CN102740214B - Howling suppression method based on feedback signal spectrum estimation - Google Patents

Abstract

The invention relates to a howling suppression method based on feedback signal spectrum estimation. The method comprises the following steps: firstly, carrying out framing, windowing and Fourier transform on a signal collected by a microphone and a loudspeaker output signal respectively so as to obtain a corresponding frequency domain signal; then, according to the microphone frequency domain signal and the loudspeaker output frequency domain signal, calculating a cross-power spectrum between the microphone and the loudspeaker output signal and a power spectrum of the loudspeaker output signal; dividing the cross-power spectrum by the power spectrum so as to obtain estimation of a feedback path transfer function amplitude spectrum; using the obtained estimation of the feedback path transfer function spectrum amplitude to calculate a current frame feedback signal amplitude spectrum; and then using the microphone signal amplitude spectrum and the obtained feedback signal amplitude spectrum to calculate a gain function; finally, according to the obtained gain function, carrying out correction on the microphone frequency domain signal so as to realize howling suppression. By using the method of the invention, convergence and tracking speeds are fast, robustness is good and algorithm complexity is low, which is good for real-time realization.

Description

A kind of inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation

Technical field

Inhibition field, particularly a kind of inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation the present invention relates to utter long and high-pitched sounds.

Background technology

The document being suppressed at of uttering long and high-pitched sounds is also referred to as " acoustic feedback inhibition ", " acoustic feedback control ", " feedback is eliminated " etc., and what they were expressed is the same meaning.As shown in Figure 1, Fig. 1 is the generation schematic diagram of uttering long and high-pitched sounds.S (n) represents near-end speech, and this is to need amplifying signal; D (n) represents that the sound sending from loud speaker feeds back to microphone, the feedback signal of being picked up by microphone again through feedback path 102.D (n) is not the signal of expecting reception, it need to be curbed.Forward direction processing module 104 has been used for uttering long and high-pitched sounds and has suppressed and the function such as automatic gain control.Input signal s (n) is G (z)/(1-G (z) F (z)) to the transfer function between output signal y (n), if meet (a) simultaneously | G (z) F (z) | >=1; (b) ∠ G (z) F (z)=2 π n, n ∈ N; This system will become unstable, on the Frequency point satisfying condition, will utter long and high-pitched sounds.Using one of public address system object is exactly that the wearer of hearing aids also wishes sound to be amplified to desired level in order to improve system gain, the amount of gain of sound reinforcement system of having uttered long and high-pitched sounds down phenomenon limits.Therefore, must take certain measure to eliminate feedback and avoid the generation of uttering long and high-pitched sounds.Position due to teller in public sound reinforcement system may change frequently.For example: in classroom, teacher gives lessons in the public address system of use, microphone is placed on teacher's clothes and the position of teacher's health in the time giving lessons is not fixed, this just causes the continuous variation of feedback path, corresponding solution must be considered this point and can follow the tracks of fast this variation, otherwise just there will be and utter long and high-pitched sounds.

The inhibition of uttering long and high-pitched sounds is a popular research topic.Current solution mainly contains:

(1) shift frequency method.Shift frequency method by destruction utter long and high-pitched sounds produce phase condition reach utter long and high-pitched sounds suppress object.But the inhibition bad of uttering long and high-pitched sounds of this method, has usually controlled uttering long and high-pitched sounds of a Frequency point, system can produce and utter long and high-pitched sounds at other Frequency point again.And shift frequency method has noticeable impact to voice quality, research and real system are tested and are shown to use shift frequency method can obtain at most the gain of 6dB.

(2) trapper method.First trapper method finds the spectrum position that feedback may occur or occur, and then these excessively strong frequency contents in signal is attenuated to reach feedback inhibition effect.But in real system, likely there are multiple feedback frequency points.Therefore, trapper method effect is limited.The use trapper method of having reported can provide the gain of 3dB-8dB.

(3) sef-adapting filter method.The operation principle of sef-adapting filter method is again from the come in transmission characteristic of such a propagation ducts of microphone after utilizing filter to carry out simulated sound to send by loud speaker, make the feedback signal that collects with microphone from filter signal out consistent, then from the input signal of microphone, this part signal is cut to reach the object of eliminating feedback.As shown in Figure 2, Fig. 2 is the whistle inhibition system block diagram based on adaptive filter algorithm.The parameter of adaptive-filtering unit 206 must be adjusted to actual transmission paths characteristic and approach, and the signal that its simulation obtains just approaches the feedback signal of reality, just can obtain the inhibition of better uttering long and high-pitched sounds.The Method And Principle that this method adopts with echo cancelltion is the same, but different from echo cancelltion is under feedback occasion, system is always in being called as the state of " both-end intercommunication ", thereby affect the convergence rate of sef-adapting filter, and the filter coefficient estimating has partially, this method is difficult to be applicable to feedback path always in situation about constantly changing.

In addition, sef-adapting filter method is divided into discontinuous utter long and high-pitched sounds inhibition and the inhibition of uttering long and high-pitched sounds continuously.The discontinuous major defect suppressing of uttering long and high-pitched sounds is that this method need to be interrupted normal input speech signal, thereby the intelligibility of voice is affected, thereby is can not be received in a lot of occasions.Suppress to carry out estimated feedback signal with continuous input signal and utter long and high-pitched sounds continuously, but convergence rate is very slow.

In sum, the current inhibition method of uttering long and high-pitched sounds robustness is bad, all probably occurs of short duration uttering long and high-pitched sounds in the time that feedback path changes.

Summary of the invention

To the object of the invention is to, in order addressing the above problem, propose a kind of inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation, thereby it to be good to reach robustness, convergence and the fast object of tracking velocity.

For achieving the above object, the present invention proposes a kind of inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation, and the method concrete steps comprise:

Step 1): the signal x (n) and the speaker output signal y (n) that respectively microphone are gathered divide frame, windowing and Fourier transform to obtain microphone frequency domain signal X (i, k) with loud speaker output frequency-region signal Y (i, k);

Step 2): according to described step 1) crosspower spectrum φ between the microphone frequency domain signal X (i, k) and loud speaker output frequency-region signal Y (i, k) calculating microphone and the speaker output signal that obtain _yxthe power spectrum φ of (i, k) and speaker output signal _yy(i, k); And by crosspower spectrum φ _yx(i, k) is divided by power spectrum φ _yy(i, k) obtains the estimation of feedback path transfer function amplitude spectrum

Step 3): utilize described step 2) estimation of the feedback path transfer function amplitude spectrum that obtains calculate present frame feedback signal amplitude spectrum

Step 4): utilize microphone signal amplitude spectrum | X (i, k) | and described step 3) obtain feedback signal amplitude spectrum calculate and obtain gain function G (i, k);

Step 5): according to described step 4) obtain gain function G (i, k) come microphone frequency domain signal X (i, k) to revise to realize the inhibition of uttering long and high-pitched sounds.

The method also further comprises:

Step 6): to described step 5) the revised microphone frequency-region signal that obtains processes the final output signal of acquisition and to output signal carry out decorrelation processing; With

Step 7): to described step 6) process result carry out gain-adjusted.

Described step 2) concrete steps that obtain feedback path transfer function amplitude spectrum comprise:

21): adopt the level and smooth method of single order autoregression, obtain crosspower spectrum φ according to formula (1) _yx(i, k);

φ _yx(i，k)＝α _Cφ _yx(i-1，k)+(1-α _C)|Y ^*(i，k)X(i，k)| (1)

Wherein, α _csmoothing factor, 0≤α _c< 1;

22): adopt the level and smooth method of single order autoregression, obtain power spectrum φ according to formula (2) _yy(i, k);

φ _yy(i，k)＝α _Rφ _yy(i-1，k)+(1-α _R)Y ^*(i，k)Y(i，k) (2)

Wherein, α _rsmoothing factor, 0≤α _r< 1;

23): utilize crosspower spectrum φ _yx(i, k) and power spectrum φ _yy(i, k) obtains feedback path transfer function amplitude spectrum according to formula (3)

$\left|{\hat{F}}^{\&prime;}(i,k)\right|=\frac{{\mathrm{\&phi;}}_{\mathrm{yx}}(i,k)}{{\mathrm{\&phi;}}_{\mathrm{yy}}(i,k)+\mathrm{\&delta;}}---\left(3\right)$

Wherein, δ is that to avoid the denominator of formula (3) be 0 to a very little positive number.

Described step 3) calculate and obtain present frame feedback signal amplitude spectrum according to formula (4)

$\left|\hat{X}(i,k)\right|=\mathrm{\&beta;}\left|{\hat{F}}^{\&prime;}(i,k)\right|\left|X(i,k)\right|---\left(4\right)$

Wherein, β is an adjustable parameter, carries out this parameter of choose reasonable according to actual conditions.

Described step 4) method of calculated gains function comprises: based on priori signal feedback ratio method with based on check back signal feedback ratio method.

Described step 4) calculate and obtain gain function G (i, k) according to formula (5) based on check back signal feedback ratio method;

$G(i,k)={\left[\frac{\max ({\left|X(i,k)\right|}^{\mathrm{\&alpha;}}-{\left|\hat{X}(i,k)\right|}^{\mathrm{\&alpha;}},0)}{{\left|X(i,k)\right|}^{\mathrm{\&alpha;}}}\right]}^{\frac{1}{\mathrm{\&alpha;}}}---\left(5\right)$

Wherein, α > 0.For reducing computation complexity, α gets 1 or 2.

Described step 5) according to formula (6), microphone frequency domain signal X (i, k) is revised to realize the inhibition of uttering long and high-pitched sounds;

$\hat{S}(i,k)=X(i,k)G(i,k)---\left(6\right).$

Described step 7) gain of employing adjusting digital signal or the gain of adjusting analogue system power amplifier.The invention has the advantages that, compared with prior art, the present invention has good robustness; Initial convergence speed is fast and in the time that feedback path changes, can follow the tracks of rapidly this variation and can not utter long and high-pitched sounds; Because all operations of the present invention is all that frequency domain carries out, therefore there is very low computation complexity.

Brief description of the drawings

Fig. 1 is the generation schematic diagram of uttering long and high-pitched sounds;

Fig. 2 is traditional whistle inhibition system block diagram based on adaptive filter algorithm;

Fig. 3 is feedback path estimating system block diagram of the present invention;

Fig. 4 is the inhibition method of the uttering long and high-pitched sounds structured flowchart based on feedback signal spectrum estimation of the present invention;

Fig. 5 is the block diagram of feedback spectrum estimation module 410 of the present invention;

Fig. 6 is a feedback path impulse response figure of actual measurement;

Fig. 7 is the amplitude spectrum of actual feedback path transfer function and the amplitude spectrum comparison diagram of the feedback path transfer function that the present invention estimates.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention will be described in detail.

The object of the invention is to have proposed a kind of robustness good, convergence and tracking velocity are fast, when effective inhibition is uttered long and high-pitched sounds, ensure the inhibition method of uttering long and high-pitched sounds that voice quality is not suffered a loss.

In order to achieve the above object, the technical scheme that the present invention takes is as follows:

Based on the inhibition method of uttering long and high-pitched sounds of feedback signal spectrum estimation, concrete steps comprise:

Step 1): respectively the microphone signal collecting and speaker output signal are carried out to windowing process and Fourier transform, time-domain signal is transformed to frequency domain;

Step 2): utilize the crosspower spectrum of level and smooth microphone signal and speaker output signal to obtain the amplitude spectrum of feedback path divided by the power spectrum of level and smooth speaker output signal, and utilize the amplitude spectrum of feedback path to calculate the amplitude spectrum of present frame feedback signal;

Step 3): utilize microphone signal amplitude spectrum and feedback signal amplitude spectrum calculated gains function;

Step 4): microphone signal frequency spectrum is revised;

Step 5): revised microphone signal frequency spectrum is carried out to inverse Fourier transform and obtain time domain output;

Step 6): optional) to step 5) and output apply a de-correlation block;

Step 7): optional) to step 6) and output apply a gain.

In technique scheme, further, step 1) described in adopt the pattern of frame processing to carry out signal processing, all processing and program are all unit based on frame.Need to carry out windowing process to the frame microphone signal receiving and the output signal of loud speaker, then use FFT to transform to frequency domain.

In technique scheme, further, step 2) described in first need to estimate the amplitude spectrum of feedback path transfer function, and utilize it further to calculate the amplitude spectrum of present frame feedback signal.The amplitude spectrum of feedback path transfer function utilizes the crosspower spectrum absolute value of level and smooth microphone signal and speaker output signal to obtain divided by the power spectrum of level and smooth speaker output signal.

In above-mentioned technical scheme, further, step 3) described gain function calculates based on check back signal feedback ratio.

In above-mentioned technical scheme, further, step 4) the frequency spectrum correction of microphone is only referred to its amplitude spectrum is revised, and keep its phase spectrum constant.

In above-mentioned technical scheme, further, step 5) be revised microphone frequency spectrum to be carried out to inverse Fourier transform obtain time-domain signal, then carry out overlapping stack as the output signal of uttering long and high-pitched sounds and suppressing with the output signal of former frame.

In above-mentioned technical scheme, further, step 6) correlation removing between the near end signal that speaker output signal and microphone receive for described de-correlation block.

In above-mentioned technical scheme, further, step 7) described gain both can be added in the power amplifier end that also can be added in analogue system in the digital signal of digital system.

As shown in Figure 4, Fig. 4 is the inhibition method of the uttering long and high-pitched sounds structured flowchart based on feedback signal spectrum estimation of the present invention.Wherein, s (n) represents near end signal, and d (n) represents feedback signal, and x (n) represents the signal that microphone collects, and y (n) represents the signal of loud speaker output.

Step 1): respectively the microphone signal collecting and speaker output signal are carried out to windowing process and Fourier transform, time-domain signal is transformed to frequency domain.

The microphone signal x (n) collecting is carried out after windowing module 402 and Fourier transform module 404, time-domain signal being transformed to frequency domain after serial to parallel conversion; Speaker output signal y (n) is carried out after windowing module 402 and Fourier transform module 404, time-domain signal being transformed to frequency domain after serial to parallel conversion.The i frame discrete Fourier transform of y (n) is designated as Y (i, k), and the i frame discrete Fourier transform of x (n) is designated as X (i, k).

Step 2): utilize the crosspower spectrum of level and smooth microphone signal x (n) and speaker output signal y (n) to obtain the amplitude spectrum of feedback path divided by the power spectrum of level and smooth speaker output signal y (n), and utilize the amplitude spectrum of feedback path to calculate the amplitude spectrum of present frame feedback signal.

As shown in Figure 3, Fig. 3 is feedback path estimating system block diagram of the present invention.The signal that microphone picks up is x (n)=d (n)+s (n), and d (n) is that speaker output signal y (n) obtains by feedback path transfer function F (z) filtering.We are objective definition function E{e ²(n) }=E{|x (n)-y (n) * f (n) | ², * represents convolution here, f (n) represents the time-domain pulse response of F (z).Minimize this target function can obtain time domain Wei Na separate form as follows:

$\hat{f}\left(n\right)=R_{\mathrm{yy}}^{-1}{r}_{\mathrm{yx}}---\left(1\right)$

Wherein, R _yyrepresent the autocorrelation matrix of y (n), r _yxrepresent the cross-correlation matrix of y (n) and x (n).(1) is transformed to frequency domain to be obtained

$F^{\&prime;}(i,k)=\frac{P_{\mathrm{yx}}(i,k)}{P_{\mathrm{yy}}(i,k)}---\left(2\right)$

Here P _yx(i, k) represents the crosspower spectrum of x (n) and y (n), P _yy(i, k) represents the power spectrum of y (n).If the delay long enough between input signal s (n) and output signal y (n), just can well carry out decorrelation to the such stationary signal in short-term of voice, can think that the correlation between s (n) and y (n) has become very weak, P so _ys(i, k) ≈ 0, so P _yx(i, k)=P _ys(i, k)+P _yd(i, k) ≈ P _yd(i, k), thus (2) formula of utilization can effectivelyly estimate the frequency spectrum of feedback path transfer function.The present invention actual utilize be the amplitude spectrum of F ' (i, k) | F ' (i, k) |.As shown in Figure 5, Fig. 5 is the block diagram of feedback spectrum estimation module 410 of the present invention.

Crosspower spectrum smooth unit 502 is calculated the crosspower spectrum absolute value of x (n) and y (n), and carries out smoothly,

φ _yx(i，k)＝α _Cφ _yx(i-1，k)+(1-α _C)|Y ^*(i，k)X(i，k)| (3)

Here use * to represent complex conjugate.We have adopted signal period figure have been carried out to autoregression smoothly as the estimation of power spectrum or crosspower spectrum, and this is mainly because the method is simple and easily realization, but this patent is not limited to the method, and the method for other any estimated power spectrums can be used.

Power spectrum smooth unit 504 is calculated the power spectrum of y (n) and is carried out smoothly,

φ _yy(i，k)＝α _Rφ _yy(i-1，k)+(1-α _R)Y ^*(i，k)Y(i，k) (4)

Here α _cand α _rbe smoothing factor, its value meets 0≤α _c< 1 and 0≤α _r< 1.

Feedback transfer function estimation unit 506 calculates feedback path transfer function amplitude spectrum, is expressed as:

$\left|{\hat{F}}^{\&prime;}(i,k)\right|=\frac{{\mathrm{\&phi;}}_{\mathrm{yx}}(i,k)}{{\mathrm{\&phi;}}_{\mathrm{yy}}(i,k)+\mathrm{\&delta;}}---\left(5\right)$

δ is a positive number that absolute value is very little in formula (5), and being used for the divisor of the formula that prevents (5) is 0.

As shown in Figure 6, Fig. 6 is the feedback path impulse response figure of an actual measurement.As shown in Figure 7, Fig. 7 is the amplitude spectrum of actual feedback path transfer function and the amplitude spectrum comparison diagram of the feedback path transfer function that the present invention estimates.Can find out the well amplitude spectrum of the feedback path transfer function of approaching to reality of amplitude spectrum of the feedback path transfer function that the present invention estimates.

Present frame feedback signal amplitude spectrum computing unit 508 is used for estimating the spectrum component of present frame feedback signal, is expressed as

$\left|\hat{X}(i,k)\right|=\mathrm{\&beta;}\left|{\hat{F}}^{\&prime;}(i,k)\right|\left|Y(i,k)\right|---\left(6\right)$

In formula (6), β is a parameter that user is adjustable, if the amplitude spectrum of feedback signal crosses has been estimated, and needs to make β < 1, if otherwise the amplitude spectrum of feedback signal owed to have estimated and needed to make β > 1.Select large compensating factor just can obtain larger amount of gain but relative meeting near-end speech quality is had to certain influence; Select the voice quality that little compensating factor can obtain but may reduce the amount of gain of sound reinforcement system, should reasonably accept or reject according to actual conditions.

In sum, we have just estimated the amplitude spectrum of feedback signal through type (6), have completed the function of feedback signal spectrum estimation module 410.

Step 3): utilize microphone signal amplitude spectrum and feedback signal amplitude spectrum calculated gains function;

The gain function of gain calculation module 412 can utilize spectrum-subtraction to obtain, and its expression formula is:

$G(i,k)={\left[\frac{\max ({\left|X(i,k)\right|}^{\mathrm{\&alpha;}}-{\left|\hat{X}(i,k)\right|}^{\mathrm{\&alpha;}},0)}{{\left|X(i,k)\right|}^{\mathrm{\&alpha;}}}\right]}^{\frac{1}{\mathrm{\&alpha;}}}---\left(7\right)$

In above formula, α represents a factor of spectrum-subtraction, α=1st, and based on amplitude spectrum subtraction, α=2nd, power spectrum subtraction is Wiener Filter Method, α also can select other numerical value, α > 0.In reality, generally select α=1 or α=2 in order to reduce computation complexity.

Although the selection of gain function also has other form, as adopted gain function ([1] the Y.Ephraim and D.Malah based on priori signal to noise ratio in document [1], " Speech enhancement using a minimum mean squareerror short-time spectral amplitude estimator, " IEEE Trans.on Acoust., Speech, SignalProcessing, vol.ASSP-32, pp.1109-1121, Dec.1984), we find to use posterior signal feedback ratio can obtain better result for whistle inhibition system by experiment, thereby recommendation formula of the present invention (7) is carried out calculated gains function.

Step 4) microphone signal frequency spectrum is revised.

The input/output relation of frequency spectrum correcting module 414 is:

$\hat{S}(i,k)=X(i,k)G(i,k)---\left(8\right)$

represent to have eliminated the microphone signal frequency spectrum after feedback signal.

Step 5) microphone signal frequency spectrum is carried out to inverse Fourier transform exported.

Inverse Fourier transform module 416 is to present frame carry out inverse Fourier transform and obtain present frame output, then carry out overlapping stack with former frame output and obtain final output signal

Step 6) optional) to step 5) and output apply a de-correlation block 418.

As step 2) as described in delay between input signal s (n) and output signal y (n) be the frame length of 3 times, corresponding 50% overlapping.This delay generally can be carried out decorrelation to the such stationary signal in short-term of voice.Because voice signal also has correlation when long, i.e. the time delay of 3 times of frame lengths sometimes is still not enough to remove completely its correlation, and this may cause the distortion of signal.For need to increasing a de-correlation block, further raising voice quality removes the correlation between s (n) and y (n), a lot of classical decorrelating methods are suggested, can be referring to document [2] (M.Ali, " Stereophonic acoustic echo cancellation system using time-varying all-pass filteringfor signal decorrelation, " in Proceedings of the ICASSP ' 98, vol.6, pp.3689-3692, Seattle, USA, May 1998) and document [3] ([3] J.Benesty, D.R.Morgan, and M.M.Sondhi, " A betterunderstanding and an improved solution to the specific problems of stereophonic acoustic echo cancellation, " IEEE Trans.Speech Audio Processing, vol.6, pp.156-165, Mar.1998).It should be noted that, choosing of de-correlation must be not damage voice quality as prerequisite.

Step 7) optional) to step 6) and output increase a gain module 420.

System gain can regulate by two kinds of modes, and the one, regulate the gain of digital signal, right in gain module 420 apply gain g, also can be by regulating analogue system to realize as the gain of power amplifier.

Should be noted that, the described in the invention inhibition method of uttering long and high-pitched sounds can realize with various ways, for example combination of hardware, software or hardware and software.Hardware platform can be FPGA, PLD or other application-specific integrated circuit ASICs.Software platform comprises DSP, ARM or other microprocessors.For example part of module of the combination of software and hardware realizes with dsp software, and part of module realizes with hardware accelerator as FFT.

It should be noted last that, above embodiment is only unrestricted in order to technical scheme of the present invention to be described.Although the present invention is had been described in detail with reference to embodiment, those of ordinary skill in the art is to be understood that, technical scheme of the present invention is modified or is equal to replacement, do not depart from the spirit and scope of technical solution of the present invention, it all should be encompassed in the middle of claim scope of the present invention.

Claims (8)

1. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation, the method concrete steps comprise:

Step 1): the signal x (n) and the speaker output signal y (n) that respectively microphone are gathered divide frame, windowing and Fourier transform to obtain microphone frequency domain signal X (i, k) with loud speaker output frequency-region signal Y (i, k);

Step 2): according to described step 1) crosspower spectrum φ between the microphone frequency domain signal X (i, k) and loud speaker output frequency-region signal Y (i, k) calculating microphone and the speaker output signal that obtain _yxthe power spectrum φ of (i, k) and speaker output signal _yy(i, k); And utilize crosspower spectrum φ _yx(i, k) and power spectrum φ _yy(i, k) calculates the estimation of feedback path transfer function amplitude spectrum its computing formula is expressed as:

$\left|{\hat{F}}^{\&prime;}(i,k)\right|=\frac{{\mathrm{\&phi;}}_{\mathrm{yx}}(i,k)}{{\mathrm{\&phi;}}_{\mathrm{yy}}(i,k)+\mathrm{\&delta;}}$

In above formula, δ is a positive number that absolute value is very little;

Step 3): utilize described step 2) estimation of the feedback path transfer function amplitude spectrum that obtains calculate present frame feedback signal amplitude spectrum its computing formula is expressed as:

$\left|\hat{X}(i,k)\right|=\mathrm{\&beta;}\left|{\hat{F}}^{\&prime;}(i,k)\right|\left|Y(i,k)\right|$

In above formula, β is a parameter that user is adjustable, if the amplitude spectrum of feedback signal crosses has been estimated, and needs to make β < 1, if otherwise the amplitude spectrum of feedback signal owed to have estimated and needed to make β > 1;

Step 4): utilize microphone signal amplitude spectrum X (i, k) and described step 3) obtain feedback signal amplitude spectrum calculate and obtain gain function G (i, k), its computing formula is expressed as:

$G(i,k)={\left[\frac{\max \left(\right|X(i,k){|}^{\mathrm{\&alpha;}}-|\hat{X}(i,k){|}^{\mathrm{\&alpha;}},0)}{{\left|X(i,k)\right|}^{\mathrm{\&alpha;}}}\right]}^{\frac{1}{\mathrm{\&alpha;}}}$

In above formula, α represents a factor of spectrum-subtraction, α=1st, and based on amplitude spectrum subtraction, α=2nd, power spectrum subtraction is Wiener Filter Method, α also can select other numerical value, α > 0;

Step 5): according to described step 4) obtain gain function G (i, k) come microphone frequency domain signal X (i, k) to revise to realize the inhibition of uttering long and high-pitched sounds, the formula table of this correction is shown:

$\hat{S}(i,k)=X(i,k)G(i,k)$

In above formula, represent to have eliminated the microphone signal frequency spectrum after feedback signal.

2. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 1, is characterized in that, the method also further comprises:

Step 6): to described step 5) the revised microphone frequency-region signal that obtains processes the final output signal of acquisition and to output signal carry out decorrelation processing; With

Step 7): to described step 6) process result carry out gain-adjusted.

3. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 1, is characterized in that described step 2) concrete steps that obtain feedback path transfer function amplitude spectrum comprise:

21): adopt the level and smooth method of single order autoregression, obtain crosspower spectrum φ according to formula (1) _yx(i, k);

φ _yx(i,k)＝α _Cφ _yx(i-1,k)+(1-α _C)|Y ^*(i,k)X(i,k)| (1)

Wherein, α _csmoothing factor, 0≤α _c< 1;

22): adopt the level and smooth method of single order autoregression, obtain power spectrum φ according to formula (2) _yy(i, k);

φ _yy(i,k)＝α _Rφ _yy(i-1,k)+(1-α _R)Y ^*(i,k)Y(i,k) (2)

Wherein, α _rsmoothing factor, 0≤α _r< 1;

23): utilize crosspower spectrum φ _yx(i, k) and power spectrum φ _yy(i, k) obtains feedback path transfer function amplitude spectrum according to formula (3) $\left|{\hat{F}}^{\&prime;}(i,k)\right|;$

$\left|{\hat{F}}^{\&prime;}(i,k)\right|=\frac{{\mathrm{\&phi;}}_{\mathrm{yx}}(i,k)}{{\mathrm{\&phi;}}_{\mathrm{yy}}(i,k)+\mathrm{\&delta;}}---\left(3\right).$

4. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 1, is characterized in that described step 3) calculate and obtain present frame feedback signal amplitude spectrum according to formula (4)

$\left|\hat{X}(i,k)\right|=\mathrm{\&beta;}\left|{\hat{F}}^{\&prime;}(i,k)\right|\left|X(i,k)\right|---\left(4\right)$

Wherein, β is an adjustable parameter, carries out this parameter of choose reasonable according to actual conditions.

5. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 1, is characterized in that described step 4) method of calculated gains function comprises: based on priori signal feedback ratio method with based on check back signal feedback ratio method.

6. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 5, is characterized in that described step 4) calculate and obtain gain function G (i, k) according to formula (5) based on check back signal feedback ratio method;

$G(i,k)={\left[\frac{\max \left(\right|X(i,k){|}^{\mathrm{\&alpha;}}-|\hat{X}(i,k){|}^{\mathrm{\&alpha;}},0)}{{\left|X(i,k)\right|}^{\mathrm{\&alpha;}}}\right]}^{\frac{1}{\mathrm{\&alpha;}}}---\left(5\right)$

Wherein, α > 0.

7. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 1, is characterized in that described step 5) according to formula (6), microphone frequency domain signal X (i, k) is revised to realize the inhibition of uttering long and high-pitched sounds:

$\hat{S}(i,k)=X(i,k)G(i,k)---\left(6\right).$

8. the inhibition method of uttering long and high-pitched sounds based on feedback signal spectrum estimation according to claim 2, is characterized in that described step 7) gain of employing adjusting digital signal or the gain of adjusting analogue system power amplifier.