CN103295582A - Noise suppression method and system - Google Patents

Abstract

The invention relates to the field of voice signal processing, and discloses a noise suppression method and system. According to the noise suppression method, the prior signal to noise ratio and the posterior signal to noise ratio of a noisy voice signal are calculated, the estimated value of a clean voice signal in the short-term spectrum amplitude is calculated according to a maximum posterior MAP standard, the ratio of the estimated value and the noisy voice signal is calculated, an attenuation factor is obtained, and then inverse fast Fourier transform IFFT treatment is conducted on the noisy voice signal according to the attenuation factor obtained through calculation, so that calculated amount is greatly reduced during noise suppression. The noise suppression method and system is more suitable for practical application, and performance is also improved. Furthermore, when the attenuation factor is calculated, the calculation of the ratio of the first order and zero order of a first kind of the modified Bessel function is involved, and at the moment, the ratio is obtained through the table look-up method so that storage space can be further reduced; or the approximate calculation method is adopted, no storage table is required, and calculation is more simplified.

Description

Noise suppressing method and system thereof

Technical field

The present invention relates to field of voice signal, particularly noise suppressing method and system thereof.

Background technology

Single Mike's voice strengthen the disposal route that (being squelch) adopts the frequency domain decay usually, comprise that Fourier transform (FFT) analysis, noise estimate, calculate modules such as elder generation/posteriority signal to noise ratio (S/N ratio), calculating decay factor, IFFT be comprehensive, and roughly flow process as shown in Figure 1.

The Noisy Speech Signal Y (w) that Mike collects can be expressed as Y (w)=X (w)+N (w), and wherein X (w) is the clean speech signal, and N (w) is acoustic noise signal; Use X, N, Y represent the short-term spectrum amplitude (Short-Time Spectral Amplitude is called for short " STSA ") of corresponding signal, i.e. X=|X (w) respectively |, N=|N (w) |, Y=|Y (w) |.Noisy Speech Signal is transformed into frequency domain through fft analysis from time domain earlier; The noise estimation module then is to estimate the variance λ of current noise as far as possible exactly _N=E{|N (w) | ²; Then calculate priori signal to noise ratio (S/N ratio) ξ and posteriority signal to noise ratio (S/N ratio) γ; Calculate decay factor then, carry out IFFT at last and comprehensively obtain voice signal through squelch.Wherein, priori signal to noise ratio (S/N ratio) ξ is the variance ratio of voice X (w) and noise N (w)

Posteriority signal to noise ratio (S/N ratio) γ is the ratio of the current signals with noise spectrum energy Y (w) that receives and noise N (w) variance $\mathrm{\γ}=\frac{{\left|Y\left(w\right)\right|}^2}{E\left\{{\left|N\left(w\right)\right|}^2\right\}}=\frac{Y^2}{{\mathrm{\λ}}_N}.$

Decay factor G is the clean speech signal estimated shared ratio in Noisy Speech Signal, like this and the Noisy Speech Signal of importing multiply each other and namely obtain result, decay factor is determining the performance of algorithm.Its computing method are core places of whole noise suppression algorithm, the final effect of algorithm will directly be influenced, least mean-square error commonly used (Minimum Mean Square Error is called for short " MMSE ") and least mean-square error (log-MMSE) the inhibiting factor algorithm of taking the logarithm calculate.

Adopting MMSE to calculate decay factor can be represented by the formula:

$G_{\mathrm{MMSE}}=\frac{\sqrt{\mathrm{\π}}}{2}\frac{\sqrt{\mathrm{\υ}}}{\mathrm{\γ}}\exp (-\frac{\mathrm{\υ}}{2})[(1+\mathrm{\υ})I_1\left(\frac{\mathrm{\υ}}{2}\right)+{\mathrm{\υI}}_0\left(\frac{\mathrm{\υ}}{2}\right)]---\left(1\right)$

Adopting Log-MMSE to calculate decay factor can be represented by the formula:

$G_{\mathrm{Log}\_\mathrm{MMSE}}=\frac{\mathrm{\ξ}}{1+\mathrm{\ξ}}\exp \left\{\frac{1}{2}{\∫}_{\mathrm{\υ}}^{\∞}\frac{e^{-t}}{t}\mathrm{dt}\right\}---\left(2\right)$

In formula 1 and the formula 2, exp{} represents exponential function, I _vExpression v rank first kind modified Bessel function,

By as can be seen top, the calculated amount of MMSE and Log-MMSE all is very big.The concrete analysis of its calculated amount is as follows:

1. for adopting MMSE to calculate decay factor, need at least 1 evolution, 2 divisions, 1 exp exponent arithmetic, several times multiplication and additive operation, and calculate I ₁And I ₀Wherein calculate I ₁And I ₀Usually adopt the method for tabling look-up, need be respectively them storage space is provided.In addition, because the domain of walker of Bessel's function is very big, cause in fixed-point algorithm, realizing very difficult.Be illustrated in figure 2 as first-order bessel function I ₁(x) and zero Bessel function I ₀(x) the numerical simulation figure in [20,20] dB scope, its numerical range is [10 ^-3, 10 ⁴⁴].

2. for adopting Log-MMSE to calculate decay factor, direct integral can't calculate, in the mathematical software matrix experiment chamber (Matrix Laboratory, be called for short " Matlab ") a Function e xpint is arranged, can be used for the integral part of exp index inside in the calculating formula 2:

$\mathrm{expint}\left(\mathrm{\υ}\right)={\∫}_{\mathrm{\υ}}^{\∞}\frac{e^{-t}}{t}\mathrm{dt}---\left(3\right)$

Approximate formula commonly used in terminal is:

$\mathrm{expint}\left(\mathrm{\&upsi;}\right)=\left\{\begin{array}{cc}-2.31{\log }_{10}\left(\mathrm{\&upsi;}\right)-0.6& \mathrm{\&upsi;}<0.1\\ -1.544{\log }_{10}\left(\mathrm{\&upsi;}\right)+0.166& 0.1\&le;\mathrm{\&upsi;}\&le;1\\ {10}^{-(0.52\mathrm{\&upsi;}+0.26)}& \mathrm{\&upsi;}>1\end{array}\right.---\left(4\right)$

Approximation method in calculating in the general employing formula 4, but total calculated amount is also very big, comprises exponent arithmetic, the log logarithm operation of exp and 10.

Summary of the invention

The object of the present invention is to provide a kind of noise suppressing method and system thereof, make in squelch, to have significantly reduced calculated amount, be more suitable for practical application, and on performance, also promote to some extent.

For solving the problems of the technologies described above, embodiments of the present invention provide a kind of noise suppressing method, comprise following steps:

After the priori signal to noise ratio (S/N ratio) that calculates Noisy Speech Signal and posteriority signal to noise ratio (S/N ratio), ask for the estimated value of clean speech signal and the ratio of Noisy Speech Signal, obtain decay factor G; Wherein, described estimated value is the estimated value according to the clean speech signal short-term spectrum amplitude of maximum a posteriori MAP criterion calculating;

According to described G described Noisy Speech Signal being carried out contrary fast fourier transform IFFT handles.

Embodiments of the present invention also provide a kind of noise suppressing system, comprise: snr computation module, decay factor are asked for module, IFFT processing module;

Described snr computation module is used for calculating priori signal to noise ratio (S/N ratio) and the posteriority signal to noise ratio (S/N ratio) of Noisy Speech Signal, and the described priori signal to noise ratio (S/N ratio) that will calculate and posteriority signal to noise ratio (S/N ratio) are exported to described decay factor and asked for module;

Described decay factor is asked for module and is used for when receiving described priori signal to noise ratio (S/N ratio) and posteriority signal to noise ratio (S/N ratio),, ask for the estimated value of clean speech signal and the ratio of Noisy Speech Signal, obtain decay factor G; Wherein, described estimated value is the estimated value according to the clean speech signal short-term spectrum amplitude of maximum a posteriori MAP criterion calculating;

Described IFFT processing module is used for asking for the G that module obtains according to described decay factor, described Noisy Speech Signal is carried out contrary fast fourier transform IFFT handle.

Embodiment of the present invention in terms of existing technologies, when calculating decay factor, the estimated value of the clean speech signal short-term spectrum amplitude of calculating with maximum a posteriori MAP criterion, the estimated value of required clean speech signal short-term spectrum amplitude when calculating decay factor.Make in squelch, to have significantly reduced calculated amount, be more suitable for practical application, and on performance, also promote to some extent.

Preferably, ask for described ratio according to following formula, obtain decay factor G:

$G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$

Wherein, ξ represents described priori signal to noise ratio (S/N ratio); γ represents described posteriority signal to noise ratio (S/N ratio); ζ represents the variable of Bessel's function; r _I(ζ) ratio of expression first kind modified Bessel function single order and zeroth order.As can be seen, calculate major part that described G adopts be operating as evolution, division, square, simple operation such as addition, do not need exponent arithmetic and power operation, and ratio r _INumerical range (ζ) is [0,1], is convenient to very much the fixed point computing.

Preferably, store all possible ζ and corresponding r in advance _IMapping table (ζ); Ask for described ratio, when obtaining decay factor G, calculating described ζ earlier, in described mapping table, searching the r corresponding with this ζ according to described ζ then _I(ζ); At last with described ξ, described γ and the described r that finds _I(ζ), the described formula of substitution obtains described G.Because that storage in advance is all possible ζ and corresponding r _IMapping table (ζ) needs to store all values of first kind modified Bessel function single order and zeroth order in the prior art, and the size of storage list can be reduced half, i.e. space storage data reduce half.

Preferably, asking for described ratio, when obtaining decay factor G, calculating described ζ earlier, judging that then whether described ζ is greater than 1; If described ζ is greater than 1, then according to formula

Calculate described r _I(ζ); If described ζ is less than or equal to 1, then according to formula r _I(ζ)=0.5 ζ calculates described r _I(ζ); At last with the described r of described ξ, described γ and calculating _I(ζ) the described formula of substitution $G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$ In, obtain described G.As seen r _ICalculating (ζ) is very simple, and does not need storage list, further saves storage space.

Preferably, when calculating described ζ, with the product of described ξ and described γ, open radical sign and multiply by the numerical value that obtains after 2 as described ζ.Simple evolution, multiplication can calculate, and realizes simple.

Description of drawings

Fig. 1 is the process flow diagram of noise suppressing method in the prior art;

Fig. 2 is first-order bessel function I ₁(x) and zero Bessel function I ₀(x) the numerical simulation figure in [20,20] dB scope;

Fig. 3 is the process flow diagram of first embodiment of the invention noise suppressing method;

Fig. 4 is the corresponding relation figure of the ratio of the independent variable of Bessel's function and first kind modified Bessel function single order and zeroth order;

Fig. 5 is the process flow diagram of noise suppressing method second embodiment of the invention;

Fig. 6 is the process flow diagram according to the 3rd embodiment noise suppressing method of the present invention;

Fig. 7 is the approximate treatment value of the ratio of first kind modified Bessel function single order and zeroth order in the 3rd embodiment of the present invention and the comparison diagram of actual value;

Fig. 8 adopts MAP, MMSE and log-MMSE criterion to calculate the objective performance comparison diagram that decay factor is carried out the voice signal of squelch;

Fig. 9 does not carry out the voice signal of squelch and adopts MAP, MMSE and the log-MMSE criterion is calculated the subjective performance comparison diagram that decay factor is carried out the voice signal of squelch;

Figure 10 is the structural drawing according to the 5th embodiment noise suppressing system of the present invention;

Figure 11 is the structural drawing according to the 6th embodiment noise suppressing system of the present invention.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing the embodiments of the present invention are explained in detail.Yet, persons of ordinary skill in the art may appreciate that in each embodiment of the present invention, in order to make the reader understand the application better many ins and outs have been proposed.But, even without these ins and outs with based on many variations and the modification of following each embodiment, also can realize each claim of the application technical scheme required for protection.

First embodiment of the present invention relates to a kind of noise suppressing method.Idiographic flow comprises following steps as shown in Figure 3:

Step 301 is carried out fast fourier transform FFT to Noisy Speech Signal.

Step 302 is carried out noise to the Noisy Speech Signal after the FFT conversion and is estimated.

Step 303, priori signal to noise ratio (S/N ratio) and the posteriority signal to noise ratio (S/N ratio) of calculating Noisy Speech Signal.

Step 301 is same as the prior art to step 303, does not repeat them here.

In step 304, ask for the estimated value of clean speech signal and the ratio of Noisy Speech Signal, obtain decay factor G; Wherein, the estimated value of clean speech signal is the estimated value according to the clean speech signal short-term spectrum amplitude of maximum a posteriori MAP criterion calculating.That is to say, with the estimated value of the clean speech signal short-term spectrum amplitude that calculates according to maximum a posteriori MAP criterion, required clean speech signal estimated value when asking for decay factor G.

Specifically, in the present embodiment, obtain decay factor G according to following formula:

$G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}---\left(5\right)$

Wherein, ξ represents the priori signal to noise ratio (S/N ratio); γ represents the posteriority signal to noise ratio (S/N ratio); ζ represents the variable of Bessel's function; r _I(ζ) ratio of expression first kind modified Bessel function single order and zeroth order.Derivation to above-mentioned formula (5) describes below:

According to multiple Gauss model, the distribution function that can obtain the short-term spectrum amplitude X of clean speech signal is

$p\left(X\right)=\left\{\begin{array}{cc}\frac{2X}{{\mathrm{\&lambda;}}_X}\exp (-\frac{X^2}{{\mathrm{\&lambda;}}_X})& X\&GreaterEqual;0\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(6\right)$

In the formula, λ _XVariance for X.

And the short-term spectrum amplitude of clean speech signal is under the condition of X, and Noisy Speech Signal is that the distribution function of Y (W) is

$p\left(Y\left(w\right)\right|X)=\frac{1}{{\mathrm{\&pi;\&lambda;}}_N}\exp (-\frac{Y^2+X^2}{{\mathrm{\&lambda;}}_N})I_0\frac{2\mathrm{XY}}{{\mathrm{\&lambda;}}_N}---\left(7\right)$

In the formula, Y is the short-term spectrum amplitude of Noisy Speech Signal, λ _NBe the variance of noise, I ₀It is zeroth order first kind modified Bessel function.

According to maximum a posteriori MAP criterion, the estimated value of calculating the short-term spectrum amplitude X of clean speech signal is

$\hat{X}=\underset{X}{\arg \max }p\left(X\right|Y\left(w\right))=\underset{X}{\arg \max }\frac{p\left(Y\left(w\right)\right|X)p\left(X\right)}{p\left(Y\left(w\right)\right)}---\left(8\right)$

Because log is monotonic quantity, therefore, the maximal value of calculating posterior probability p (X|Y (w)) can be by calculating its logarithm log{p (X|Y (w)) } maximal value simplify calculating.

Definition:

J＝log{p(X|Y(w))}＝log{p(Y(w)|X)}+log{p(X)}-log{p(Y(w))} (9)

Distribution function 6 and the above-mentioned formula 9 of 7 substitutions can be got:

$J=-\frac{Y^2+X^2}{{\mathrm{\&lambda;}}_N}-\frac{X^2}{{\mathrm{\&lambda;}}_N}+\log X+\log \left[I_0\left(\frac{2\mathrm{YX}}{{\mathrm{\&lambda;}}_N}\right)\right]+\mathrm{cons}\tan t---\left(10\right)$

In the formula, constant is constant.

Make the independent variable of Bessel's function be

Ask the local derviation of the X of J, and utilize the characteristic of first kind correction Bezier (Bessel) function:

$\frac{\&PartialD;}{\&PartialD;\mathrm{\&zeta;}}{I}_n\left(\mathrm{\&zeta;}\right)=\frac{n}{\mathrm{\&zeta;}}{I}_n\left(\mathrm{\&zeta;}\right)+I_{n+1}\left(\mathrm{\&zeta;}\right)$ n＝0，1，2.... (12)

Obtain at last:

$\frac{\&PartialD;J}{\&PartialD;X}=-\frac{2X}{{\mathrm{\&lambda;}}_X}-\frac{2X}{{\mathrm{\&lambda;}}_N}+\frac{1}{X}+\frac{2Y}{{\mathrm{\&lambda;}}_N}\frac{I_1\left(\mathrm{\&zeta;}\right)}{I_0\left(\mathrm{\&zeta;}\right)}---\left(13\right)$

Make formula 13 equal zero, obtain about the super quadratic equation of X as follows through distortion:

$0=X^2-\frac{{\mathrm{Y\&lambda;}}_X}{({\mathrm{\&lambda;}}_X+{\mathrm{\&lambda;}}_N)}\frac{I_1\left(\mathrm{\&zeta;}\right)}{I_0\left(\mathrm{\&zeta;}\right)}X-\frac{1}{2}\frac{{\mathrm{\&lambda;}}_X{\mathrm{\&lambda;}}_N}{{\mathrm{\&lambda;}}_X+{\mathrm{\&lambda;}}_N}---\left(14\right)$

Because the variable of Bessel's function

Comprise unknown variable X, therefore above-mentioned super quadratic equation does not have exact solution.In the present embodiment, the variable in the hypothesis Bessel's function is known earlier, and on this basis, according to the root formula of quadratic equation, two solutions that obtain formula 14 are

${\mathrm{<figure-callout\; id="1"\; label="X"\; filenames="HDA0000140307820000041.png,HDA0000140307820000051.png"\; state="\{\{state\}\}">X</figure-callout>}}_{\mathrm{1,2}}=\frac{{\mathrm{Y\&lambda;}}_X}{2({\mathrm{\&lambda;}}_X+{\mathrm{\&lambda;}}_N)}\frac{I_1\left(\mathrm{\&zeta;}\right)}{I_0\left(\mathrm{\&zeta;}\right)}\&PlusMinus;\sqrt{{\left(\frac{{\mathrm{Y\&lambda;}}_X}{2({\mathrm{\&lambda;}}_X+{\mathrm{\&lambda;}}_N)}\frac{I_1\left(\mathrm{\&zeta;}\right)}{I_0\left(\mathrm{\&zeta;}\right)}\right)}^2+\frac{{\mathrm{\&lambda;}}_X{\mathrm{\&lambda;}}_N}{2({\mathrm{\&lambda;}}_X+{\mathrm{\&lambda;}}_N)}}---\left(15\right)$

With the solution that obtains, divided by the amplitude Y of signals with noise:

$G=\frac{{\mathrm{<figure-callout\; id="1"\; label="X"\; filenames="HDA0000140307820000041.png,HDA0000140307820000051.png"\; state="\{\{state\}\}">X</figure-callout>}}_{\mathrm{1,2}}}{Y}---\left(16\right)$

Remove impossible negative value solution, the expression that finally obtains G is:

$G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$

In the formula

Ratio for first kind modified Bessel function single order and zeroth order.

After asking for decay factor G, enter step 105, carry out contrary fast fourier transform IFFT according to the Noisy Speech Signal of the decay factor G that calculates and handle.This step is same as the prior art, does not repeat them here.

In terms of existing technologies, first embodiment of the present invention calculates the estimated value of clean speech signal short-term spectrum amplitude according to maximum a posteriori MAP criterion in squelch, ask for the ratio of this estimated value and Noisy Speech Signal, obtain the computing formula of decay factor, by this computing formula as can be seen, calculate major part that decay factor G adopts be operating as evolution, division, square, simple operation such as addition, do not need exponent arithmetic and power operation, and as shown in Figure 4, ratio r _INumerical range (ζ) is [0,1], is convenient to very much the fixed point computing.Therefore in squelch, significantly reduce calculated amount, be more suitable for practical application.

In addition, it will be understood by those skilled in the art that the concrete numerical value of wishing to get decay factor G, need to calculate earlier the ratio r of priori signal to noise ratio (S/N ratio) ξ, posteriority signal to noise ratio (S/N ratio) γ and first kind modified Bessel function single order and zeroth order _I(ζ); And to calculate r _I(ζ), then need to calculate earlier the independent variable of Bessel's function

Calculate for further simplifying, can obtain the estimated value of ζ according to priori signal to noise ratio (S/N ratio) ξ and posteriority signal to noise ratio (S/N ratio) γ:

$\widehat{\mathrm{\&zeta;}}\&ap;2\sqrt{\frac{E\left[X^2\right]}{{\mathrm{\&lambda;}}_N}\frac{Y^2}{{\mathrm{\&lambda;}}_N}}=2\sqrt{\mathrm{\&xi;Y}}---\left(17\right)$

Because priori signal to noise ratio (S/N ratio) ξ and posteriority signal to noise ratio (S/N ratio) γ calculate in step before, therefore simple evolution, multiplication can calculate the independent variable ζ of Bessel's function, realizes very simple.

Second embodiment of the present invention relates to a kind of noise suppressing method.Second embodiment has been done further improvement on the basis of first embodiment, and its improvements mainly are: in second embodiment of the invention, store all possible ζ and corresponding r in advance _IMapping table (ζ); Asking for ratio r _IIn the time of (ζ), calculate the independent variable ζ of Bessel's function earlier, in mapping table, search the r corresponding with this ζ according to ζ then _I(ζ); At last with ξ, γ and r _I(ζ), substitution formula (5) calculates G, and idiographic flow as shown in Figure 5.Because that storage in advance is all possible ζ and corresponding r _IMapping table (ζ) therefore after calculating ζ, can directly find corresponding r rapidly by this mapping table _I(ζ), and need to store all values of first kind modified Bessel function single order and zeroth order in the prior art, the size of the storage list of second embodiment of the present invention can be reduced half, i.e. space storage data reduce half.

The 3rd embodiment of the present invention relates to a kind of noise suppressing method.The 3rd embodiment has been done further improvement on the basis of first embodiment, its improvements mainly are: in third embodiment of the invention, after calculating the independent variable ζ of Bessel's function, judge that whether ζ is greater than 1, if ζ is greater than 1, then according to formula

Calculate r _I(ζ); If ζ is less than or equal to 1, then according to formula r _I(ζ)=0.5 ζ calculates r _I(ζ); At last with ξ, γ and r _IIn (ζ) the substitution formula (5), calculate G, idiographic flow as shown in Figure 6.In the present embodiment, ratio r _ICalculating (ζ) is very simple, and does not need storage list, can further save storage space.And, use formula

With formula r _I(ζ)=0.5 r that calculates of ζ _I(ζ), with the r of reality _I(ζ) compare, almost can ignore in the error that performance exists.As shown in Figure 7, solid line is represented r among the figure _IActual value (ζ), long dot-and-dash line represents to adopt formula

The r that calculates _I(ζ) approximate value, short dot-and-dash line represents to adopt formula r _I(ζ)=0.5 _ζThe r that calculates _I(ζ) approximate value, (1,0.5) is long dot-and-dash line and short dashdotted joint.As can be seen from the figure, the r that in most of zone, calculates according to formula _I(ζ) approximate value curve and true ratio curve fit tightly, and have some little errors near 0dB, its maximum error＜0.07, and exist the scope of error very narrow, can accept in the scope.

In terms of existing technologies, the 3rd embodiment of the present invention is when reducing calculated amount, and it is main, objective performance also promotes to some extent.Objective performance refers to adopt certain objective algorithm to come assess performance, adopt segmental signal-to-noise ratio to promote (Segmental SNR Improvement) as evaluation criterion in the present embodiment, as shown in Figure 8, horizontal ordinate is represented input signal-to-noise ratio among the figure, ordinate is represented the amplitude that segmental signal-to-noise ratio promotes, as seen from the figure, adopt the MAP criterion than adopting MMSE and log-MMSE criterion to calculate all liftings to some extent of objective performance that decay factor is carried out the voice signal of squelch.Subjective performance refers to use subjective people to go to listen and handles back sound, adopt International Telecommunications Union's telecommunication standards group (International Telecommunication Union-Telecommunication Standardization Sector in the present embodiment, abbreviation " ITU-T ") P.862 Perceptual Evaluation of Speech Quality standard (Perceptual Evaluation of Speech Quality, be called for short " PESQ ") as evaluation criterion, as shown in Figure 9, horizontal ordinate is represented input signal-to-noise ratio among the figure, ordinate is represented the voice signal after squelch is carried out the P.862PESQ result of standard evaluation, as seen from the figure, the voice signal that adopts MAP criterion calculating decay factor to carry out squelch carries out squelch than employing MMSE and log-MMSE criterion, and all liftings to some extent of objective performance of not carrying out the voice signal of squelch.When carrying out performance simulation in the present embodiment, adopt 30 sentence (15 male voices among the received pronunciation storehouse NOIZEUS, 15 female voices), sampling rate is 8KHz, 16bit quantizes, with white noise on their aliasings, and be divided into 0dB, 5dB, 10dB and four grades of 15dB, the processing frame length is 20ms, Duplication is 50%, and Hamming window (Hamming) adopts directly decision-making (Decision-Direct) algorithm estimation priori signal to noise ratio (S/N ratio), noise variance adopts the simplest hard decision tracking mode to ask for, and the judgement mode adopts log-likelihood ratio.

Need to prove, above the step of the whole bag of tricks divide, just clear in order to describe, can merge into a step during realization or some step is split, be decomposed into a plurality of steps, as long as comprise identical logical relation, all in the protection domain of this patent; To adding inessential modification in the algorithm or in the flow process or introduce inessential design, but the core design that does not change its algorithm and flow process is all in the protection domain of this patent.

Four embodiment of the invention relates to a kind of noise suppressing system, comprises: the fast fourier transform module, noise estimation module, snr computation module, decay factor are asked for module, IFFT processing module.

Wherein the fast fourier transform module is used for Noisy Speech Signal is carried out fast fourier transform FFT, and the Noisy Speech Signal after the FFT conversion is exported to the noise estimation module.

The noise estimation module is used for that the Noisy Speech Signal after the FFT conversion is carried out noise and estimates.

The snr computation module is used for calculating priori signal to noise ratio (S/N ratio) and the posteriority signal to noise ratio (S/N ratio) of Noisy Speech Signal, and the priori signal to noise ratio (S/N ratio) calculated and posteriority signal to noise ratio (S/N ratio) is exported to decay factor ask for module.

Decay factor is asked for module and is used for asking for the estimated value of clean speech signal and the ratio of Noisy Speech Signal when receiving priori signal to noise ratio (S/N ratio) and posteriority signal to noise ratio (S/N ratio), obtains decay factor G; Wherein, the estimated value of clean speech signal is the estimated value according to the clean speech signal short-term spectrum amplitude of maximum a posteriori MAP criterion calculating.In the present embodiment, the computing formula of decay factor G is:

$G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$

Wherein, ξ represents the priori signal to noise ratio (S/N ratio); γ represents the posteriority signal to noise ratio (S/N ratio); ζ represents the variable of Bessel's function; r _I(ζ) ratio of expression first kind modified Bessel function single order and zeroth order.

The IFFT processing module is used for asking for the G that module obtains according to decay factor, Noisy Speech Signal is carried out contrary fast fourier transform IFFT handle.

Be not difficult to find that present embodiment is the system embodiment corresponding with first embodiment, present embodiment can with the enforcement of working in coordination of first embodiment.The correlation technique details of mentioning in first embodiment is still effective in the present embodiment, in order to reduce repetition, repeats no more here.Correspondingly, the correlation technique details of mentioning in the present embodiment also can be applicable in first embodiment.

What deserves to be mentioned is that each involved in present embodiment module is logic module, in actual applications, a logical block can be a physical location, also can be the part of a physical location, can also realize with the combination of a plurality of physical locations.In addition, for outstanding innovation part of the present invention, will not introduce not too close unit with solving technical matters relation proposed by the invention in the present embodiment, but this does not show the unit that does not have other in the present embodiment.

Fifth embodiment of the invention relates to a kind of noise suppressing system, as shown in figure 10.The 5th embodiment has been done further improvement on the basis of the 4th embodiment, and its improvements mainly are: in fifth embodiment of the invention, decay factor is asked for module and is comprised following submodule:

Sub module stored is used for storing in advance all possible ζ and corresponding r _IMapping table (ζ);

The independent variable calculating sub module is used for calculating ζ;

Search submodule, be used at sub module stored stored relation table, search the r corresponding with this ζ according to ζ _I(ζ);

The G calculating sub module is used for ξ, γ and r _I(ζ), substitution formula 5 calculates G.

In addition, it will be understood by those skilled in the art that the independent variable calculating sub module when calculating ζ, can obtain ζ after opening radical sign and multiply by 2 with the product of ξ and γ.

Because second embodiment is corresponding mutually with present embodiment, thus present embodiment can with the enforcement of working in coordination of second embodiment.The correlation technique details of mentioning in second embodiment is still effective in the present embodiment, and the technique effect that can reach in second embodiment can be realized in the present embodiment too, in order to reduce repetition, repeats no more here.Correspondingly, the correlation technique details of mentioning in the present embodiment also can be applicable in second embodiment.

Sixth embodiment of the invention relates to a kind of noise suppressing system, as shown in figure 11.The 6th embodiment has been done further improvement on the basis of four embodiments, and its improvements mainly are: in sixth embodiment of the invention, decay factor is asked for module and is comprised following submodule:

The independent variable calculating sub module is used for calculating ζ;

Judge submodule, be used for judging that whether ζ is greater than 1;

r _I(ζ) calculating sub module is used for judging that submodule judgement ζ is greater than 1 o'clock, according to formula

Calculate r _I(ζ); Judging that submodule judgement ζ is less than or equal at 1 o'clock, according to formula r _I(ζ)=0.5 ζ calculates r _I(ζ);

The G calculating sub module is used for ξ, γ and r _I(ζ), in the substitution formula 5, calculate G.

In addition, it will be understood by those skilled in the art that the independent variable calculating sub module when calculating ζ, can obtain ζ after opening radical sign and multiply by 2 with the product of ξ and γ.

Because the 3rd embodiment is corresponding mutually with present embodiment, thus present embodiment can with the enforcement of working in coordination of the 3rd embodiment.The correlation technique details of mentioning in the 3rd embodiment is still effective in the present embodiment, and the technique effect that can reach in the 3rd embodiment can be realized in the present embodiment too, in order to reduce repetition, repeats no more here.Correspondingly, the correlation technique details of mentioning in the present embodiment also can be applicable in the 3rd embodiment.

Persons of ordinary skill in the art may appreciate that the respective embodiments described above are to realize specific embodiments of the invention, and in actual applications, can do various changes to it in the form and details, and without departing from the spirit and scope of the present invention.

Claims (12)

1. a noise suppressing method is characterized in that, comprises following steps:

After the priori signal to noise ratio (S/N ratio) that calculates Noisy Speech Signal and posteriority signal to noise ratio (S/N ratio), ask for the estimated value of clean speech signal and the ratio of Noisy Speech Signal, obtain decay factor G; Wherein, described estimated value is the estimated value according to the clean speech signal short-term spectrum amplitude of maximum a posteriori MAP criterion calculating;

According to described G described Noisy Speech Signal being carried out contrary fast fourier transform IFFT handles.

2. noise suppressing method according to claim 1 is characterized in that, asks for described ratio according to following formula, obtains decay factor G:

$G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$

Wherein, ξ represents described priori signal to noise ratio (S/N ratio); γ represents described posteriority signal to noise ratio (S/N ratio); ζ represents the variable of Bessel's function; r _I(ζ) ratio of expression first kind modified Bessel function single order and zeroth order.

3. noise suppressing method according to claim 2 is characterized in that, also comprises following steps:

Store all possible ζ and corresponding r in advance _IMapping table (ζ);

Ask for described ratio, obtain comprising following substep in the step of decay factor G:

Calculate described ζ;

In described mapping table, search the r corresponding with this ζ according to described ζ _I(ζ);

With described ξ, described γ and the described r that finds _I(ζ), the described formula of substitution obtains described G.

4. noise suppressing method according to claim 2 is characterized in that, asks for described ratio, obtains comprising following substep in the step of decay factor G:

Calculate described ζ;

If described ζ is greater than 1, then according to formula

Calculate described r _I(ζ); If described ζ is less than or equal to 1, then according to formula r _I(ζ)=0.5 ζ calculates described r _I(ζ);

Described r with described ξ, described γ and calculating _I(ζ) the described formula of substitution $G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$ In, obtain described G.

5. according to claim 3 or 4 described noise suppressing methods, it is characterized in that, when calculating described ζ, with the product of described ξ and described γ, open radical sign and multiply by the numerical value that obtains after 2 as described ζ.

6. noise suppressing method according to claim 1 is characterized in that, also comprises following steps:

Before the priori signal to noise ratio (S/N ratio) and posteriority signal to noise ratio (S/N ratio) of calculating Noisy Speech Signal, Noisy Speech Signal is carried out fast fourier transform FFT;

Noisy Speech Signal after described FFT conversion is carried out noise to be estimated.

7. a noise suppressing system is characterized in that, comprises: snr computation module, decay factor are asked for module, IFFT processing module;

Described snr computation module is used for calculating priori signal to noise ratio (S/N ratio) and the posteriority signal to noise ratio (S/N ratio) of Noisy Speech Signal, and the described priori signal to noise ratio (S/N ratio) that will calculate and posteriority signal to noise ratio (S/N ratio) are exported to described decay factor and asked for module;

Described decay factor is asked for module and is used for asking for the estimated value of clean speech signal and the ratio of Noisy Speech Signal when receiving described priori signal to noise ratio (S/N ratio) and posteriority signal to noise ratio (S/N ratio), obtains decay factor G; Wherein, described estimated value is the estimated value according to the clean speech signal short-term spectrum amplitude of maximum a posteriori MAP criterion calculating;

Described IFFT processing module is used for asking for the G that module obtains according to described decay factor, described Noisy Speech Signal is carried out contrary fast fourier transform IFFT handle.

8. noise suppressing system according to claim 7 is characterized in that, described decay factor is asked for module and asked for described ratio according to following formula, obtains decay factor G:

$G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$

Wherein, ξ represents described priori signal to noise ratio (S/N ratio); γ represents described posteriority signal to noise ratio (S/N ratio); ζ represents the variable of Bessel's function; r _I(ζ) ratio of expression first kind modified Bessel function single order and zeroth order.

9. noise suppressing system according to claim 8 is characterized in that, described decay factor is asked for module and comprised following submodule:

Sub module stored is used for storing in advance all possible ζ and corresponding r _IMapping table (ζ);

The independent variable calculating sub module is used for calculating described ζ;

Search submodule, be used for the described mapping table in described sub module stored storage, search the r corresponding with this ζ according to the described ζ that calculates _I(ζ);

The G calculating sub module is used for described ξ, described γ and described r _I(ζ), the described formula of substitution obtains described G.

10. noise suppressing system according to claim 8 is characterized in that, described decay factor is asked for module and comprised following submodule:

The independent variable calculating sub module is used for calculating described ζ;

Judge submodule, be used for judging that whether the described ζ that calculates is greater than 1;

r _I(ζ) calculating sub module is used for judging that at described judgement submodule described ζ is greater than 1 o'clock, according to formula Calculate described r _I(ζ); Judge that at described judgement submodule described ζ is less than or equal at 1 o'clock, according to formula r _I(ζ)=0.5 ζ calculates described r _I(ζ);

The G calculating sub module is used for described ξ, described γ and described r _I(ζ), substitution formula $G=\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)+\sqrt{{\left\{\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})}{r}_I\left(\mathrm{\&zeta;}\right)\right\}}^2+\frac{\mathrm{\&xi;}}{2(1+\mathrm{\&xi;})\mathrm{\&gamma;}}}$ In, obtain described G.

11., it is characterized in that described independent variable calculating sub module with the product of described ξ and described γ, obtains described ζ after opening radical sign and multiply by 2 according to claim 9 or 10 described noise suppressing systems when calculating described ζ.

12. noise suppressing system according to claim 7 is characterized in that, this system also comprises: fast fourier transform module and noise estimation module;

Described fast fourier transform module is used for Noisy Speech Signal is carried out fast fourier transform FFT, and the Noisy Speech Signal after described FFT conversion is exported to the noise estimation module;

Described noise estimation module is used for that the Noisy Speech Signal after described FFT conversion is carried out noise and estimates.

Images