A multi-microphone-based wind noise suppression method and device

Abstract

The present invention discloses a wind noise suppression method and device based on multiple microphones. The method comprises the following steps: step S1, selecting an output microphone signal according to the low-frequency energy of each microphone; step S2, suppressing the wind noise frequency of the output microphone signal according to the energy relationship between the output microphone signal and each frequency point of other microphone signals; step S3, performing differential analysis on the main microphone and other microphone signals, taking the low-frequency signal for wind noise detection and wind noise magnitude estimation, and performing wind noise suppression in combination with the continuity characteristics of the wind noise.

Description

Technical Field

[0001] The present invention relates to the technical field of audio signal processing, and in particular to a method and device for suppressing wind noise based on multiple microphones. Background Art

[0002] When communication equipment or recording equipment is used outdoors, it will inevitably be affected by wind noise, which will reduce the clarity of sound pickup. Wind noise suppression is generally divided into two methods: active and passive. Passive wind noise suppression is to cover the microphone with a windscreen to reduce the wind speed entering the microphone, thereby reducing the wind noise energy collected by the microphone; active wind noise suppression is to detect, estimate and suppress wind noise on the digital signal collected by the microphone. Generally speaking, active wind noise suppression is more likely to cause loss of target audio signals than passive wind noise suppression, and passive wind noise suppression is easily limited by the size and geometry of the equipment. The present invention discusses active wind noise suppression.

[0003] Active wind noise suppression algorithms can be divided into single-microphone wind noise suppression and multi-microphone wind noise suppression according to the number of microphones. Among them, single-microphone wind noise suppression uses the characteristics of wind noise (spectral flatness, high-low frequency energy ratio, center of mass position, etc.) for detection, estimation and suppression. Because wind noise usually has greater energy than the target signal and is extremely unstable, it is difficult to take into account both the amount of wind noise suppression and the amount of target signal retention with only one microphone signal; multi-microphone wind noise suppression, based on the single-microphone, can utilize the relevant information between microphone signals to provide a guarantee for the wind noise suppression effect.

[0004] The wind noise signals between multiple microphones are very different from other audio signals: because wind noise is generated by the direct push of gas flow on the microphone membrane, wind noise is generally not generated at the same time for different microphones. Even if wind noise occurs at the same time, the amplitude and phase information of multiple wind noises are difficult to be consistent. To this end, there are two methods to suppress wind noise between multiple microphones: one is to use the correlation between microphones to detect and suppress wind noise, but the correlation will use more multiplication and division operations, the calculation amount is relatively large, and the correlation will use smoothing operations, resulting in insufficient timeliness in estimating wind noise and affecting the detection and suppression effect; second, multiple microphones can be placed in different directions, so that the probability of wind noise appearing in all microphones at the same time is small, and the microphone with less wind noise can be selected as the algorithm output. This method can reduce the impact of wind noise and almost no loss of target signal. The disadvantage is that as the number of microphones decreases, the wind noise suppression effect will deteriorate, and the requirements for the microphone direction limit the use of this method, and frequent switching of output microphone signals will inevitably affect the listening experience. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a wind noise suppression method and device based on multiple microphones, so as to ensure the wind noise suppression effect of the multiple microphones while making the calculation amount relatively small.

[0006] To achieve the above and other objectives, the present invention provides a wind noise suppression method based on multiple microphones, comprising the following steps:

[0007] Step S1, selecting an output microphone signal according to the low-frequency energy of each microphone;

[0008] Step S2 : suppressing the wind noise frequency point of the output microphone signal according to the energy magnitude relationship between the output microphone signal and each frequency point of other microphone signals.

[0009] Preferably, step S1 further comprises:

[0010] Step S100, performing frame division and time-frequency transformation on each microphone signal to obtain a frequency domain signal, taking the value in the low frequency interval of the frequency domain signal to calculate the low frequency amplitude and / or the low frequency average amplitude;

[0011] Step S101 : comparing the low-frequency amplitudes and / or low-frequency average amplitudes of the microphone signals, and selecting a frequency domain signal with low energy to switch to as the output microphone signal.

[0012] Preferably, no secondary switching is performed within a specified time after the switching.

[0013] Preferably, in step S101 , when microphone switching is performed, weight smoothing is performed first within a few frames before the switching and then the complete switching is performed.

[0014] Preferably, step S2 further comprises:

[0015] Step S200, calculating the frequency domain amplitude vector of the output microphone signal and other microphone signals;

[0016] Step S201, comparing the magnitude of each frequency point of the frequency domain amplitude vector of the output microphone signal with that of other microphone signals, and calculating the gain factor of each frequency point of the output microphone signal according to the comparison result;

[0017] Step S202 : selecting a minimum gain factor for each frequency point, and using the gain factor to suppress the frequency domain signal of the output microphone signal.

[0018] Preferably, step S2 includes:

[0019] Step S2a, the microphone that outputs the microphone signal is used as the main microphone, a secondary microphone is selected, and the frequency domain amplitude vector of the main microphone signal and the secondary microphone signal is calculated;

[0020] Step S2b, comparing the magnitude of each frequency point of the frequency domain amplitude vector of the main microphone signal and the auxiliary microphone signal, and calculating the gain factor of each frequency point of the main microphone signal according to the comparison result;

[0021] Step S2c, performing suppression processing on the frequency domain signal of the main microphone signal using the gain factor;

[0022] Step S2d: Return to step S2a to select another secondary microphone until all secondary microphones are selected.

[0023] Preferably, the magnitude of each frequency point of the frequency domain amplitude of the two microphones is compared, and if the magnitude of each frequency point is satisfied,

[0024] ampD1(k)>ampD2(k)*thr2

[0025] Where k represents the current frame frequency index, thr2 represents the amplitude ratio threshold of the primary microphone and the secondary microphone at the kth frequency point, and the gain factor gain1(k) of the frequency point is:

[0026] gain1(k)=ampD2(k) / ampD1(k)

[0027] Otherwise, the gain factor of the frequency point is 1.

[0028] Preferably, the method further comprises:

[0029] Step S3: Differentiate the frequency domain signals of the main microphone and other microphones, detect the wind noise frame, and estimate and suppress the wind noise level.

[0030] Preferably, step S3 further comprises:

[0031] Step S300: Differentiate the frequency domain signals of the main microphone and the other microphones to obtain a differential signal diffD, calculate the low-frequency energy mean diffDM of the differential signal diffD, and determine whether it is a wind noise frame based on the value of the low-frequency energy mean diffDM;

[0032] Step S301: If the current frame is determined to be a wind noise frame, the audio average frequency domain amplitude smoothD is not updated, and the wind noise level is estimated based on the low-frequency energy mean diffDM value and the smoothD value; otherwise, the audio average frequency domain amplitude smoothD is updated.

[0033] Step S303 : Calculate a wind noise suppression factor according to the estimated wind noise amplitude, and apply the wind noise suppression factor to the frequency spectrum to be output.

[0034] To achieve the above object, the present invention further provides a wind noise suppression device based on multiple microphones, comprising:

[0035] An output microphone signal selection unit, configured to select an output microphone signal according to the low-frequency energy of each microphone;

[0036] The first wind noise suppression unit is configured to suppress the wind noise frequency point of the output microphone signal according to the energy magnitude relationship between the output microphone signal and each frequency point of other microphone signals.

[0037] Compared with the existing technology, the present invention provides a wind noise suppression method and device based on multiple microphones. By exchanging the output microphones, the minimum energy frequency point is taken as the output, and the wind noise frequency point of the output microphone signal is suppressed according to the energy relationship between the output microphone signal and each frequency point of other microphone signals, and the wind noise is estimated by using the differential results of the microphone signals. It can not only suppress wind noise in most scenarios, but also reduce the computational complexity of the algorithm, and has high practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flowchart of the steps of a multi-microphone-based wind noise suppression method of the present invention;

[0039] Figure 2 This is a system structure diagram of a wind noise suppression device based on multiple microphones according to the present invention;

[0040] Figure 3 This is a structural diagram of the first wind noise suppression unit in the first embodiment of the present invention;

[0041] Figure 4 2 is a structural diagram of a first wind noise suppression unit in a second embodiment of the present invention;

[0042] Figure 5 Flowchart of a multi-microphone wind noise suppression method according to an embodiment of the present invention;

[0043] Figure 6 Detailed flowchart of step 1 in an embodiment of the present invention.

[0044] Figure 7 is a detailed flow chart of step 2 in an embodiment of the present invention;

[0045] Figure 8 Detailed flowchart of step three in an embodiment of the present invention;

[0046] Figure 9 2 is a structural diagram of a wind noise suppression device based on multiple microphones in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following describes the embodiments of the present invention using specific examples and accompanying drawings. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through other different specific examples, and the details in this specification may be modified and altered based on different viewpoints and applications without departing from the spirit of the present invention.

[0048] Figure 1 This is a flow chart of the steps of a wind noise suppression method based on multiple microphones of the present invention. Figure 1 As shown, the present invention provides a wind noise suppression method based on multiple microphones, comprising the following steps:

[0049] Step S1: Select an output microphone signal according to the low-frequency energy of each microphone.

[0050] In the present invention, step S1 further includes:

[0051] Step S100 : performing frame division and time-frequency transformation on each microphone signal to obtain a frequency domain signal D, and taking the value in the low frequency interval of the frequency domain signal D to calculate the low frequency amplitude and / or low frequency average amplitude.

[0052] Specifically, assuming there are n microphones, for each microphone, its time domain signal d, is framed and discrete Fourier transformed (DFT) to obtain the frequency domain signal D, and the value in the low-frequency interval of the frequency domain signal D is taken to calculate the low-frequency amplitude and or the low-frequency average amplitude EM.

[0053] In one embodiment of the present invention, a low-frequency cutoff frequency point K is selected, and the low-frequency amplitude sum M below the cutoff frequency of the frequency domain signal D of each microphone signal is calculated:

[0054]

[0055] In another embodiment of the present invention, a low-frequency cutoff frequency point K is selected, and the low-frequency average amplitude EM of the frequency domain signal D of each microphone signal below the cutoff frequency is calculated:

[0056]

[0057] Where EM represents the low-frequency average amplitude of each microphone, k represents the frequency subscript, and abs represents the modulo operation.

[0058] Step S101 : compare the low-frequency amplitudes and / or low-frequency average amplitudes of the microphone signals, select the frequency domain signal with the smallest energy as the output microphone signal, and define the corresponding microphone as the main microphone.

[0059] In a specific embodiment of the present invention, taking the comparison of low-frequency average amplitudes as an example, a microphone is first selected as a main microphone and a microphone is selected as a secondary microphone. After the above step S100, the time domain signal d1 of the main microphone is framed and discrete Fourier transformed (DFT) to obtain a frequency domain signal D1, and the value in the low-frequency interval of the frequency domain signal D1 is taken to obtain a low-frequency average amplitude EM1. The time domain signal d2 of the secondary microphone is framed and discrete Fourier transformed (DFT) to obtain a frequency domain signal D2, and the value in the low-frequency interval of the frequency domain signal D2 is taken to obtain a low-frequency average amplitude EM2. The low-frequency average amplitudes EM1 and EM2 are specifically as follows:

[0060]

[0061]

[0062] When the consistency between the two microphones is not very bad, if there is wind noise in one of the microphones, the low-frequency energy corresponding to the microphone is relatively large. Therefore, in the present invention, the microphone with smaller low-frequency energy is selected as the output microphone.

[0063] Specifically, in this step, the low-frequency average amplitude EM1 of the main microphone signal is compared with the low-frequency average amplitude EM2 of the auxiliary microphone signal. If the low-frequency amplitude averages of the two microphones satisfy the following relationship:

[0064] EM1>EM2*thr1 (3)

[0065] This indicates that the low-frequency energy of the primary microphone is greater than that of the secondary microphone. The probability of wind noise in the primary microphone is high, so the primary and secondary microphones are switched. thr1 represents the threshold of the ratio of the primary microphone energy to the secondary microphone amplitude.

[0066] Of course, the method of using the low-frequency amplitudes of the two microphones and calculating the energy ratio to select the frequency domain signal with smaller energy as the output microphone signal is the same and will not be described in detail here.

[0067] After the above process, the output microphone signal has been determined between the two microphones, and the corresponding microphone is used as the main microphone. When the number of microphones exceeds 2, the low-frequency amplitude and / or low-frequency average amplitude of the main microphone is compared one by one with the low-frequency amplitude and / or low-frequency average amplitude of other microphones, and the frequency domain signal with small energy is selected as the output microphone signal, and finally the frequency domain signal with the smallest energy is selected as the output microphone signal.

[0068] Preferably, when calculating the low-frequency amplitude mean or amplitude sum, time domain smoothing can also be performed. Taking the calculation of the low-frequency amplitude mean as an example, the specific method is as follows:

[0069]

[0070]

[0071] Where gamma is a number between 0 and 1, indicating a smoothing factor.

[0072] Preferably, in order to avoid affecting the listening experience due to frequent microphone switching, no secondary switching is performed within a specified time after the switching (the specified time can be set to 1 second, for example).

[0073] Preferably, to avoid waveform distortion caused by microphone switching, weight smoothing can be performed within the first few frames of the switch. Specifically, when switching the auxiliary microphone signal to the main microphone signal for output, there may be a problem of signal discontinuity and distortion. To avoid this problem, the present invention does not switch directly to the auxiliary microphone signal when switching. Instead, the weight smoothing is performed within the first few frames of the switch, for example, the auxiliary microphone signal accounts for 0.5, the main microphone signal accounts for 0.5, and after a few frames, the full switch to the auxiliary microphone signal as output is completed.

[0074] As can be seen, after the above microphone switching operation, the wind noise energy in the primary microphone is generally smaller than that in the secondary microphone. However, there are still two problems: 1) Since the switch cannot be switched again for a period of time after the switch, the wind noise energy of the primary microphone may be greater than that of the secondary microphone during this period; 2) If both microphones contain wind noise at the same time, the effectiveness of the microphone switching method will be limited. Therefore, further treatment is required.

[0075] Step S2 : suppressing the wind noise frequency point of the output microphone signal according to the energy magnitude relationship between the output microphone signal and each frequency point of other microphone signals.

[0076] In one embodiment of the present invention, step S2 further includes:

[0077] Step S200: Calculate the frequency domain amplitude vector of the output microphone signal and other microphone signals.

[0078] In a specific embodiment of the present invention, the microphone corresponding to the output microphone signal determined in step S1 is defined as the main microphone. Assume that it is the microphone of the frequency domain signal D1, and the other microphones are the microphones of the frequency domain signal D2. The frequency domain amplitude vectors ampD1 and ampD2 of the two microphones are calculated.

[0079] ampD1=abs(D1) (6)

[0080] ampD2=abs(D2) (7)

[0081] Where abs represents the modulo operation.

[0082] Step S201 : comparing the magnitude of each frequency point of the frequency domain amplitude vectors of the output microphone signal and other microphone signals, and calculating the gain factor of each frequency point of the output microphone signal according to the comparison result.

[0083] Specifically, for the main microphone of the frequency domain signal D1 and the auxiliary microphone of the frequency domain signal D2, compare the magnitude of each frequency point of the frequency domain amplitude of the two microphones. If the magnitude of each frequency point is satisfied,

[0084] ampD1(k)>ampD2(k)*thr2 (8)

[0085] Where k represents the frequency index of the current frame, and thr2 represents the amplitude ratio threshold of the primary microphone and the secondary microphone at the kth frequency (thr2 can be a fixed value or a value that changes with the frequency). This indicates that there is a high probability of wind noise at the kth frequency of the primary microphone signal in the current frame, and wind noise suppression is required. At this time, the gain factor gain1(k) of this frequency is:

[0086] gain1(k)=ampD2(k) / ampD1(k) (9)

[0087] Otherwise, the gain factor of the frequency point is 1.

[0088] In the case where there are multiple secondary microphones, each frequency point of the output microphone signal has multiple gain factors.

[0089] Step S202 : selecting a minimum gain factor for each frequency point, and using the gain factor to suppress the frequency domain signal of the output microphone signal.

[0090] In the present invention, through step S1, the microphone corresponding to the output microphone signal is used as the main microphone, and the frequency domain signal D1 of the main microphone is multiplied by the selected gain factor as the output, as follows:

[0091] D1(k)=D1(k)*gain1(k) (10)

[0092] According to the above formulas (9) and (10), if it is determined that there is wind noise at the kth frequency point of the main microphone, the amplitude of the kth frequency point of the main microphone is suppressed to the same as that of the secondary microphone. The advantage of doing so is that it can ensure that the wind noise output by the main microphone is minimized at each frequency point and that the voice is not lost to the greatest extent possible.

[0093] In another embodiment of the present invention, step S2 further includes:

[0094] Step S2a: Select a pair of microphones and calculate the frequency domain amplitude vector of the output microphone signal and the signal of the pair of microphones.

[0095] In a specific embodiment of the present invention, the microphone corresponding to the output microphone signal determined in step S1 is defined as the main microphone. Assuming that it is the microphone of the frequency domain signal D1, a secondary microphone of the frequency domain signal D2 is selected, and the frequency domain amplitude vectors ampD1 and ampD2 of the two microphones are calculated.

[0096] ampD1=abs(D1)

[0097] ampD2=abs(D2)

[0098] Where abs represents the modulo operation.

[0099] Step S2b: comparing the output microphone signal with each frequency point of the frequency domain amplitude vector of the microphone signal, and calculating the gain factor of each frequency point of the output microphone signal according to the comparison result.

[0100] Specifically, for the main microphone of the frequency domain signal D1 and the auxiliary microphone of the frequency domain signal D2, compare the magnitude of each frequency point of the frequency domain amplitude of the two microphones. If the magnitude of each frequency point is satisfied,

[0101] ampD1(k)>ampD2(k)*thr2

[0102] Where k represents the frequency index of the current frame, and thr2 represents the amplitude ratio threshold of the primary microphone and the secondary microphone at the kth frequency (thr2 can be a fixed value or a value that changes with the frequency). This indicates that there is a high probability of wind noise at the kth frequency of the primary microphone signal in the current frame, and wind noise suppression is required. At this time, the gain factor gain1(k) of this frequency is:

[0103] gain1(k)=ampD2(k) / ampD1(k)

[0104] Otherwise, the gain factor of the frequency point is 1.

[0105] In the case where there are multiple secondary microphones, each frequency point of the output microphone signal has multiple gain factors.

[0106] Step S2c: using a gain factor to suppress the frequency domain signal of the output microphone signal.

[0107] In the present invention, through step S1, the microphone corresponding to the output microphone signal is used as the main microphone, and the frequency domain signal D1 of the main microphone is multiplied by the selected gain factor as the output, as follows:

[0108] D1(k)=D1(k)*gain1(k)

[0109] According to the above formula, if wind noise is detected at the kth frequency point of the main microphone, the amplitude of the kth frequency point of the main microphone will be suppressed to the same level as that of the secondary microphone. This ensures that the wind noise output by the main microphone is minimized at each frequency point while also minimizing voice loss.

[0110] Step S2d: Return to step S2a to select another secondary microphone until all secondary microphones are selected.

[0111] After the above steps, it can be ensured that the final output has the minimum wind noise at each frequency point. However, there may still be a problem: if both microphones have a large wind noise at the same time, then even if the output is the minimum wind noise at each frequency point of the two microphones, the energy of the wind noise will still be relatively large. Therefore, preferably, after step S2, the wind noise suppression method based on multiple microphones of the present invention also includes:

[0112] Step S3: Differentiate the frequency domain signals of the main microphone and other microphones, detect the wind noise frame, and estimate and suppress the wind noise level.

[0113] Specifically, step S3 further includes:

[0114] Step S300 : Differentiate the frequency domain signals of the main microphone and other microphones to obtain a differential signal diffD, calculate the low-frequency energy mean diffDM of the differential signal diffD, and determine whether it is a wind noise frame based on the value of the low-frequency energy mean diffDM.

[0115] Specifically, assuming the frequency domain signal D1 of the main microphone and the frequency domain signal D2 of the other microphone, the frequency domain signals D1 and D2 of the two microphones are differentiated, and the difference between the frequency domain signals D1 and D2 of the two microphones is as follows:

[0116] diffD=D1-D2 (11)

[0117] The difference signal diffD has the following characteristics: if the two microphones are not far apart (for example, <0.1m), the energy of speech and other active sounds must be very small in the low-frequency band. The frequency domain signals D1 and D2 have real and imaginary components, which can be converted into amplitude and phase in polar coordinate form. When the wavelength of the sound is much larger than the distance between the microphones, the amplitude and phase of the two frequency domain signals D1 and D2 should be close to equal, so the diffD signal after differentiation is close to 0. Because wind noise is generated independently on the two microphone membranes, the amplitude and phase are independent, so the requirement that diffD is close to 0 after differentiation is not satisfied. And because wind noise is mainly distributed in low frequencies, the low-frequency part of the diffD signal will form a large energy contrast in the presence and absence of wind noise. This can be used as a basis for wind noise detection. Therefore, the low-frequency energy mean diffDM of the difference signal diffD is calculated:

[0118]

[0119] Among them, K2 is the upper limit of the frequency value of the low frequency band.

[0120] For the low-frequency energy mean diffDM of the differential signal diffD, if

[0121] diffDM>thr3 (13)

[0122] The current frame is determined to be a wind noise frame and requires additional wind noise suppression, where thr3 is the wind noise frame detection threshold.

[0123] Preferably, if there are multiple low-frequency energy mean diffDM values, the multiple low-frequency energy mean diffDM values are averaged, and then the current frame is determined to be a wind noise frame based on the averaged low-frequency energy mean diffDM value. Specifically, when there are multiple secondary microphones, the primary microphone and each secondary microphone perform a differential analysis to obtain a series of low-frequency energy mean diffDM values, and the average of these multiple low-frequency energy mean diffDM values is calculated.

[0124] This determination method essentially leverages the low-frequency correlation of the two microphone signals, but compared to specific correlation calculation methods, it requires much less computation and is latency-free. In addition to this method for determining wind noise frames, it can also be combined with other methods such as high- and low-frequency energy ratios and centroid location for a joint determination.

[0125] Step S301: If the current frame is determined to be a wind noise frame, the audio average frequency domain amplitude smoothD is not updated, and the wind noise level is estimated based on the low-frequency energy mean diffDM value and smoothD. Otherwise, the audio average frequency domain amplitude smoothD of the output microphone signal is updated.

[0126] Because the low-frequency part of the differential signal diffD is wind noise, the low-frequency energy mean diffDM itself can be used as a reference for estimating the magnitude of wind noise. If the current frame is judged to be non-wind noise, the spectrum amplitude vector smoothD of the non-wind noise is updated:

[0127] smoothD=delta*smoothD+(1-delta)*ampD1 (14)

[0128] Among them, ampD1 is the spectrum amplitude of the main microphone in the current frame, delta is the smoothing factor with a value between 0 and 1, and smoothD represents the average spectrum amplitude of the audio signal when there is no wind noise.

[0129] If the current frame is determined to be a wind noise frame, stop updating the smoothD value and use the non-wind noise spectrum amplitude vector smoothD and the low-frequency energy mean diffDM to estimate the wind noise size:

[0130] windAmp1=ampD1-smoothD (15)

[0131] windAmp=min(diffDM,windAmp1) (16)

[0132] Wherein, formula (15) indicates that the frequency domain amplitude of the wind noise frame minus the average amplitude of the non-wind noise frame is used as a preliminary estimate of the wind noise amplitude spectrum, and formula (16) indicates that each frequency point of the preliminary estimated wind noise amplitude spectrum is compared with the low-frequency wind noise magnitude estimate obtained in step S300 (diffDM is the low-frequency energy mean, which is also a preliminary estimate of the wind noise magnitude when there is wind noise), and the minimum value is taken as the final wind noise estimate.

[0133] Step S303 : Calculate a wind noise suppression factor according to the estimated wind noise amplitude, and apply the wind noise suppression factor to the frequency spectrum to be output.

[0134] Specifically, step S303 further includes:

[0135] Step S303a: determining the frequency band of the wind noise based on the continuity of the wind noise spectrum.

[0136] Because wind noise is mainly concentrated in low frequencies, it is necessary to estimate the frequency band of wind noise when determining it as a wind noise frame. The method is based on the continuity of the wind noise spectrum, that is, the wind noise energy decreases from low frequency to high frequency. First, the amplitude spectrum of the wind noise frame is divided into several frequency bands. 11 ,ampD 12 ,…ampD 1N , if satisfied

[0137] ampD 1n<ampD 1n+1 (17)

[0138] It can be determined that the n+1 frequency band and above are not wind noise frequencies, and wind noise suppression is performed on the n frequency band and below.

[0139] Step S303b: for the wind noise frame in the determined frequency band where the wind noise is located, a wind noise suppression factor is calculated according to the spectrum amplitude of the current frame and the estimated wind noise amplitude, and the wind noise suppression factor is applied to the spectrum to be output.

[0140] Assuming that according to formula (17) we can judge that the frequency points k below the K3 frequency point are all the frequency points where wind noise exists, then for the frequency points 1 to K3, we have

[0141] gain2(k)=(ampD1(k)-windAmp(k)) / ampD1(k) (18)

[0142] D1(k)=D1(k)*gain2(k) (19)

[0143] That is, the wind noise suppression factor gain2 is first calculated based on the spectrum amplitude ampD1(k) of the current frame and the estimated wind noise amplitude windAmp(k), and then the wind noise suppression factor gain2 is applied to the spectrum to be output, that is, it is multiplied by the frequency domain output signal D1 of step S2.

[0144] Figure 2 This is a schematic diagram of the structure of a wind noise suppression device based on multiple microphones in the present invention. Figure 2 As shown, the present invention provides a wind noise suppression device based on multiple microphones, comprising:

[0145] The output microphone signal selection unit 20 is used to select the output microphone signal according to the low-frequency energy of each microphone.

[0146] In the present invention, the output microphone signal selection unit 20 further includes:

[0147] The time-frequency conversion and low-frequency energy calculation unit 201 is used to perform frame division and time-frequency conversion on each microphone signal to obtain a frequency domain signal D, and calculate the low-frequency amplitude and / or low-frequency average amplitude by taking the value in the low-frequency interval of the frequency domain signal D.

[0148] Specifically, assuming there are n microphones, for each microphone, its time domain signal d, is framed and discrete Fourier transformed (DFT) to obtain the frequency domain signal D, and the value in the low-frequency interval of the frequency domain signal D is taken to calculate the low-frequency amplitude and or the low-frequency average amplitude EM.

[0149] In one embodiment of the present invention, a low-frequency cutoff frequency point K is selected, and the low-frequency amplitude sum M below the cutoff frequency of the frequency domain signal D of each microphone signal is calculated:

[0150]

[0151] In another embodiment of the present invention, a low-frequency cutoff frequency point K is selected, and the low-frequency average amplitude EM of the frequency domain signal D of each microphone signal below the cutoff frequency is calculated:

[0152]

[0153] Where EM represents the low-frequency average amplitude of each microphone, k represents the frequency subscript, and abs represents the modulo operation.

[0154] The output microphone switching unit 202 is used to compare the low-frequency amplitude and / or low-frequency average amplitude of each microphone signal, select the frequency domain signal with the smallest energy as the output microphone signal, and define the corresponding microphone as the main microphone.

[0155] In a specific embodiment of the present invention, taking the comparison of low-frequency average amplitudes as an example, a microphone is first selected as a main microphone and a microphone is selected as a secondary microphone. After the time-frequency conversion and low-frequency energy calculation unit 201, the time domain signal d1 of the main microphone is framed and discrete Fourier transformed (DFT) to obtain a frequency domain signal D1, and the value in the low-frequency interval of the frequency domain signal D1 is taken to obtain a low-frequency average amplitude EM1. The time domain signal d2 of the secondary microphone is framed and discrete Fourier transformed (DFT) to obtain a frequency domain signal D2, and the value in the low-frequency interval of the frequency domain signal D2 is taken to obtain a low-frequency average amplitude EM2. The low-frequency average amplitudes EM1 and EM2 are specifically as follows:

[0156]

[0157]

[0158] When the consistency between the two microphones is not very bad, if there is wind noise in one of the microphones, the low-frequency energy corresponding to the microphone is relatively large. Therefore, in the present invention, the microphone with smaller low-frequency energy is selected as the output microphone.

[0159] Specifically, the output microphone switching unit 202 compares the low-frequency average amplitude EM1 of the main microphone signal with the low-frequency average amplitude EM2 of the auxiliary microphone signal. If the low-frequency amplitude averages of the two microphones satisfy the following relationship:

[0160] EM1>EM2*thr1

[0161] This indicates that the low-frequency energy of the primary microphone is greater than that of the secondary microphone. The probability of wind noise in the primary microphone is high, so the primary and secondary microphones are switched. thr1 represents the threshold of the ratio of the primary microphone energy to the secondary microphone amplitude.

[0162] Of course, the method of using the low-frequency amplitudes of the two microphones and calculating the energy ratio to select the frequency domain signal with smaller energy as the output microphone signal is the same and will not be described in detail here.

[0163] After the above process, the output microphone signal has been determined between the two microphones, and the corresponding microphone is used as the main microphone. When the number of microphones exceeds 2, the low-frequency amplitude and / or low-frequency average amplitude of the main microphone is compared one by one with the low-frequency amplitude and / or low-frequency average amplitude of other microphones, and the frequency domain signal with small energy is selected as the output microphone signal, and finally the frequency domain signal with the smallest energy is selected as the output microphone signal.

[0164] Preferably, when calculating the low-frequency amplitude mean or amplitude sum, time domain smoothing can also be performed. Taking the calculation of the low-frequency amplitude mean as an example, the specific method is as follows:

[0165]

[0166]

[0167] Where gamma is a number between 0 and 1, indicating a smoothing factor.

[0168] Preferably, in order to avoid affecting the listening experience due to frequent microphone switching, no secondary switching is performed within a specified time after the switching (the specified time can be set to 1 second, for example).

[0169] Preferably, in order to avoid waveform distortion caused by switching, weight smoothing can be performed within the first few frames of switching.

[0170] As can be seen, after the above microphone switching operation, the wind noise energy in the primary microphone is generally smaller than that in the secondary microphone. However, there are still two problems: 1) Since the switch cannot be switched again for a period of time after the switch, the wind noise energy of the primary microphone may be greater than that of the secondary microphone during this period; 2) If both microphones contain wind noise at the same time, the effectiveness of the microphone switching method will be limited. Therefore, further treatment is required.

[0171] The first wind noise suppression unit 21 is configured to suppress the wind noise frequency points of the output microphone signal according to the energy magnitude relationship between the output microphone signal and each frequency point of other microphone signals.

[0172] In one embodiment of the present invention, Figure 3 As shown, the first wind noise suppression unit 21 further includes:

[0173] The frequency domain amplitude vector calculation unit 210 is configured to calculate the frequency domain amplitude vectors of the output microphone signal and other microphone signals.

[0174] In a specific embodiment of the present invention, the microphone corresponding to the output microphone signal determined by the output microphone signal selection unit 20 is defined as the main microphone. Assume that it is the microphone of the frequency domain signal D1, and the other microphones are the microphones of the frequency domain signal D2. The frequency domain amplitude vectors ampD1 and ampD2 of the two microphones are calculated.

[0175] ampD1=abs(D1)

[0176] ampD2=abs(D2)

[0177] Where abs represents the modulo operation.

[0178] The gain factor calculation unit 211 is configured to compare the magnitude of each frequency point of the frequency domain amplitude vectors of the output microphone signal and other microphone signals, and calculate the gain factor of each frequency point of the output microphone signal according to the comparison result.

[0179] Specifically, for the main microphone of the frequency domain signal D1 and the auxiliary microphone of the frequency domain signal D2, compare the magnitude of each frequency point of the frequency domain amplitude of the two microphones. If the magnitude of each frequency point is satisfied,

[0180] ampD1(k)>ampD2(k)*thr2

[0181] Where k represents the frequency index of the current frame, and thr2 represents the amplitude ratio threshold of the primary microphone and the secondary microphone at the kth frequency (thr2 can be a fixed value or a value that changes with the frequency). This indicates that there is a high probability of wind noise at the kth frequency of the primary microphone signal in the current frame, and wind noise suppression is required. At this time, the gain factor gain1(k) of this frequency is:

[0182] gain1(k)=ampD2(k) / ampD1(k)

[0183] Otherwise, the gain factor of the frequency point is 1.

[0184] In the case where there are multiple secondary microphones, each frequency point of the output microphone signal has multiple gain factors.

[0185] The suppression processing unit 212 is configured to select a minimum gain factor for each frequency point, and perform suppression processing on the frequency domain signal of the output microphone signal using the gain factor.

[0186] In the present invention, the microphone corresponding to the output microphone signal is used as the main microphone, and the frequency domain signal D1 of the main microphone is multiplied by the selected gain factor as the output, specifically as follows:

[0187] D1(k)=D1(k)*gain1(k)

[0188] According to the above formula, if wind noise is detected at the kth frequency point of the main microphone, the amplitude of the kth frequency point of the main microphone will be suppressed to the same level as that of the secondary microphone. This ensures that the wind noise output by the main microphone is minimized at each frequency point while also minimizing voice loss.

[0189] In another embodiment of the present invention, Figure 4 As shown, the first wind noise suppression unit 21 further includes:

[0190] The frequency domain amplitude vector calculation unit 21a selects a pair of microphones and calculates the frequency domain amplitude vector of the output microphone signal and the signal of the pair of microphones.

[0191] In a specific embodiment of the present invention, the microphone corresponding to the output microphone signal is defined as the main microphone. Assume that it is the microphone of the frequency domain signal D1. Select a secondary microphone of the frequency domain signal D2. Calculate the frequency domain amplitude vectors ampD1 and ampD2 of the two microphones.

[0192] ampD1=abs(D1)

[0193] ampD2=abs(D2)

[0194] Where abs represents the modulo operation.

[0195] The gain factor calculation unit 21b is configured to compare the output microphone signal with each frequency point of the frequency domain amplitude vector of the microphone signal, and calculate the gain factor of each frequency point of the output microphone signal according to the comparison result.

[0196] Specifically, for the main microphone of the frequency domain signal D1 and the auxiliary microphone of the frequency domain signal D2, compare the magnitude of each frequency point of the frequency domain amplitude of the two microphones. If the magnitude of each frequency point is satisfied,

[0197] ampD1(k)>ampD2(k)*thr2

[0198] Where k represents the frequency index of the current frame, and thr2 represents the amplitude ratio threshold of the primary microphone and the secondary microphone at the kth frequency (thr2 can be a fixed value or a value that changes with the frequency). This indicates that there is a high probability of wind noise at the kth frequency of the primary microphone signal in the current frame, and wind noise suppression is required. At this time, the gain factor gain1(k) of this frequency is:

[0199] gain1(k)=ampD2(k) / ampD1(k)

[0200] Otherwise, the gain factor of the frequency point is 1.

[0201] In the case where there are multiple secondary microphones, each frequency point of the output microphone signal has multiple gain factors.

[0202] The suppression processing unit 21c is configured to perform suppression processing on the frequency domain signal of the output microphone signal by using a gain factor.

[0203] In the present invention, the microphone corresponding to the output microphone signal is used as the main microphone, and the frequency domain signal D1 of the main microphone is multiplied by the selected gain factor as the output, specifically as follows:

[0204] D1(k)=D1(k)*gain1(k)

[0205] According to the above formula, if wind noise is detected at the kth frequency point of the main microphone, the amplitude of the kth frequency point of the main microphone will be suppressed to the same level as that of the secondary microphone. This ensures that the wind noise output by the main microphone is minimized at each frequency point while also minimizing voice loss.

[0206] Traverse the processing unit 21d and return to the frequency domain amplitude vector calculation unit 21a to select another secondary microphone until all secondary microphones are selected.

[0207] The above process ensures that the final output has the minimum wind noise at each frequency point. However, there may still be a problem: if both microphones have large wind noise at the same time, then even if the output is the minimum wind noise at each frequency point of the two microphones, the energy of the wind noise will still be relatively large.

[0208] Therefore, preferably, the wind noise suppression device based on multiple microphones of the present invention further includes:

[0209] The second wind noise suppression unit 22 is configured to perform differentiation on the frequency domain signals of the main microphone and other microphones, detect wind noise frames, and estimate and suppress the wind noise level.

[0210] Specifically, the second wind noise suppression unit 22 further includes:

[0211] The differential unit 220 is configured to perform a differential on the frequency domain signals of the main microphone and the other microphones to obtain a differential signal diffD, and calculate a low-frequency energy mean diffDM of the differential signal diffD.

[0212] The wind noise frame judging unit 221 is configured to judge whether the frame is a wind noise frame according to the low-frequency energy mean diffDM value.

[0213] Specifically, assuming the frequency domain signal D1 of the main microphone and the frequency domain signal D2 of the other microphone, the frequency domain signals D1 and D2 of the two microphones are differentiated, and the difference between the frequency domain signals D1 and D2 of the two microphones is as follows:

[0214] diffD=D1-D2

[0215] The difference signal diffD has the following characteristics: if the two microphones are not far apart (for example, <0.1m), the energy of speech and other active sounds must be very small in the low-frequency band. The frequency domain signals D1 and D2 have real and imaginary components, which can be converted into amplitude and phase in polar coordinate form. When the wavelength of the sound is much larger than the distance between the microphones, the amplitude and phase of the two frequency domain signals D1 and D2 should be close to equal, so the diffD signal after differentiation is close to 0. Because wind noise is generated independently on the two microphone membranes, the amplitude and phase are independent, so the requirement that diffD is close to 0 after differentiation is not satisfied. And because wind noise is mainly distributed in low frequencies, the low-frequency part of the diffD signal will form a large energy contrast in the presence and absence of wind noise. This can be used as a basis for wind noise detection. Therefore, the low-frequency energy mean diffDM of the difference signal diffD is calculated:

[0216]

[0217] For the low-frequency energy mean diffDM of the differential signal diffD, if

[0218] diffDM>thr3

[0219] The current frame is determined to be a wind noise frame and requires additional wind noise suppression.

[0220] Preferably, if there are multiple low-frequency energy mean diffDM values, the multiple low-frequency energy mean diffDM values are averaged, and then the current frame is determined to be a wind noise frame based on the averaged low-frequency energy mean diffDM value. Specifically, when there are multiple secondary microphones, the primary microphone and each secondary microphone perform a differential analysis to obtain a series of low-frequency energy mean diffDM values, and the average of these multiple low-frequency energy mean diffDM values is calculated.

[0221] This determination method essentially leverages the low-frequency correlation of the two microphone signals, but compared to specific correlation calculation methods, it requires much less computation and is latency-free. In addition to this method for determining wind noise frames, it can also be combined with other methods such as high- and low-frequency energy ratios and centroid location for a joint determination.

[0222] The wind noise magnitude estimation unit 222 is configured to estimate the wind noise magnitude based on the low-frequency energy mean diffDM value and smoothD without updating the audio average frequency domain amplitude smoothD when determining that the current frame is a wind noise frame; otherwise, update the audio average frequency domain amplitude smoothD.

[0223] Because the low-frequency part of the differential signal diffD is wind noise, the low-frequency energy mean diffDM itself can be used as a reference for estimating the magnitude of wind noise. If the current frame is judged to be non-wind noise, the spectrum amplitude vector smoothD of the non-wind noise is updated:

[0224] smoothD=delta*smoothD+(1-delta)*ampD1

[0225] Among them, ampD1 is the spectrum amplitude of the main microphone in the current frame, delta is the smoothing factor with a value between 0 and 1, and smoothD represents the average spectrum amplitude of the audio signal when there is no wind noise.

[0226] If the current frame is determined to be a wind noise frame, the smoothD value is stopped from being updated, and the wind noise magnitude is estimated using the non-wind noise spectrum amplitude vector smoothD and the low-frequency energy mean diffDM. That is, the frequency domain amplitude of the wind noise frame minus the average amplitude of the non-wind noise frame is used as a preliminary estimate of the wind noise amplitude spectrum. Then, each frequency point of the preliminary estimated wind noise amplitude spectrum is compared with the low-frequency wind noise magnitude estimate obtained by the difference unit 220, and the minimum value is taken as the final wind noise estimation value, as follows:

[0227] windAmp1=ampD1-smoothD

[0228] windAmp=min(diffDM,windAmp1)

[0229] The wind noise suppression unit 223 is configured to calculate a wind noise suppression factor according to the estimated wind noise amplitude, and apply the wind noise suppression factor to the frequency spectrum to be output.

[0230] Specifically, the wind noise suppression unit 223 is specifically used to:

[0231] The frequency band of the wind noise is determined based on the continuity of the wind noise spectrum.

[0232] Because wind noise is mainly concentrated in low frequencies, it is necessary to estimate the frequency band of wind noise when determining it as a wind noise frame. The method is based on the continuity of the wind noise spectrum, that is, the wind noise energy decreases from low frequency to high frequency. First, the amplitude spectrum of the wind noise frame is divided into several frequency bands. 11 ,ampD 12 ,…ampD 1N , if satisfied

[0233] ampD 1n <ampD 1n+1

[0234] It can be determined that the n+1 frequency band and above are not wind noise frequencies, and wind noise suppression is performed on the n frequency band and below.

[0235] For the wind noise frame in the determined frequency band where the wind noise is located, a wind noise suppression factor is calculated according to the spectrum amplitude of the current frame and the estimated wind noise amplitude, and the wind noise suppression factor is applied to the spectrum to be output.

[0236] Assuming that according to formula (17) we can judge that the frequency points k below the K3 frequency point are all the frequency points where wind noise exists, then for the frequency points 1 to K3, we have

[0237] gain2(k)=(ampD1(k)-windAmp(k)) / ampD1(k)

[0238] D1(k)=D1(k)*gain2(k)

[0239] That is, the wind noise suppression factor gain2 is first calculated based on the spectrum amplitude ampD1(k) of the current frame and the estimated wind noise amplitude windAmp(k), and then the wind noise suppression factor gain2 is applied to the spectrum to be output, that is, it is multiplied by the frequency domain output signal D1 of the first wind noise suppression unit 21.

[0240] Example 1

[0241] In this embodiment, taking the dual microphones mic1 and mic2 as an example, Figure 5 As shown, a wind noise suppression method based on multiple microphones, the steps are as follows:

[0242] Step 1: Select the output microphone signal according to the low-frequency energy of the microphone.

[0243] like Figure 6 As shown, step one includes:

[0244] A. Assume that mic1 is the main microphone. Perform frame division and time-frequency transformation on the main microphone signal to obtain the frequency domain signal D1 of the main microphone. Take the value in the low-frequency range of the frequency domain signal D1 to calculate the low-frequency average amplitude EM1.

[0245] B. Similar to step A, assuming that mic2 is the secondary microphone, perform frame division and time-frequency transformation on the secondary microphone signal to obtain the frequency domain signal D2 of the secondary microphone. Take the value in the low-frequency range of the frequency domain signal D2 to calculate the low-frequency average amplitude EM2.

[0246] C. Compare the low-frequency average amplitude EM1 of the main microphone signal with the low-frequency average amplitude EM2 of the auxiliary microphone signal. If EM1 is greater than EM2, switch the auxiliary microphone signal with the main microphone signal, that is, use the frequency domain signal with smaller energy as the output signal D1 of this step.

[0247] It should be noted that the sum of low-frequency energy can also be compared in step C. After switching the output signal, it cannot be switched again for a period of time. The first few frames of the signal before switching can be the weighted output of D1 and D2.

[0248] Step 2: Preliminary suppression of the output signal is performed based on the magnitude of each frequency point of the two microphone signals.

[0249] like Figure 7 As shown, step two includes:

[0250] D. Calculate the frequency domain amplitude ampD1 of signal D1 and calculate the frequency domain amplitude ampD2 of signal D2.

[0251] E. Compare the magnitude relationship of each frequency point of ampD1 and ampD2. If the frequency point amplitude of ampD1 is greater than the frequency point amplitude of ampD2, calculate the gain factor of the frequency point; otherwise, the gain factor is 1.

[0252] F. Multiply the frequency domain signal D1 by the gain factor as the output.

[0253] When comparing the magnitude of each frequency point in the above step E, the gain factor is calculated only when ampD1 is greater than ampD2 by a certain value, and the gain factor should be between 0 and 1.

[0254] Step 3: Differentiate the two microphones to detect the wind noise frame and estimate and suppress the wind noise level.

[0255] like Figure 8 As shown, step three includes:

[0256] G. Differentiate signal D1 from signal D2 to obtain signal diffD.

[0257] H. Calculate the low-frequency energy mean diffDM of the signal diffD.

[0258] I. Determine whether it is a wind noise frame based on the diffDM value.

[0259] J. If the frame is determined to be a wind noise frame in step I, then the audio average frequency domain amplitude smoothD is not updated. Otherwise, smoothD is updated.

[0260] K. If step I determines that the frame is a wind noise frame, the amplitude of the wind noise spectrum is estimated according to the value of diffDM and the value of smoothD.

[0261] L. If the frame is determined to be a wind noise frame in step I, a wind noise suppression factor is calculated according to the estimated wind noise amplitude, and the wind noise suppression factor is multiplied by the frequency domain output signal D1 in step F.

[0262] Example 2

[0263] In this embodiment, a wind noise suppression device based on multiple microphones is provided, such as Figure 9 As shown, the device includes: a time-frequency conversion unit 10, a low-frequency energy calculation and comparison unit 11, an output microphone signal switching unit 12, a main and sub-microphone frequency point energy comparison unit 13, a wind noise suppression 1 unit 14, a difference unit 15, a wind noise frame determination unit 16, a wind noise size estimation unit 17, and a wind noise suppression 2 unit 18.

[0264] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Any skilled artisan may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be as set forth in the appended claims.

Claims

1. A wind noise suppression method based on multiple microphones, comprising the following steps: Step S1, selecting an output microphone signal according to the low-frequency energy of each microphone; Step S2, suppressing the wind noise frequency of the output microphone signal according to the energy magnitude relationship between the output microphone signal and each frequency point of other microphone signals; Step S3, performing differential analysis on the frequency domain signals of the main microphone and other microphones, detecting wind noise frames, and estimating and suppressing the wind noise level; Step S2 further comprises: Step S200, calculating the frequency domain amplitude vector of the output microphone signal and other microphone signals; Step S201, comparing the magnitude of each frequency point of the frequency domain amplitude vector of the output microphone signal with that of other microphone signals, and calculating the gain factor of each frequency point of the output microphone signal according to the comparison result; Step S202 : selecting a minimum gain factor for each frequency point, and using the gain factor to suppress the frequency domain signal of the output microphone signal.

2. The multi-microphone wind noise suppression method according to claim 1, wherein: Step S1 further comprises: Step S100, performing frame division and time-frequency transformation on each microphone signal to obtain a frequency domain signal, taking the value in the low frequency interval of the frequency domain signal to calculate the low frequency amplitude and / or the low frequency average amplitude; Step S101 : comparing the low-frequency amplitudes and / or low-frequency average amplitudes of the microphone signals, and selecting a frequency domain signal with low energy to switch to as the output microphone signal.

3. The multi-microphone wind noise suppression method according to claim 2, wherein: No secondary switching will be performed within the specified time after the switching.

4. The multi-microphone wind noise suppression method according to claim 2, wherein: In step S101 , when microphone switching is performed, weight smoothing is first performed within a few frames before the switching and then the complete switching is performed.

5. The multi-microphone wind noise suppression method according to claim 2, wherein: Step S2 includes: Step S2a, the microphone that outputs the microphone signal is used as the main microphone, a secondary microphone is selected, and the frequency domain amplitude vector of the main microphone signal and the secondary microphone signal is calculated; Step S2b, comparing the magnitude of each frequency point of the frequency domain amplitude vector of the main microphone signal and the auxiliary microphone signal, and calculating the gain factor of each frequency point of the main microphone signal according to the comparison result; Step S2c, performing suppression processing on the frequency domain signal of the main microphone signal using the gain factor; Step S2d: Return to step S2a to select another secondary microphone until all secondary microphones are selected.

6. A wind noise suppression method based on multiple microphones according to claim 1 or 5, characterized in that , compare the frequency domain amplitude of the two microphones at each frequency point, if it satisfies ； in, Indicates the current frame frequency index, Indicates the The gain factor of the frequency point is the amplitude ratio threshold of the main microphone and the auxiliary microphone. for: ； Otherwise, the gain factor of the frequency point is 1.

7. The multi-microphone wind noise suppression method according to claim 1, wherein: Step S3 further comprises: Step S300: Differentiate the frequency domain signals of the main microphone and other microphones to obtain a differential signal. , and calculate the differential signal The low-frequency energy mean , and according to The size of the value determines whether it is a wind noise frame; Step S301: If the current frame is determined to be a wind noise frame, the average frequency domain amplitude of the audio is not updated. , according to the low-frequency energy mean The size of the value and Size estimates the size of wind noise, otherwise updates the audio average frequency domain amplitude ; Step S303 : Calculate a wind noise suppression factor according to the estimated wind noise amplitude, and apply the wind noise suppression factor to the frequency spectrum to be output.

8. A wind noise suppression device based on multiple microphones, comprising: An output microphone signal selection unit, configured to select an output microphone signal according to the low-frequency energy of each microphone; a first wind noise suppression unit, configured to suppress a wind noise frequency point of the output microphone signal according to an energy magnitude relationship between each frequency point of the output microphone signal and other microphone signals; The second wind noise suppression unit first uses the main microphone signal to differentiate with the other microphones, and uses the low-frequency signal of the differential signal to detect wind noise frames and preliminarily estimate the wind noise level. Taking into account the extremely unstable and spectral continuity characteristics of wind noise, it further suppresses wind noise; The first wind noise suppression unit further comprises: a frequency domain amplitude vector calculation unit, configured to calculate the frequency domain amplitude vectors of the output microphone signal and other microphone signals; a gain factor calculation unit, configured to compare the magnitude of each frequency point of the frequency domain amplitude vectors of the output microphone signal with those of other microphone signals, and calculate the gain factor of each frequency point of the output microphone signal according to the comparison result; The suppression processing unit is used to select a minimum gain factor for each frequency point and perform suppression processing on the frequency domain signal of the output microphone signal using the gain factor.

Images