Howling suppression device and howling suppression method

Abstract

The present invention relates to a howling suppression device and a howling suppression method, in which a voice signal input by a microphone and an amplified signal fed back by a loudspeaker are mixed and then divided into two signals, one of which undergoes a howling analysis to obtain the probability of howling occurring at each frequency point, and the other undergoes a delay process before both signals are sent to a dynamic filter group for filtering. The obtained howling probability is introduced into the generation process of the filter coefficients in the dynamic filter group, and the size of the suppression gain in the filter coefficients is adjusted according to the size of the howling probability. In addition, the present application separates the detection node and the processing node, which greatly reduces the computing power bottleneck brought by Fourier and inverse Fourier transforms compared to the existing technology, reduces the computing power requirements while improving the howling suppression effect, and avoids sound quality damage and pop sound problems caused by sudden changes in howling suppression parameters through the specific switching logic of the two filter groups provided in the dynamic filter group.

Description

Technical Field

[0001] The present invention relates to the technical field of sound signal processing, and in particular to a howling suppression device and a howling suppression method. Background Art

[0002] In scenarios where a microphone (for voice collection) and an external speaker operate simultaneously, such as in KTVs, in-car karaoke systems, or at home, the microphone picks up sound, amplifies the signal, and then plays it back out through the speaker. However, the sound emitted by the speaker is captured again by the microphone, forming a closed audio loop. When the amplitude-frequency and phase responses of this loop meet certain conditions, the feedback signal will be continuously self-amplified, causing severe distortion and clipping of the signal played back by the speaker, resulting in howling. Due to environmental limitations (such as in specific indoor and car spaces), the amplitude-frequency and phase-frequency responses of the feedback path from the speaker sound signal to the microphone cannot be flexibly adjusted. Therefore, specific signal processing is required between the microphone acquisition and the speaker output to prevent self-amplification of the signal throughout the loop and thus avoid howling.

[0003] Patent application number 202280045253.3, titled "Howling Suppression Device, Howling Suppression Method, and Howling Suppression Program," describes a technical solution for obtaining howling suppression parameters (which require Fourier transformation to the frequency domain) through a frequency-domain adaptive filter. Furthermore, these howling suppression parameters must be effective in the frequency domain, meaning the input signal must undergo an additional Fourier transform to the frequency domain. After frequency-domain processing, they must be inversely Fourier transformed back to the time domain before being output. Furthermore, to avoid sound quality degradation and popping caused by real-time updates of howling suppression parameters, all howling suppression parameters must be smoothed, which is cumbersome and computationally intensive. The howling suppression algorithm and solution must be integrated into a specific DSP or SOC chip. If the computing power requirements are high, this may exceed the computing power limit of the DSP or SOC, making it impossible to implement in commercial products. Furthermore, due to the smoothing of the howling suppression parameters, there is a lag in the effective suppression of howling, which may result in the initial howling sound not being suppressed. Summary of the Invention

[0004] A primary object of the present invention is to overcome at least one of the above-mentioned drawbacks, and to provide a howling suppression device having a simple overall structure and requiring less computing power, while being able to effectively improve the howling suppression effect.

[0005] Another main object of the present invention is to overcome at least one of the above-mentioned drawbacks, and to provide a howling suppression method that can adjust the gain according to the probability of howling, thereby improving the howling suppression effect while reducing the computing power requirement, and can achieve early suppression of howling.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] The present invention provides a howling suppression device, comprising:

[0008] The signal acquisition module collects the voice signal input by the microphone and the amplified signal fed back by the speaker and forms a digital input signal;

[0009] a howling suppression detection module, connected to one output of the signal acquisition module, configured to detect the frequency characteristics of the digital input signal and analyze the howling probability corresponding to the occurrence of howling at each frequency point, and form a filter coefficient for adjusting the gain of the filter based on the detection and analysis results, wherein the filter coefficient includes the filter frequency, the filter bandwidth, and the corresponding suppression gain;

[0010] a buffer connected to another output of the signal acquisition module, for performing delay processing on the digital input signal and outputting the delayed digital signal to the dynamic filter bank;

[0011] and the dynamic filter group, adjusting the filtering frequency, filtering bandwidth and suppression gain of each filter in the dynamic filter group according to the filtering coefficient, and performing filtering processing on each frequency point in the delayed digital signal under the corresponding filtering suppression gain through the adjusted dynamic filter group, so that howling suppression can be completed after filtering processing by the dynamic filter group.

[0012] According to one embodiment of the present invention, the output end of the signal acquisition module is connected to a first limiter, which is used to limit the maximum amplitude of the digital input signal. The output limited digital signal is divided into two paths and sent to the howling suppression detection module and the buffer respectively.

[0013] According to one embodiment of the present invention, an internal gain module is provided between the signal acquisition module and the signal acquisition module, for performing overall gain adjustment on the digital input signal.

[0014] According to one embodiment of the present invention, the rear side of the dynamic filter bank is connected to a second limiter, and the second limiter is used to limit the maximum amplitude of the output signal of the dynamic filter bank.

[0015] According to one embodiment of the present invention, the dynamic filter group includes two or more filter groups, the howling suppression detection module generates corresponding filter coefficients in sequence according to digital input signals of different frames, and each filter group alternately receives the filter coefficients generated in sequence and adjusts the filter frequency, filter bandwidth and suppression gain of each filter in each filter group.

[0016] In particular, the present application also provides a howling suppression method, which includes the following working steps:

[0017] Collect the voice signal input from the microphone and the amplified signal fed back from the speaker, and convert them into digital input signals;

[0018] detecting a frequency characteristic of the digital input signal, analyzing a probability corresponding to howling occurring at each frequency point in the digital input signal, and correspondingly binding a frequency value, a gain, and a probability of howling occurring at each frequency point of the digital input signal as a howling suppression parameter; and forming a filter coefficient for adjusting the gain of a filter according to the howling suppression parameter, wherein the filter coefficient includes a filter frequency, a filter bandwidth, and a corresponding suppression gain;

[0019] Performing delay processing on the digital input signal to form a delayed digital signal;

[0020] The filter frequency, filter bandwidth and suppression gain of the dynamic filter group are adjusted according to the filter coefficient, and the delayed digital signal is filtered by the dynamic filter group to achieve howling suppression.

[0021] According to one embodiment of the present invention, in the process of analyzing the digital input signal to obtain the howling suppression parameter, and forming the filter coefficient for gain adjustment of the filter according to the howling suppression parameter,

[0022] The frequency feature extraction algorithm is used to extract and analyze the frequency features of each frequency point, obtain the howling probability P(k) of howling at each frequency point, and form a waveform of the howling probability P(k);

[0023] Analyze each continuous waveform in the waveform of the howling probability P(k), determine the maximum probability value and the corresponding target frequency point in each continuous waveform, and record them as filter coefficients.

[0024] According to one embodiment of the present invention, after determining the maximum probability value and the corresponding target frequency point in each continuous waveform, two limit frequency values ​​±ndB on the left and right sides of the target frequency point are determined according to the preset attenuation value ndB, and the bandwidth value between the two limit frequency values ​​is used as the filter bandwidth of the filter coefficient, where n is a preset constant.

[0025] According to one embodiment of the present invention, when analyzing each continuous waveform in the waveform of the howling probability P(k), if the maximum probability value in the continuous waveform, i.e., the peak, is a continuous value, the frequency point corresponding to the center value of the continuous peak is taken as the target frequency point.

[0026] According to one embodiment of the present invention, the maximum suppression gain of the filter in the dynamic filter bank is preset to Gmax, and the suppression gain in the filter coefficient is the product of the howling probability P(k) at the target frequency point and the maximum suppression gain Gmax.

[0027] According to one embodiment of the present invention, the dynamic filter group includes two or more filter groups, each filter group includes several filters, each filter adjusts its own filtering frequency, filtering bandwidth and suppression gain according to the filtering coefficient, and each filter is dynamically adjusted corresponding to a filtering frequency, filtering bandwidth and corresponding suppression gain. The delayed digital signal is processed by each filter group and then summed to obtain the optimized processed signal.

[0028] According to one embodiment of the present invention, the dynamic filter bank includes two filter banks, denoted as a first filter bank and a second filter bank, wherein the first filter bank filters the delayed digital signal and then multiplies it by a proportional coefficient ratio, and the second filter bank filters the delayed digital signal and then multiplies it by 1-ratio, wherein the proportional coefficient ratio switches between 0 and 1. Preferably, the switching period of the proportional coefficient ratio is equal to the output period of the filter coefficient.

[0029] According to one embodiment of the present invention, the output signal after the delayed digital signal is filtered by the first filter group is recorded as vout1, the output signal after the delayed digital signal is filtered by the second filter group is recorded as vout2, and the output signal after the delayed digital signal is filtered by the dynamic filter group is recorded as vout, vout=ratio*vout1+(1-ratio)*vout2,

[0030] When the proportional coefficient ratio=0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the first filter group and the filter frequency, filter bandwidth and suppression gain of the first filter group are adjusted according to the filter coefficient. After the first filter group is updated, the proportional coefficient ratio gradually changes from 0 to 1. When the proportional coefficient ratio becomes 1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal.

[0031] When the proportional coefficient ratio=1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the second filter group and the filtering frequency, filtering bandwidth and suppression gain of the second filter group are adjusted according to the filter coefficient. After the update of the second filter group is completed, the proportional coefficient ratio gradually changes from 1 to 0. When the proportional coefficient ratio becomes 0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal.

[0032] According to one embodiment of the present invention, the method generates corresponding filter coefficients in sequence according to digital input signals of different frames, and sends them to the dynamic filter group in sequence. Each filter group in the dynamic filter group alternately receives each of the filter coefficients. Each filter group adjusts the filter frequency, filter bandwidth and suppression gain of each filter according to the received filter coefficient. The adjusted filter group receives the digital input signal corresponding to the filter coefficient and performs filtering processing.

[0033] Compared with the prior art, the advantages and beneficial effects of the howling suppression device and howling suppression method of the patent application of this invention are:

[0034] The present application mixes the voice signal input by the microphone and the amplified signal fed back by the loudspeaker and divides the mixed signal into two signals. One signal is subjected to a howling analysis to obtain the probability of howling occurring at each frequency point, and the other signal is subjected to a delay process before being sent together to a dynamic filter group for filtering. The obtained howling probability is introduced into the generation process of the filter coefficients in the dynamic filter group, and the size of the suppression gain in the filter coefficients is adjusted according to the size of the howling probability. Unlike the traditional howling judgment with an identification probability of only 0 or 1 and a fixed suppression gain, the present application can reduce the error processing rate of howling suppression and improve the output sound quality of the loudspeaker.

[0035] In addition, the dynamic filter bank introduced in this application detects and analyzes howling suppression parameters in the frequency domain (detection node) and performs noise suppression on acoustic signals in the time domain (processing node). This separation of detection and processing nodes significantly reduces the computing power bottlenecks associated with Fourier and inverse Fourier transforms compared to existing technologies, lowering computing power requirements while improving howling suppression effectiveness. Furthermore, the specific switching logic between the two filter banks within the dynamic filter bank avoids sound quality impairments and popping issues caused by sudden changes in howling suppression parameters.

[0036] In addition, in this application, the digital input signal is delayed through a buffer. In this application, the detection node and the processing node are separated. After the digital input signal is delayed, the output sound signal can be suppressed in advance, avoiding the occurrence of the first howling sound caused by the slow change of the filter coefficient, thereby improving the sound quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings:

[0038] Figure 1 This is an overall principle block diagram of a howling suppression device according to an embodiment of the present application;

[0039] Figure 2 is a principle block diagram of a howling suppression detection module according to one embodiment of the present application;

[0040] Figure 3 is a principle block diagram of a dynamic filter bank according to one embodiment of the present application;

[0041] Figure 4 1 is a schematic diagram of the frequency domain energy distribution of a digital input signal after Fourier transformation during the filter coefficient calculation process according to one embodiment of the present application;

[0042] Figure 51 is a schematic diagram of a waveform curve of howling probability during the calculation of filter coefficients according to an embodiment of the present application;

[0043] Figure 6 3 is a schematic diagram of the comparison results of input and output signals after being processed by the howling suppression method according to an embodiment of the present application.

[0044] The following are the descriptions of the reference numerals:

[0045] 1. Microphone, 2. Speaker;

[0046] 3. Signal acquisition module, 4. Internal gain module, 5. First limiter, 6. Buffer, 9. Second limiter;

[0047] 7. Howling suppression detection module, 71. Fourier transform unit, 72. Howling probability calculation unit, 73. Filter coefficient calculation unit;

[0048] 8. Dynamic filter bank, 81. Filter, 82. Multiplier, 83. Adder. DETAILED DESCRIPTION

[0049] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0051] Example 1:

[0052] This embodiment describes a howling suppression device, such as Figure 1 As shown, it includes:

[0053] Signal acquisition module 3, collects the voice signal input by microphone 1 and the amplified signal fed back by speaker 2 and forms a digital input signal;

[0054] A howling suppression detection module 7 is connected to one output of the signal acquisition module 3 and is used to detect the frequency characteristics of the digital input signal and analyze the howling probability corresponding to the occurrence of howling at each frequency point, and form a filter coefficient for gain adjustment of the filter 81 based on the detection and analysis results, wherein the filter coefficient includes the filter frequency, the filter bandwidth, and the corresponding suppression gain;

[0055] a buffer 6 connected to another output of the signal acquisition module 3, for performing delay processing on the digital input signal and outputting the delayed digital signal to the dynamic filter bank 8;

[0056] And the dynamic filter group 8 adjusts the filtering frequency, filtering bandwidth and suppression gain of each filter 81 in the dynamic filter group 8 according to the filtering coefficient, and the adjusted dynamic filter group 8 performs filtering processing on each frequency point in the delayed digital signal under the corresponding filtering suppression gain. After filtering processing by the dynamic filter group 8, howling suppression can be completed.

[0057] The voice signal input by microphone 1 and the amplified signal fed back by loudspeaker 2 are mixed and divided into two signals. One signal is subjected to howling analysis to obtain the probability of howling occurring at each frequency point, and the other signal is subjected to delay processing and then sent to the dynamic filter group 8 for filtering. The obtained howling probability is introduced into the generation process of the filter coefficient in the dynamic filter group 8, and the size of the suppression gain in the filter coefficient is adjusted according to the size of the howling probability. Different from the traditional howling judgment with an identification probability of only 0 or 1 and a fixed suppression gain, the present application can reduce the error processing rate of howling suppression and improve the output sound quality of the loudspeaker 2.

[0058] The output end of the signal acquisition module 3 is connected to a first limiter 5, which is used to limit the maximum amplitude of the digital input signal to prevent overflow or clipping of the digital input signal itself. The limited digital signal output by the first limiter 5 is divided into two paths and sent to the howling suppression detection module 7 and the buffer 6 respectively. In addition, an internal gain module 4 is provided between the signal acquisition module 3 and the signal acquisition module 3 for adjusting the overall gain of the digital input signal.

[0059] In one embodiment, a second limiter 9 may also be connected to the rear side of the dynamic filter group 8. The second limiter 9 is used to limit the maximum amplitude of the output signal of the dynamic filter group 8 to prevent overflow or clipping of the output signal caused by processing by the dynamic filter group 8.

[0060] Specifically, if Figure 2 As shown, the howling suppression detection module 7 includes a Fourier transform unit 71, a howling probability calculation unit 72 and a filter coefficient calculation unit 73. The Fourier transform unit 71 processes the digital input signal to obtain the distribution of frequency energy parameters. The obtained frequency domain energy parameters are sent to the howling probability calculation unit 72 to realize frequency domain feature extraction and calculate the howling probability of each frequency point. According to the obtained howling probability of each frequency point, the filter coefficient calculation unit 73 generates a probability waveform and calculates the filter frequency and the corresponding suppression gain.

[0061] The howling probability calculation unit 72 may integrate one or more functional modules such as Ptpr, Papr, Pnpr, and Cnfr to comprehensively calculate the howling probability.

[0062] like Figure 3 As shown, the dynamic filter bank 8 may include two filter banks, each of which includes a plurality of filters 81. Each filter 81 adjusts its own filter frequency, filter bandwidth, and suppression gain according to the filter coefficients. The delayed digital signal is processed by each filter bank and then summed to obtain an optimized signal. Combined with the specific switching logic of the two filter banks in the dynamic filter bank, sound quality degradation and popping problems caused by sudden changes in howling suppression parameters are avoided.

[0063] like Figure 3 As shown, each filter 81 in the two filter groups is configured with a different filtering frequency range and suppression gain according to the filtering coefficient. The delayed digital signal is filtered by each filter 81 in each filter group in turn. A multiplier 82 is provided at the back end of each filter group. The signal vout1 after filtering by one filter group is multiplied by the proportional coefficient ratio by a multiplier 82, and the signal vout2 after filtering by the other filter group is multiplied by 1-ratio by another multiplier 82, wherein the proportional coefficient ratio switches between 0 and 1, and the switching period of the proportional coefficient ratio is equal to the output period of the filtering coefficient. The adder 83 connected to the two multipliers 82 sums the processed signals to obtain the final signal vout, that is, vout=(1-ratio)*vout1+ratio*vout2, and the proportional coefficient ratio switches between 0 and 1. It is only necessary to control the switching period of the proportional coefficient ratio to be consistent with the output period of the filter coefficient. At the same time, the delayed digital input signal is sent to the adjusted filter group, so that the digital input signal can be accurately and effectively filtered to complete howling suppression.

[0064] Specifically, the output signal after the delayed digital signal is filtered by the first filter group is recorded as vout1, the output signal after the delayed digital signal is filtered by the second filter group is recorded as vout2, and the output signal after the delayed digital signal is filtered by the dynamic filter group is recorded as vout, vout=ratio*vout1+(1-ratio)*vout2,

[0065] When the proportional coefficient ratio=0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the first filter group and the filter frequency, filter bandwidth and suppression gain of the first filter group are adjusted according to the filter coefficient. After the first filter group is updated, the proportional coefficient ratio gradually changes from 0 to 1. When the proportional coefficient ratio becomes 1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal.

[0066] When the proportional coefficient ratio=1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the second filter group and the filtering frequency, filtering bandwidth and suppression gain of the second filter group are adjusted according to the filter coefficient. After the update of the second filter group is completed, the proportional coefficient ratio gradually changes from 1 to 0. When the proportional coefficient ratio becomes 0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal.

[0067] The dynamic filter group 8 may also include more than two filter groups, as long as the filter group performing real-time filtering and the filter group performing filter coefficient update are separated, and both should be protected by the present application.

[0068] The howling suppression detection module 7 sequentially generates corresponding filter coefficients based on the digital input signals of different frames, and sequentially sends them to the dynamic filter group. Each filter group in the dynamic filter group alternately receives each of the filter coefficients. Each filter group adjusts the filter frequency, filter bandwidth and suppression gain of each filter according to the received filter coefficient. The adjusted filter group receives the digital input signal corresponding to the filter coefficient and performs filtering processing.

[0069] Example 2:

[0070] The present application also provides a howling suppression method, the framework of which is based on the howling suppression device described in Example 1. The howling suppression method includes the following steps:

[0071] Collect the voice signal input by microphone 1 and the amplified signal fed back by speaker 2, and convert them into digital input signals;

[0072] Detecting the frequency characteristics of the digital input signal, analyzing the probability of howling occurring at each frequency point in the digital input signal, and correspondingly binding the frequency value, gain, and probability of howling occurring at each frequency point of the digital input signal as a howling suppression parameter, and forming a filter coefficient for gain adjustment of the filter 81 according to the howling suppression parameter, wherein the filter coefficient includes a filter frequency, a filter bandwidth, and a corresponding suppression gain;

[0073] Performing delay processing on the digital input signal to form a delayed digital signal;

[0074] The filter frequency, filter bandwidth and suppression gain of the dynamic filter group 8 are adjusted according to the filter coefficient, and the delayed digital signal is filtered by the dynamic filter group 8 to achieve howling suppression.

[0075] The voice signal input by microphone 1 and the amplified signal fed back by loudspeaker 2 are mixed and divided into two signals. One signal is subjected to howling analysis to obtain the probability of howling occurring at each frequency point, and the other signal is subjected to delay processing and then sent to the dynamic filter group 8 for filtering. The obtained howling probability is introduced into the generation process of the filter coefficient in the dynamic filter group 8, and the size of the suppression gain in the filter coefficient is adjusted according to the size of the howling probability. Different from the traditional howling judgment with an identification probability of only 0 or 1 and a fixed suppression gain, the present application can reduce the error processing rate of howling suppression and improve the output sound quality of the loudspeaker 2.

[0076] In the process of analyzing the digital input signal to obtain howling suppression parameters and forming filter coefficients for gain adjustment of the filter 81 based on the howling suppression parameters, the frequency characteristics of each frequency point are extracted and analyzed respectively by a frequency feature extraction algorithm to obtain the howling probability P(k) of howling at each frequency point, and form a waveform of the howling probability P(k); each continuous waveform in the waveform of the howling probability P(k) is analyzed to determine the maximum probability in each continuous waveform and the corresponding target frequency point and record them as the filter coefficient.

[0077] After determining the maximum probability value and the corresponding target frequency point in each continuous waveform, two extreme frequency values ​​±ndB to the left and right of the target frequency point are determined based on the preset attenuation value ndB. The frequency point between the two extreme frequency values ​​is used as the filtering frequency of the filter coefficient, where n is a preset constant. In addition, when analyzing each continuous waveform in the waveform of the howling probability P(k), if the maximum probability value, i.e., the peak, in the continuous waveform is a continuous value, the frequency point corresponding to the center value of the continuous peak is taken as the target frequency point. The maximum suppression gain of the filter 81 in the dynamic filter bank 8 is preset to Gmax, and the suppression gain in the filter coefficient is the product of the howling probability P(k) at the target frequency point and the maximum suppression gain Gmax.

[0078] The digital input signal is a time domain digital signal that can be played normally. It is converted into a frequency domain signal through Fourier transform and then analyzed to obtain the howling suppression parameters. The horizontal axis of the frequency domain signal is the frequency, and the vertical axis is the energy amplitude corresponding to each frequency point, which reflects the energy distribution of different frequencies after the digital input signal (time domain signal) is converted into the frequency domain. Figure 4 shown.

[0079] Figure 4 This diagram shows the frequency domain energy distribution of the digital input signal after Fourier transform. The energy at 600 Hz is higher, indicating that the primary frequency components of this digital input signal are concentrated around 600 Hz. This frequency domain energy distribution diagram allows for frequency feature extraction. Using a frequency feature extraction algorithm, the frequency features of each frequency point are extracted and analyzed to determine the howling probability P(k) at each frequency point. k represents the frequency index, and different k values ​​represent different frequencies.

[0080] Common frequency feature extraction algorithms include peak power, peak-to-average power ratio, and peak-to-nearest neighbor power ratio. You can choose one of these to calculate the howling frequency, or you can combine two or more frequency feature extraction algorithms to calculate the howling frequency and then perform mean calculation, standard deviation calculation, or product calculation. This allows you to more accurately calculate the howling frequency by combining the results of each frequency feature extraction algorithm. This allows you to more accurately determine the howling probability.

[0081] This embodiment uses Ptpr, Papr, Pnpr, Cnfr and other modules to extract frequency features, and each module outputs the howling probability p(k, i) of each frequency point, where i represents the probability output by different feature extraction modules.

[0082] The Ptpr module detects the peak power at each frequency and employs two thresholds, TH1 and TH2, where TH1 is the upper threshold and TH2 is the lower threshold. When the peak power at the corresponding frequency is less than or equal to TH2, p(k, 1) for that frequency is 0; when the peak power at the corresponding frequency is greater than or equal to TH1, p(k, 1) for that frequency is 1. When the peak power pp(k) at the corresponding frequency is between TH1 and TH2, a function mapping is used to obtain the probability p(k, 1). For example, a linear mapping of p(k, 1) = (pp(k) - TH2) / (TH1 - TH2) is used. More complex mappings are also possible, but are not limited here and can be adjusted based on user needs and system design requirements.

[0083] Papr module calculates the peak energy pp(k) at each frequency point and the average power ppa of the entire frequency band (where ). Similarly, if ratio(k) is less than or equal to TH2 (this TH2 is different from TH2 in Ptpr, and the thresholds of each module are independently adjustable), p(k, 2) = 0; if ratio(k) is greater than or equal to TH1 (this TH1 is different from TH1 in Ptpr, and the thresholds of each module are independently adjustable), p(k, 2) = 1. When ratio(k) is between TH1 and TH2, the probability p(k, 2) can be obtained using a function mapping, such as the linear mapping p(k, 2) = (ratio(k) - TH2) / (TH1 - TH2). Other more complex mapping relationships can also be used, and are not specifically limited here.

[0084] The Pnpr module detects whether the energy of several adjacent frequency points is greater than a threshold. When the energy of the m adjacent frequency points, pp(km), pp(k-m+1), ..., pp(k), pp(k+1), ..., pp(k+m), is less than or equal to threshold TH2 (this threshold TH2 is different from threshold TH2 in Ptpr and can be set independently for each module), the corresponding pp3(k, j) = 0 (where j represents the index from km to k+m). When the energy of the m adjacent frequency points, pp(km), pp(k-m+1), ..., pp(k), pp(k+1), ..., pp(k+m), is greater than or equal to threshold TH1 (this threshold TH2 is different from threshold TH2 in Ptpr and can be set independently for each module), the corresponding pp3(k, j) = 1. When pp(km), pp(k-m+1), ..., pp(k), pp(k+1), ..., pp(k+m) are between TH1 and TH2, a function mapping can be used to obtain the probability pp3(k, j), such as the linear mapping pp3(k, j) = (pp(k+j) - TH2) / (TH1 - TH2). Other more complex mapping relationships can also be used, which are not specifically limited here. The final probability of the corresponding frequency point is .

[0085] The Cnfr module is a module that detects the energy characteristics of the same frequency points in adjacent frames. Considering the real-time nature of actual scenarios, such as karaoke scenes, each time the digital signal in the time domain is converted into a frequency signal, only a relatively short frame of data can be obtained. If the frame length is too long, the output delay of the filter coefficient output by the howling suppression detection module 7 will be very long, which cannot meet the real-time requirements of the karaoke scene. Therefore, the data length of a frame is usually set to between 5ms and 20ms. However, howling generally lasts for a relatively long time. Therefore, we use the information of consecutive adjacent frames to determine the probability of the howling frequency point, which will be more accurate and prevent false detection. The Cnfr module detects the frequency peaks pp(k, -n), pp(k, -n+1), ..., pp(k, 0), pp(k, 1), ..., pp(k, n) of the n adjacent frames before and after. When the detection result of pp4(k, j) (j represents the index of different frames) is less than or equal to TH2 (different from TH2 of other modules, can be set independently), the corresponding probability p4(k, j) = 0; when the detection result of pp4(k, j) (j represents the index of different frames) is greater than or equal to TH1 (different from TH1 of other modules, can be set independently), the corresponding probability p4(k, j) = 1. When the value of pp4(k, j) is between TH1 and TH2, a function mapping can be used to obtain the probability pp4(k, j), such as using a linear mapping pp4(k, j) = (pp(k, j) - TH2) / (TH1 - TH2). Other more complex mapping relationships can also be used, and there is no specific restriction here. The final probability of the corresponding frequency point is .

[0086] Of course, the probability of the howling frequency can also be determined by other features to obtain p(k, 5), p(k, 6), etc. This is not limited here. What is protected in this application is that the frequency domain features are combined with the frequency feature extraction algorithm to provide an estimated value of the howling probability corresponding to each frequency point, rather than a simple yes or no, and the howling probability value is subsequently applied to the generation of the filter coefficient for controlling and adjusting the dynamic filter 81.

[0087] The howling probability p(k) of each frequency point is obtained and input into the filter coefficient calculation unit 73. The filter coefficient calculation unit 73 is used to extract the waveform information of the howling probability p(k). Figure 5 As shown in the figure, the curve of the howling probability p(k) has two continuous peaks. For each continuous peak, only one filter 81 is used to process it (not every frequency point will use the filter 81). This can greatly reduce the number of filters 81 and avoid the overlapping of filters 81 due to being too close, resulting in a more severe overvoltage due to the superposition of filters 81. Therefore, for Figure 5 As shown in the p(k) curve, in this embodiment, two filters 81 are respectively provided in each filter bank of the dynamic filter bank 8 for filtering. Accordingly, only two sets of filter coefficients are required. The maximum suppression gain of each filter 81 is Gmax (open and adjustable). The filter 81 can be a notch filter 81 or a wave limiter.

[0088] The logic for generating the filter coefficient is as follows: find the maximum value and the corresponding frequency in the continuous wave peak. If there is a continuous identical maximum value, the frequency is the center frequency point of the continuous maximum value. The preset attenuation value ndB is 3dB. Then, based on the preset attenuation value, determine the two extreme frequency values ​​of ±3dB on the left and right sides of the target frequency point. The frequency point between the two extreme frequency values ​​is used as the filter frequency of the filter coefficient. The suppression gain in the filter coefficient is the product of the howling probability P (k) of the target frequency point and the maximum suppression gain Gmax.

[0089] by Figure 5Taking the probability waveform of the howling frequency point shown as an example, the maximum probability of the peak of the first continuous waveform is 1, and the value of the frequency center point fc1=600Hz. Find the frequencies fc11 and fc12 corresponding to the maximum value attenuated by 3dB relative to the maximum value. In the first continuous waveform, the maximum value attenuated by 3dB is 0.707, and the fc11 and fc12 corresponding to 0.707 are 440Hz and 730Hz respectively. Therefore, the -3dB bandwidth fb1=fc12-fc11=390Hz. Therefore, the filter frequency fc1=600 of the Peak filter 81 corresponding to the first continuous peak, the filter bandwidth fb1=390, and the suppression gain G1=-Gamx*max(p(k))=-Gmax*1=-Gmax.

[0090] Similarly, the parameters of the second group of filters 81 can be obtained as follows: filter frequency fc2=3100 Hz, filter bandwidth fb2=fc22-fc21=3230 Hz-2920 Hz=310 Hz, and suppression gain G1=-Gamx*max(p(k))=-Gmax*0.75=-0.75Gmax.

[0091] The filter coefficients obtained by the above calculation are sent to the dynamic filter bank 8, and each filter 81 in the dynamic filter bank 8 is configured. After the configuration is completed, the filter 81 filters the delayed digital input signal in the time domain to suppress the howling frequency point, thereby avoiding the occurrence of howling.

[0092] The filter coefficient changes with the detection of frequency domain characteristics of the digital input signal in different frames, and this change is discontinuous. If the filter 81 adjusted according to the filter coefficient is directly used for output time domain digital signal processing, noise and pop sounds will be introduced into the final output signal. Therefore, a special parameter smoothing scheme is required to cooperate with the dynamic filter group 8 for processing.

[0093] In this embodiment, the dynamic filter bank 8 includes two filter banks, each of which includes a plurality of filters 81. Each filter 81 adjusts its own filter frequency, filter bandwidth, and suppression gain according to the filter coefficient. Each filter 81 dynamically adjusts its filter frequency, filter bandwidth, and corresponding suppression gain. The delayed digital signal is processed by each filter bank and then summed to obtain an optimized signal. In the two filter banks of the dynamic filter bank 8, one filter bank filters the delayed digital signal and then multiplies it by a proportional coefficient ratio, and the other filter bank filters the delayed digital signal and then multiplies it by 1-ratio. The proportional coefficient ratio switches between 0 and 1, and the switching period of the proportional coefficient ratio is equal to the output period of the filter coefficient. The method sequentially generates corresponding filter coefficients based on digital input signals of different frames, and sequentially sends them to the dynamic filter group. Each filter group in the dynamic filter group alternately receives each of the filter coefficients. Each filter group adjusts the filter frequency, filter bandwidth and suppression gain of each filter according to the received filter coefficient. The adjusted filter group receives the digital input signal corresponding to the filter coefficient and performs filtering processing.

[0094] The voice input may be intermittent or continuous, but is generally input in a time-sequential order. The corresponding digital input signal is also acquired in a time-sequential manner. Accordingly, filter coefficients are generated one by one in the time sequence. Each filter coefficient is alternately sent to the two filter banks of the dynamic filter bank. Each filter 81 in each filter bank is configured with a different filter frequency range and suppression gain according to the filter coefficient. The delayed digital signal is filtered by each filter 81 in each filter bank in turn. A multiplier 82 is provided at the back end of each filter bank. After filtering by one filter bank, the signal vout1 is multiplied by a proportional coefficient ratio by a multiplier 82. After filtering by another filter bank, the signal vout2 is multiplied by 1-ratio by another multiplier 82. The proportional coefficient ratio switches between 0 and 1. Simply by controlling the switching period of the proportional coefficient ratio to coincide with the output period of the filter coefficient, and sending the delayed digital input signal to the adjusted filter bank, the digital input signal can be accurately and effectively filtered to achieve howling suppression.

[0095] The dynamic filter bank 8 comprises two filter banks, which, for ease of description, are referred to as the first filter bank and the second filter bank. Assuming that the coefficients of filter 81 in the first filter bank are detected based on the previous frame of digital input signal, when the coefficients of filter 81 in the next frame change, the coefficients of filter 81 corresponding to the next frame of digital input signal are updated in the second filter bank. The proportional coefficient, ratio, is a parameter that changes slowly and continuously from 0 to 1. Initially, ratio = 0, so the output signal vout is actually the output vout1 of the first filter bank. When ratio finally reaches 1, vout is actually the output vout2 of EQ group 2. Because ratio changes slowly and continuously, the final output signal vout does not introduce noise or popping due to sudden signal changes during the transition from vout1 to vout2. When the detected parameters in the next frame change and require updating, they are updated in EQ group 1. That is, when the coefficients of filter 81 detected by the howling suppression detection module 7 change, the parameters of the two filter banks are updated alternately. The outputs of both filter banks are summed using a proportional coefficient, ratio, which changes from 0 to 1, to obtain the final output signal.

[0096] Specifically, the output signal after the delayed digital signal is filtered by the first filter group is recorded as vout1, the output signal after the delayed digital signal is filtered by the second filter group is recorded as vout2, and the output signal after the delayed digital signal is filtered by the dynamic filter group is recorded as vout, vout=ratio*vout1+(1-ratio)*vout2,

[0097] When the proportional coefficient ratio=0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the first filter group and the filter frequency, filter bandwidth and suppression gain of the first filter group are adjusted according to the filter coefficient. After the first filter group is updated, the proportional coefficient ratio gradually changes from 0 to 1. When the proportional coefficient ratio becomes 1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal.

[0098] When the proportional coefficient ratio=1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the second filter group and the filtering frequency, filtering bandwidth and suppression gain of the second filter group are adjusted according to the filter coefficient. After the update of the second filter group is completed, the proportional coefficient ratio gradually changes from 1 to 0. When the proportional coefficient ratio becomes 0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal.

[0099] The dynamic filter group 8 may also include more than two filter groups, as long as the filter group performing real-time filtering and the filter group performing filter coefficient update are separated, and both should be protected by the present application.

[0100] Each update of the filter coefficient is performed before a single filter group is put into filtering processing, and the proportional coefficient ratio is slowly and continuously changed from 0 to 1 or from 1 to 0, so that pop sound caused by signal mutation will not be introduced. The howling suppression method of the present application effectively avoids the smooth update of the filter coefficients in the traditional technology through the setting of the dual filter group and the proportional coefficient ratio, because in the traditional technology, the calculation amount of converting the filter coefficients (filter frequency fc, filter bandwidth fb, suppression gain Gain) into the filter numerator and denominator coefficients is very large. If the coefficients of the filter 81 need to be smoothed, a large number of filter coefficients need to be converted into filter numerator and denominator coefficients, thereby greatly optimizing the calculation amount of the howling suppression algorithm.

[0101] In addition, because the dynamic filter group 8 switches from the previous filter 81 state to the next filter state with a smooth transition (the proportional coefficient ratio has a process of slowly changing from 0 to 1), there will be a lag in the actual effectiveness compared to the detection, resulting in the first sound of howling not being suppressed in time. Therefore, this embodiment can also add a buffer 6 on the front side of the dynamic filter group 8, and the buffer 6 can delay the digital input signal and output the delayed digital signal to the dynamic filter group 8, relatively advancing the detection time point, thereby effectively avoiding the problem of suppressing the first sound of howling. The length of the buffer 6 is open and configurable, and there is no restriction here. However, because there are requirements for the delay of the entire loop in working conditions such as karaoke scenes, the length of the buffer 6 should not be set too long.

[0102] In summary, the dynamic filter group 8 introduced in the present application realizes detection and analysis of howling suppression parameters in the frequency domain (detection node) and performs noise suppression on the sound signal in the time domain (processing node), which separates the detection node and the processing node. Compared with the existing technology, it greatly reduces the computing power bottleneck brought by Fourier and inverse Fourier transforms, reduces the computing power requirements, and improves the howling suppression effect. The specific switching logic of the two filter groups set in the dynamic filter group 8 avoids sound quality damage and pop sound problems caused by sudden changes in the howling suppression parameters.

[0103] like Figure 6 The figure shows the comparison results of the input and output signals after being processed by the howling suppression method of this application. The left channel is the input voice signal, and the right channel is the final output signal. Figure 6 It can be seen that the waveform of the output signal is almost the same as the input signal, and there is no howling caused by self-excitation. The reason why the output is slightly different from the input is that the microphone 1 collects the sound emitted by the speaker 2 and fed back through the room or the car space, but because of the algorithm processing, it has been eliminated to a small enough extent that the sound quality and effect of the output signal are not significantly changed. If the echo cancellation algorithm is added, the difference can be further optimized, but the purpose of this application is to suppress the generation of howling and does not involve echo cancellation. Therefore, the difference is within the acceptable range of this application.

[0104] The above embodiments are only for illustrating the technical concept and features of the present invention. Their purpose is to enable people familiar with this technology to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the scope of protection of the present invention.

Claims

1. A howling suppression device, characterized in that: include: The signal acquisition module collects the voice signal input by the microphone and the amplified signal fed back by the speaker and forms a digital input signal; a howling suppression detection module, connected to one output of the signal acquisition module, configured to detect the frequency characteristics of the digital input signal and analyze the howling probability corresponding to the occurrence of howling at each frequency point, and to bind the frequency value, gain, and howling probability of each frequency point of the digital input signal as howling suppression parameters, and to form a filter coefficient for adjusting the gain of the filter based on the howling suppression parameters, wherein the filter coefficient includes a filter frequency, a filter bandwidth, and a corresponding suppression gain; a buffer connected to another output of the signal acquisition module, for performing delay processing on the digital input signal and outputting the delayed digital signal to the dynamic filter bank; and the dynamic filter group, adjusting the filtering frequency, filtering bandwidth and suppression gain of each filter in the dynamic filter group according to the filtering coefficient, and performing filtering processing on each frequency point in the delayed digital signal under the corresponding filtering suppression gain through the adjusted dynamic filter group, so that howling suppression can be completed after filtering processing by the dynamic filter group.

2. The howling suppression device according to claim 1, characterized in that: The output end of the signal acquisition module is connected to a first limiter, which is used to limit the maximum amplitude of the digital input signal. The output limited digital signal is divided into two paths and sent to the howling suppression detection module and the buffer respectively.

3. The howling suppression device according to claim 2, characterized in that: An internal gain module is provided between the signal acquisition module and the signal acquisition module, for performing overall gain adjustment on the digital input signal.

4. The howling suppression device according to claim 1, characterized in that: The rear side of the dynamic filter bank is connected to a second limiter, and the second limiter is used to limit the maximum amplitude of the output signal of the dynamic filter bank.

5. The howling suppression device according to claim 1, characterized in that: The dynamic filter group includes two or more filter groups. The howling suppression detection module generates corresponding filter coefficients in sequence according to digital input signals of different frames. Each filter group alternately receives the filter coefficients generated in sequence and adjusts the filter frequency, filter bandwidth and suppression gain of each filter in each filter group.

6. A howling suppression method, characterized in that: The following steps are included: Collect the voice signal input from the microphone and the amplified signal fed back from the speaker, and convert them into digital input signals; detecting a frequency characteristic of the digital input signal, analyzing a probability corresponding to howling occurring at each frequency point in the digital input signal, and correspondingly binding a frequency value, a gain, and a probability of howling occurring at each frequency point of the digital input signal as a howling suppression parameter; and forming a filter coefficient for adjusting the gain of a filter according to the howling suppression parameter, wherein the filter coefficient includes a filter frequency, a filter bandwidth, and a corresponding suppression gain; Performing delay processing on the digital input signal to form a delayed digital signal; The filter frequency, filter bandwidth and suppression gain of the dynamic filter group are adjusted according to the filter coefficient, and the delayed digital signal is filtered by the dynamic filter group to achieve howling suppression.

7. The howling suppression method according to claim 6, characterized in that: In the process of analyzing the digital input signal to obtain a howling suppression parameter, and forming a filter coefficient for adjusting the gain of the filter according to the howling suppression parameter, The frequency feature extraction algorithm is used to extract and analyze the frequency features of each frequency point, obtain the howling probability P(k) of howling at each frequency point, and form a waveform of the howling probability P(k); Analyze each continuous waveform in the waveform of the howling probability P(k), determine the maximum probability value and the corresponding target frequency point in each continuous waveform, and record them as filter coefficients.

8. The howling suppression method according to claim 7, characterized in that: After determining the maximum probability value and the corresponding target frequency point in each continuous waveform, the two limit frequency values ​​±ndB on the left and right sides of the target frequency point are determined according to the preset attenuation value ndB, and the bandwidth value between the two limit frequency values ​​is used as the filter bandwidth of the filter coefficient, where n is a preset constant.

9. The howling suppression method according to claim 7 or 8, characterized in that: When analyzing each continuous waveform in the waveform of the howling probability P(k), if the maximum probability value in the continuous waveform, ie, the peak, is a continuous value, the frequency point corresponding to the center value of the continuous peak is taken as the target frequency point.

10. The howling suppression method according to claim 7 or 8, characterized in that: The maximum suppression gain of the filters in the dynamic filter bank is preset to Gmax, and the suppression gain in the filter coefficients is the product of the howling probability P(k) at the target frequency point and the maximum suppression gain Gmax.

11. The howling suppression method according to claim 6, characterized in that: The dynamic filter group includes two or more filter groups, each filter group includes several filters, each filter adjusts its own filtering frequency, filtering bandwidth and suppression gain according to the filtering coefficient, each filter corresponds to a filtering frequency, filtering bandwidth and corresponding suppression gain and is dynamically adjusted, and the delayed digital signal is processed by each filter group and then summed to obtain the optimized processed signal.

12. The howling suppression method according to claim 11, characterized in that: The dynamic filter group includes two filter groups, denoted as a first filter group and a second filter group, wherein the first filter group filters the delayed digital signal and then multiplies it by a proportional coefficient ratio, and the second filter group filters the delayed digital signal and then multiplies it by 1-ratio, wherein the proportional coefficient ratio switches between 0 and 1.

13. The howling suppression method according to claim 12, characterized in that: The switching period of the proportional coefficient ratio is equal to the output period of the filter coefficient.

14. The howling suppression method according to claim 12 or 13, characterized in that: The output signal after the delayed digital signal is filtered by the first filter group is recorded as vout1, the output signal after the delayed digital signal is filtered by the second filter group is recorded as vout2, and the output signal after the delayed digital signal is filtered by the dynamic filter group is recorded as vout, vout=ratio*vout1+(1-ratio)*vout2, When the proportional coefficient ratio=0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the first filter group and the filter frequency, filter bandwidth and suppression gain of the first filter group are adjusted according to the filter coefficient. After the first filter group is updated, the proportional coefficient ratio gradually changes from 0 to 1. When the proportional coefficient ratio becomes 1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal. When the proportional coefficient ratio=1, the digital input signal to be processed is sent to the first filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout1 after the first filter group filters the delayed digital signal. If the filter coefficient is updated, the updated filter coefficient is sent to the second filter group and the filtering frequency, filtering bandwidth and suppression gain of the second filter group are adjusted according to the filter coefficient. After the update of the second filter group is completed, the proportional coefficient ratio gradually changes from 1 to 0. When the proportional coefficient ratio becomes 0, the digital input signal to be processed is sent to the second filter group for filtering processing, and the output signal vout after the dynamic filter group filters the delayed digital signal is equal to the output signal vout2 after the second filter group filters the delayed digital signal.

15. The howling suppression method according to any one of claims 12 to 14, characterized in that: The method sequentially generates corresponding filter coefficients based on digital input signals of different frames, and sequentially sends them to the dynamic filter group. Each filter group in the dynamic filter group alternately receives each of the filter coefficients. Each filter group adjusts the filter frequency, filter bandwidth and suppression gain of each filter according to the received filter coefficient. The adjusted filter group receives the digital input signal corresponding to the filter coefficient and performs filtering processing.

Images