] Figure 5 A flowchart illustrating a method for spectrum analysis according to an example embodiment of the present disclosure is shown.

[0093] Frequency domain transformation operations can be performed on microphone signals using various methods. For example, the Fast Fourier Transform (FFT) can be used to perform frequency domain transformation operations on microphone signals. Figure 5In the example, the frequency domain transformation operation of the microphone signal can be performed using the short-time Fourier transform (STFT). However, it should be understood that the frequency domain transformation method of this disclosure is not limited to the short-time Fourier transform, and can be any other frequency domain transformation method.

[0094] Reference Figure 5 In operating the S510, window analysis can be performed on the microphone signal to determine which window function to use. w(n) To process microphone signals. Window function. w(n) It can be used to reduce spectral leakage. Window function. w(n) It can be selected and set in various ways.

[0095] In operation S520, a short-time Fourier transform can be performed on the microphone signal to generate a frequency-domain microphone signal. For example, this can be based on the equation corresponding to the short-time Fourier transform. To generate frequency domain microphone signals. It is the k-th frequency sample, and It is a frame index. The power spectrum at the k-th frequency is Considering the impact of windowing on amplitude, a scaling factor needs to be used. To perform amplitude correction. However, the above equation is exemplary, and in some embodiments, other equations may be used to perform Fourier transform.

[0096] In operation of the S530, the auto-power spectrum (APS) of the frequency domain microphone signal can be calculated. The amplitude at frequency points on one side of the spectrum will be used to calculate the final signal power. .

[0097] In operating the S540, the peak value of the frequency domain microphone signal can be found by analyzing its self-power spectrum. Through equations To find the peak power. It can be used to detect candidate whistling frequency points in subsequent modules.

[0098] Figure 6 A flowchart illustrating a method for detecting howling frequencies according to an example embodiment of the present disclosure is shown.

[0099] Reference Figure 6 To reduce the complexity of detection, the spectrum of the frequency domain microphone signal can be divided into multiple sub-bands and the frequency range in which feedback occurs can be preliminarily detected.

[0100] In operation of S610, the spectrum of the frequency domain microphone signal can be divided into multiple sub-bands.

[0101] The spectrum of a frequency-domain microphone signal can be divided into multiple sub-bands using any partitioning method. For example, just as an example, Figure 7 As shown, the fixed frequency width of each sub-band can be set to 500Hz. Therefore, the 8kHz spectrum can be evenly divided into 16 sub-bands. The spectrum of each sub-band can be represented as... ,in, W and S represent the sub-band width and the number of sub-bands, respectively. W can be expressed by the equation... To determine, and S can be determined by the equation To determine.

[0102] In operation S621, the peak-to-average power ratio can be calculated for each of the multiple sub-bands. .

[0103] PAPR is a time feature based on counting how many frames of frequencies in past signal frames fall within a candidate whistling frequency set. The formula for calculating PAPR is: ,in, If PAPR exceeds a predefined threshold, the i-th candidate whistling component will be identified as a whistling component. That is, PAPR feature is probably the most widely used feature in whistling detection, and different thresholds are used based on estimates of the background noise spectrum, source signal spectrum, reverberation time, and acoustic feedback path response.

[0104] In operation of S622, the peak-to-average power ratio of each sub-band can be used as a basis. To determine the peak-to-mean power ratio with a relatively large value.

[0105] According to at least one example embodiment, for subband detection, the equation can be used. To perform a simple whistling detection method, where, The calculation results can be entered into... middle.

[0106] In operation S623, the peak-to-average power ratio with a relatively large value in the subband (e.g., the peak-to-average power ratio with the maximum value) can be compared with a first threshold Th1. The first threshold Th1 can be preset.

[0107] When the peak-to-average power ratio with a relatively large value is greater than the first threshold Th1 (S623, yes), the frequency point corresponding to the peak-to-average power ratio with a relatively large value can be determined in operation S624.

[0108] In operation S625, the inter-frame peak duration can be calculated for the frequency point corresponding to the peak-to-average power ratio with a relatively large value. Inter-frame peak duration is a time-dependent characteristic based on counting how many frames of frequencies in past signal frames fall within the candidate howling frequency set, and can be calculated using the equation... It means that, among them, Denotes the set of candidate whistling frequencies, and for example This indicates that four frames are counted consecutively. For example, this equation can be used in subband detection. Use statistical intersection to simplify calculations.

[0109] In operation S626, it is possible to determine whether the frequency point corresponding to the relatively large peak-to-average power ratio is the first candidate howling frequency point.

[0110] In the above description, operations S621 to S626 can be referred to as coarse howling detection based on a first howling feature, and the first howling feature includes peak-to-mean power ratio and inter-frame peak duration. However, the above description is merely exemplary, and the first howling feature can include various howling features.

[0111] According to at least one example embodiment of this disclosure, in a coarse howling detection phase based on a first howling feature, simultaneous howling and speech are simulated. When acoustic feedback occurs, both spectral and temporal features are used to distinguish between howling and speech. Conversely, if no acoustic feedback occurs, coarse howling detection may yield erroneous results (i.e., coarse howling detection merely determines that howling may occur), so the next step is to reduce detection error.

[0112] In operation S631, the power peak value of the candidate howling band, including the first candidate howling subband, can be determined. The first candidate howling frequency point is included in the first candidate howling subband among multiple subbands.

[0113] For example, the first stage of acquisition The results can indicate This indicates that a feedback loop may occur in the second sub-band. A candidate feedback loop can be configured to include the second sub-band. According to at least one example embodiment, to avoid feedback frequencies located at sub-band boundaries, a candidate feedback loop can be configured to include a first candidate feedback sub-band and a sub-band adjacent to the first candidate feedback sub-band. For example, as... Figure 8 As shown, the detected spectral range can be extended, for example, by three times the width. The detected spectrum can be represented as... ,in, ,and .

[0114] In operation S632, the peak neighbor-to-neighbor power ratio of the frequency points in the candidate howling band can be calculated based on the peak power of the candidate howling band.

[0115] Peak neighbor-to-neighbor power ratio can be expressed as Where m represents the number of adjacent frequency points, This indicates the frequency point located at the peak.

[0116] In operation S633, the peak threshold power ratio of the frequency points in the candidate howling band can be calculated based on the power peak of the frequency points in the candidate howling band.

[0117] Peak-threshold power ratio can be expressed as .

[0118] In operation S634, the calculated peak neighbor-to-neighbor power ratio can be compared with a second threshold Th2. The second threshold Th2 can be preset.

[0119] In operation S635, the calculated peak threshold power ratio can be compared with a third threshold Th3. The third threshold Th3 can be preset.

[0120] When the peak neighbor power of a frequency point in a candidate howling band is determined to be greater than the second threshold Th2 (S634, yes) and the peak threshold power ratio of a frequency point in a candidate howling band is greater than the third threshold Th3 (S635, yes), the frequency point in the candidate howling band can be set as a candidate howling frequency point in operation S636.

[0121] In operation S637, inter-frame peak persistence can be calculated for candidate howling frequency points.

[0122] Here, the inter-frame peak durability can be expressed by the following equation: .

[0123] In operation S638, the candidate howling frequency point can be determined as the second candidate howling frequency point where howling occurs based on the inter-frame peak duration.

[0124] Here, the second candidate howling frequency can be considered as the frequency at which howling occurs.

[0125] According to at least one example embodiment, when using inter-frame peak persistence, signals that are easily confused with howling, such as trills and whistles, are not detected, and only a general subband range is detected. Therefore, IPMP can avoid false detections, thereby improving the accuracy of detection.

[0126] In the above description, operations S631 to S638 can be referred to as fine-grained howling detection based on a second howling feature, and the second howling feature includes peak neighbor-to-neighbor power ratio, peak threshold power ratio, and inter-frame peak persistence. However, the above description is merely exemplary, and the second howling feature can include various howling features. The computational complexity of the first howling feature may be lower than that of the second howling feature.

[0127] Figure 9 A flowchart illustrating howling suppression operation according to an example embodiment of the present disclosure is shown.

[0128] Candidate List The system stores all currently detected howling frequencies. In operation S910, it is determined whether the detected howling frequency exists in the candidate list. If it is determined that the detected howling frequency is in the candidate list (S910, yes), the process can proceed to the following operation S940, and the candidate howling gain can be calculated.

[0129] If it is determined that the detected howling frequency is not in the candidate list (S910, No), then the candidate list is updated in operation S920.

[0130] In operation S930, a new gain calculation can be performed based on the updated candidate list.

[0131] The new gain calculation can include calculating the range of howl suppression gain. For example, refer to Figure 10 This illustrates multi-band suppression divided into three frequency bands: high, medium, and low. The gain width of each band is set separately. As a non-limiting example only, the gain suppression width of the low-frequency sub-band (0-2kHz) can be 100Hz, the medium-frequency sub-band (2-4kHz) can be 1000Hz, and the high-frequency sub-band (>4kHz) can be 2000Hz. It is a frequency range calculated based on different gain widths, and can be expressed by equations. Let W represent the gain width.

[0132] According to at least one example embodiment, a first gain width is applied to the second candidate howling frequency point within a first frequency range, and a second gain width is applied to the second candidate howling frequency point within a second frequency range. The frequencies in the first frequency range are lower than the frequencies in the second frequency range, and the first gain width is smaller than the second gain width.

[0133] In operation of S940, candidate howling gains can be calculated based on the results of the new gain calculation.

[0134] Reference Figure 10 The candidate howling gain can include gain update and gain smoothing. Gain update consists of three parts: gain range update, horizontal (time domain) smoothing, and vertical (frequency domain) smoothing. In the gain range update, based on the results of the previous stage... The algorithm updates the range of frequency points that need to be processed. In horizontal smoothing, the algorithm utilizes the release and attack times to smooth the gain. At that time, through the equation To set and update the initial suppression gain C, where, ,and In vertical smoothing, the gain can be smoothed using the sigmoid function. The sigmoid function can be expressed as: ,in In some embodiments, gain smoothing can be achieved through parameter T. T and T A To control.

[0135] When operating the S950, howling suppression can be performed based on candidate howling gain.

[0136] For example, through equations To perform howling gain suppression processing.

[0137] Although reference Figure 9 A flowchart of howling suppression operation is described, but this disclosure is not limited thereto. Howling suppression operation can be performed in any and various ways. Various methods may include, but are not limited to, phase modulation, gain control, and adaptive feedback cancellation.

[0138] Figure 11 A schematic diagram of a howling suppression method according to an example embodiment of the present disclosure is shown.

[0139] Reference Figure 11 The howling suppression method according to the example embodiments of this disclosure may include one or more operations described above.

[0140] The howling suppression method according to an example embodiment of this disclosure may include dual-filter bank processing performed by a second-order adaptive filter bank. The dual-filter bank processing performed by the second-order adaptive filter bank includes first dual-filter bank processing of a first dual-filter bank and second dual-filter bank processing of a second dual-filter bank. The description above regarding the first and second dual-filter bank processing also applies here to the first dual-filter bank processing of the first dual-filter bank and the second dual-filter bank processing of the second dual-filter bank. Therefore, for the sake of brevity, repeated descriptions will be omitted.

[0141] The howling suppression method according to an example embodiment of this disclosure may include a frequency domain transformation operation (e.g., an STFT-based frequency domain transformation operation). The frequency domain transformation operation described above can also be applied to the frequency domain transformation operation herein. Therefore, for the sake of brevity, repeated descriptions will be omitted.

[0142] The howling suppression method according to an example embodiment of this disclosure may include two-level howling detection. Two-level howling detection may include coarse howling detection and fine howling detection. The description above regarding coarse howling detection and fine howling detection also applies to coarse howling detection and fine howling detection herein. Therefore, for the sake of brevity, repeated descriptions will be omitted. Peak picking and discriminant feature calculation may be computational operations in coarse howling detection and fine howling detection.

[0143] The howling suppression method according to an example embodiment of this disclosure may include candidate howling component selection and candidate howling gain. Candidate howling component selection may include... Figure 9 The operation regarding gain shown in the figure can include candidate howling gains. Figure 10 The candidate howling gain. Therefore, for the sake of brevity, repeated descriptions will be omitted.

[0144] The howling suppression method according to an example embodiment of this disclosure may include an inverse transform corresponding to the STFT (e.g., ISTFT synthesis). The inverse transform corresponding to the STFT can be used to generate a speaker source signal. The speaker source signal can then be transmitted to the first speaker after passing through a gain G1 of the first speaker, and to the second speaker after passing through a gain G2 of the second speaker.

[0145] Thus, the signal forms a closed loop, effectively eliminating feedback and ensuring the quality of the audio signal within this loop.

[0146] Figure 12 A block diagram of a howling suppression device according to an example embodiment of the present disclosure is shown.

[0147] Reference Figure 12 The howling suppression device 1200 may include at least one memory 1210 and at least one processor 1220. The at least one processor 1220 may include components configured according to the above-mentioned... Figures 1 to 11 The described howling suppression method encodes a central processing unit (CPU), audio processor, audio signal processor, application-specific integrated circuit (ASIC), and / or hardware control logic.

[0148] At least one memory 1210 may store one or more instructions. When executed by at least one processor 1220, the one or more instructions are configured to cause the at least one processor 1220 to: perform a frequency domain transformation operation on a microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal; divide the frequency domain microphone signal into multiple sub-bands; detect a first candidate howling frequency point from the multiple sub-bands by performing coarse howling detection on the multiple sub-bands; perform fine howling detection on a candidate howling band including the first candidate howling sub-band to detect a second candidate howling frequency point from the candidate howling band, wherein the first candidate howling frequency point is included in the first candidate howling sub-band among the multiple sub-bands; and perform howling suppression operation on the second candidate howling frequency point.

[0149] Reference Figures 1 to 11 One or more of the described whistling suppression methods can be performed by the whistling suppression device 1200. More specifically, refer to Figures 1 to 11The described one or more howling suppression methods can be executed by at least one processor 1220 in conjunction with one or more instructions stored in one or more memories 1210. Therefore, for the sake of brevity, repeated descriptions will be omitted.

[0150] Furthermore, the feedback suppression device 1200 may be included in an electronic device with a microphone. The electronic device may also include one or more speakers.

[0151] The howling suppression method according to at least one example embodiment may employ a stereo adaptive digital filter as a pre-filter to eliminate linear feedback from multiple speakers to the microphone. The howling suppression method according to at least one example embodiment may use sub-band-based howling pre-detection on multiple frequency bands, and then perform precise howling peak detection on the identified local sub-bands. The howling suppression method according to at least one example embodiment may apply different gain suppression scales to different frequencies to minimize damage to the desired signal, thereby achieving effective howling suppression.

[0152] Furthermore, the feedback suppression method according to at least one example embodiment can support multi-channel speaker systems. The feedback suppression method according to at least one example embodiment can quickly detect feedback frequency peaks using a two-stage feedback detection block. The feedback suppression method according to at least one example embodiment can provide higher audio quality.

[0153] The embodiments of this disclosure provide a non-transitory computer-readable storage medium storing a computer program or instructions that, when executed by at least one processor, can perform or implement the steps and corresponding content of the foregoing method embodiments.

[0154] Embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the foregoing method embodiments.

[0155] The terms "first," "second," "third," "fourth," "1," "2," etc., used in this disclosure, claims, and accompanying drawings are for distinguishing similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in a sequence other than that shown in the figures or text.

[0156] It should be understood that although arrows indicate various operation steps in the flowcharts of the embodiments of this disclosure, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this disclosure, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured as required, and the embodiments of this disclosure do not limit this.

[0157] The above text and accompanying drawings are provided as examples only to help the reader understand this disclosure. They are not intended and should not be construed as limiting the scope of this disclosure in any way. Although certain embodiments and examples have been provided, it will be apparent to those skilled in the art, based on the content disclosed herein, that changes can be made to the illustrated embodiments and examples, and other similar implementations based on the technical concept of this disclosure can be adopted without departing from the scope of this disclosure, and these modifications and modifications are also within the protection scope of the embodiments of this disclosure.

Claims

1. A method for suppressing howling, the method comprising: Perform a frequency domain transformation operation on the microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal; Divide the frequency domain microphone signal into multiple sub-bands; The first candidate howling frequency point where howling occurs is detected by performing coarse howling detection on the multiple subbands; By performing fine-grained howling detection on the candidate howling band, including the first candidate howling frequency point, a second candidate howling frequency point causing howling is detected from the candidate howling band; and Perform a howling suppression operation on the second candidate howling frequency.

2. The howling suppression method according to claim 1, wherein, Coarse howl detection includes howl detection based on a first howl feature, and fine howl detection includes howl detection based on a second howl feature. The computational complexity of the first howling feature is lower than that of the second howling feature.

3. The howling suppression method according to claim 2, wherein, The first howling feature includes peak-to-average power ratio and inter-frame peak duration, while the second howling feature includes peak-to-neighbor power ratio, peak-to-threshold power ratio, and inter-frame peak duration.

4. The howling suppression method according to claim 1, wherein, The whistling suppression method also includes: Acquire the first speaker source signal and the second speaker source signal; Before performing the frequency domain transformation operation: A first microphone signal is generated by removing the linear echo corresponding to the first speaker source signal from the microphone signal through first dual filter bank processing based on the first dual filter bank and the first speaker source signal; and The second microphone signal is generated by removing the linear echo corresponding to the second speaker source signal from the first microphone signal through second dual-filter bank processing based on the second dual-filter bank and the second speaker source signal. The step of performing a frequency domain transformation operation on the microphone signal includes: performing a frequency domain transformation operation on the second microphone signal to generate a frequency domain microphone signal.

5. The howling suppression method according to claim 4, wherein, One of the first speaker source signal and the second speaker source signal corresponds to the target speaker source signal. The dual filter bank in the first and second dual filter banks that corresponds to the target loudspeaker source signal corresponds to the target dual filter bank, and The dual-filter bank processing corresponding to the target loudspeaker source signal in the first dual-filter bank processing and the second dual-filter bank processing corresponds to the target dual-filter bank processing. The target dual filter bank includes a first adaptive filter and a second adaptive filter. The target dual filter bank processing includes first-stage adaptive filter processing based on a first adaptive filter and second-stage adaptive filter processing based on a second adaptive filter. In the first-stage adaptive filter processing, the filter coefficients of the first adaptive filter are dynamically updated using the filter coefficients of the second adaptive filter. Furthermore, the linear echo corresponding to the target loudspeaker source signal includes the signal output from the first adaptive filter that receives the target loudspeaker source signal. In the second-stage adaptive filter processing, the filter coefficients of the second adaptive filter are updated based on the error between the estimated feedback signal and the actual feedback signal of the second adaptive filter. The estimated feedback signal of the second adaptive filter includes the signal output from the second adaptive filter that receives the target speaker source signal, and the actual feedback signal of the second adaptive filter includes the microphone signal.

6. The howling suppression method according to claim 5, wherein, When the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter decreases to the threshold error level, the filter coefficients of the second adaptive filter are passed to the filter coefficients of the first adaptive filter.

7. The howling suppression method according to claim 5, wherein, When the error between the estimated feedback signal of the second adaptive filter and the actual feedback signal of the second adaptive filter is greater than a threshold, the filter coefficients of the first adaptive filter are passed to the filter coefficients of the second adaptive filter.

8. The howling suppression method according to claim 1, wherein, Performing howling suppression operations on the second candidate howling frequency point includes: Based on the second candidate howling frequency point within the first frequency range, the first gain width is applied to the second candidate howling frequency point; and Based on the second candidate howling frequency point within the second frequency range, the second gain width is applied to the second candidate howling frequency point. Wherein, the frequencies in the first frequency range are lower than the frequencies in the second frequency range, and The first gain width is smaller than the second gain width.

9. The howling suppression method according to claim 1, wherein, The candidate howling band includes a first candidate howling subband and a subband adjacent to the first candidate howling subband. The first candidate howling subband includes the first candidate howling frequency point.

10. A howling suppression device, the howling suppression device comprising: At least one processor, and At least one memory, storing one or more instructions. Wherein, the one or more instructions, when executed by the at least one processor, are configured to cause the at least one processor to: Perform a frequency domain transformation operation on the microphone signal to generate a frequency domain microphone signal corresponding to the microphone signal; Divide the frequency domain microphone signal into multiple sub-bands; The first candidate howling frequency point where howling occurs is detected by performing coarse howling detection on the multiple subbands; By performing fine-grained howling detection on the candidate howling band, including the first candidate howling frequency point, a second candidate howling frequency point causing howling is detected from the candidate howling band; and Perform a howling suppression operation on the second candidate howling frequency.

Images