CN101426058A - System and method for improving quality of multichannel audio call - Google Patents

Abstract

A system for improving multi-channel audio speech quality, is provided with at least two audio collecting devices and at least two audio playback apparatus at local, wherein, each audio collecting device correspondingly collects an audio signal, each audio playback apparatus correspondingly outputs a reference signal. The system for improving multi-channel audio speech quality further includes several reecho path processing modules, wherein, the audio signal collected by each audio collecting device and the reference signal output by each audio playback apparatus adopt a hierarchical combination mode to realize the improvement of multi-channel audio speech quality by these reecho processing modules, the noise in the audio signal can be effectively removed by using a hierarchical combination mode to estimate residual reecho and noise of multi-channel audio signals, thereby greatly improving the speech quality and realizing channel expansion easily.

Description

A kind of system and method that improves quality of multichannel audio call

Technical field

The present invention relates to a kind of system and method that improves quality of multichannel audio call.Relate in particular to a kind of system and method for eliminating the quality of multichannel audio call of noise and echo.

Background technology

The video conferencing system of main flow has been brought into use the Audiotechnica of dual track or multichannel at present, and echo wherein, noise jamming also become problem maximum in the system.

Usually, the generation of echo is owing to exist the acoustics loop between loud speaker and the microphone, promptly after the loudspeaker plays of signal from conversation one end through the conversation other end, can be transmitted back to conversation one end by the microphone collection of this end simultaneously, the speaker of an end of causing thus conversing can hear the echo of oneself, and then has had a strong impact on speech quality.In addition, derive from stable state or the astable noise signal that the microphone surrounding enviroment produce and also speech quality is had very big influence.

And the interference between the multichannel audio call must have calculating and the realization difficulty more complicated than single channel.Multichannel (comprising stereo double channel) echo cancellation is the emphasis that whole audio quality improves, conventional solution is based on the method for sef-adapting filter and removes interchannel correlation, by removing the correlation of each path filter, interchannel frequency shift (FS), and wait and realize eliminating function carrying out nonlinear processor by staggered comb filter.

Please refer to the schematic diagram of existing stereo double channel echo cancelling system shown in Figure 1, the core component of this system is exactly a middle multichannel echo cancellation module, and this module has 4 signals inputs, is respectively the x1 from left and right sides microphone, x2 and from the d1 of left and right sides loud speaker, d2.Will there be 4 kinds of acoustics transitive relations in this module memory, the combination bang path of respectively corresponding different mic and loud speaker.

But this method is used an independently sef-adapting filter, and multichannel carries out echo and noise treatment to be difficult to expand to more from two passages.

Therefore, solve the voice communication quality that multichannel comprises stereo double channel on the whole and become the technical task that those skilled in the art need to be resolved hurrily.

Summary of the invention

The technical problem to be solved in the present invention is: a kind of system that improves quality of multichannel audio call is provided, and expanding to more with the echo in the effective elimination conversation and noise and the system of being convenient to from two passages, multichannel carries out echo and noise treatment.

In order to solve the problems of the technologies described above, the present invention adopts following technical scheme: a kind of system that improves quality of multichannel audio call, be used for the local system that is provided with at least two audio collecting devices and at least two audio playing apparatus simultaneously, wherein, audio signal of the corresponding collection of each audio collecting device, reference signal of the corresponding output of each audio playing apparatus, it is characterized in that: described system further comprises some echo path processing modules, the mode of the reference signal employing hierarchical composition of the audio signal of each audio collecting device collection and the output of each audio playing apparatus realizes improving quality of multichannel audio call by the processing of above-mentioned echo processing module, and described echo path processing module number is to determine according to the product number of the number of audio collecting device and audio playing apparatus.

As a preferred embodiment of the present invention, the principle that a plurality of audio signals are redistributed by phase place synthesizes the processing by above-mentioned echo processing module of the mode that adopts above-mentioned hierarchical composition after two channel stereo and realizes improving quality of multichannel audio call.

As a preferred embodiment of the present invention, be connected with echo cancellation module after each the echo processing module in the described system that improves quality of multichannel audio call.

The method of improving quality of multichannel audio call of the present invention is characterized in that may further comprise the steps:

1) adopt at least two microphones as signal input part, the corresponding collection playback equipment of each microphone;

2) signal that adopts at least two loudspeaker plays to export;

3) adopt a plurality of sef-adapting filter circulation carrying out calculating of echo path;

4) signal to described loudspeaker plays carries out preemphasis respectively;

5) signal that loud speaker (Spk1...Spkn) is corresponding is respectively as with reference to signal (X1...Xn), the signal Mic1 of Spk1 and microphone 1 carries out single transfer function and calculates, and will export the result at last as echo estimated value Y1, after first transfer function is calculated, obtained removing the signal Mic1_1=(Mic1-Y1) after the echo of Mic1 and Spk1; Calculate second transfer function then, use Mic1_1 and Spk2, export the result at last and be echo estimated value Y2, obtained removing the signal Mic1_2=(Mic1_1-Y2) after the echo of Mic1 and Spk2 respectively as input signal and reference signal; After the echo transfer function Yn of n reference signal Spkn and Mic1_n-1 correspondence being calculated so repeatedly always, just obtained normal signal, the final signal of first via microphone after just eliminating: Mic1_n=(Mic1_n-1-Yn) through multichannel echo corresponding to Mic1;

6) corresponding sef-adapting filter of the calculating of each transfer function, after the input time-domain signal is converted into frequency-region signal, by microphone signal and loudspeaker signal are handled respectively, calculate correlation and pass through sef-adapting filter function calculation filter weight coefficient w, add up and the rub-out signal e that exports calculates the echo signal Y of simulation by weight coefficient w, and then frequency-region signal is converted into time-domain signal y, obtain output signal e rub-out signal just after cutting echo signal y with microphone signal, constantly revise so repeatedly and calculating filter coefficient w, double counting and study reach the effect in estimated echo path;

7) be separate between each transfer function, the calculating of all corresponding linear regression learning parameter of the calculating of each transfer function, the calculations incorporated of the weight coefficient by this parameter and sef-adapting filter solve the renewal of the filter coefficient of both sides when speaking simultaneously and the problem of calculating.

As a preferred embodiment of the present invention, loudspeaker signal is carried out preemphasis respectively handle the raising output signal-to-noise ratio.

As a preferred embodiment of the present invention, after each transfer function is calculated, output is removed after the echo signal, still needing that output signal is carried out further residual echo suppresses to calculate, this part is calculated the residual echo that obtains according to sef-adapting filter and residual echo estimating part and is revealed parameter, carries out the inhibition once more of echo.

As a preferred embodiment of the present invention, the corresponding noise absorber of each microphone, this part is estimated to calculate according to noise, and in conjunction with gain control noise is suppressed.

As a preferred embodiment of the present invention, each output signal is given the system in another space by holding wire or network delivery.

As a preferred embodiment of the present invention, according to the size and the position in space, place 3,4 or 5 s' microphone, and synthesize by the principle that phase place is redistributed stereo, and and stereo sound card mutual.

In sum, the system and method that improves quality of multichannel audio call of the present invention carries out the estimation of echo by the mode that adopts hierarchical composition, has improved the quality of multi-channel audio communication effectively, has guaranteed the sense of reality and the telepresenc of voice communication.Can realize the perfect effect of stereo or multichannel collecting and playback.And can solve echo and noise problem under stereo or the multichannel, each passage is carried out echo cancellation and noise eliminating respectively, improve speech quality greatly and easily sound channel is handled expanding to multichannel processing.Thereby in video conference, realized the audio subsystem of " net is true ".

Description of drawings

Fig. 1 is the inside schematic diagram of existing stereo echo cancellation module;

Fig. 2 is the principle schematic of one tunnel signal processing in the system that improves quality of multichannel audio call of the present invention;

Fig. 3 is the principle schematic of another road signal processing in the system that improves quality of multichannel audio call of the present invention;

Fig. 4 is the principle schematic of stereophony Audio Processing in the system that improves quality of multichannel audio call of the present invention;

Fig. 5 is converted into the principle schematic of two channel stereo for quadraphony acquired signal.

Embodiment

System's technical problem solved by the invention of improving quality of multichannel audio call of the present invention is to provide a kind of multi-channel echo noise canceling system, and be combined into the system of a cover Audio Processing, with the sense of reality that reaches multi-channel audio communication and the audio quality of high-fidelity.

In order to solve the problem of echo and noise in the multichannel, the present invention proposes a kind of multi-channel echo and eliminates and noise canceling system.This system possesses easy expansion and easily realizes and possess echo noise cancellation efficiently.Thoroughly solved the difficult problem of acoustic processing in the multi-channel system.

Whole system has adopted following principle and mentality of designing:

1〉adopt 2 (more than) microphone as signal input part, the corresponding collection playback equipment (AD/DA chip or sound card) of each microphone.Only in this way just can collect real multichannel audio signal.So at first we require the sense of reality of signal and high-quality at the information source end.

2〉adopt 2 (more than) audio amplifier (loud speaker) play the signal of output.Equally only in this way we could reduce the multi channel signals of gathering.General stereo L channel and two audio amplifiers of R channel of corresponding to.Then need the support of AD/DA chip as for 2.1,5.1 playing environments such as grade.

3〉adopt a plurality of sef-adapting filter circulation carrying out calculating of echo path.Method is as follows: 2 tunnel independent audio are gathered mic1, mic2, and 2 tunnel reference signal ref1, the ref2 that export to loud speaker give sef-adapting filter Y1, Y2, Y3, Y4 respectively; Y1 is used for calculating the path model of mic1 and ref1, and Y2 is used for calculating the path model of mic1 and ref2, and Y3 is used for calculating the path model of mic2 and ref1, and Y4 is used for calculating the path model of mic2 and ref2.

4〉computational process of the corresponding transfer function of path model, relate to a separate space, and this separate space contains a plurality of loud speakers and a plurality of microphone, and form a plurality of audio frequency transfer systems, in the audio frequency transfer system,, gathered by a plurality of microphones again a plurality of speaker playbacks because having gone out the multichannel or the stereo audio of regeneration; Each transfer function by estimation a plurality of audio frequency transfer system correspondences as above, wherein, carry out preemphasis respectively by signal to described loudspeaker plays, the corresponding signal of loud speaker (Spk1...Spkn) is respectively as with reference to signal (X1...Xn), the signal Mic1 of Spk1 and microphone 1 carries out single transfer function and calculates, and will export the result at last as echo estimated value Y1, after first transfer function is calculated, obtained removing the signal Mic1_1=(Mic1-Y1) after the echo of Mic1 and Spk1; Calculate second transfer function then, use Mic1_1 and Spk2, export the result at last and be echo estimated value Y2, obtained removing the signal Mic1_2=(Mic1_1-Y2) after the echo of Mic1 and Spk2 respectively as input signal and reference signal; After the echo transfer function Yn of n reference signal Spkn and Mic1_n-1 correspondence being calculated so repeatedly always, just obtained normal signal, the final signal of first via microphone after just eliminating: Mic1_n=(Mic1_n-1-Yn) through multichannel echo corresponding to Mic1;

5〉as above be the recursive calculation of a plurality of transfer functions of a microphone correspondence in the separate space.The computational process of same a plurality of microphones is duplicate.Each corresponding output signal is passed through the system that holding wire or network delivery are given another space at last.

6〉corresponding sef-adapting filter of the calculating of each transfer function.After the input time-domain signal is converted into frequency-region signal, by microphone signal and loudspeaker signal are handled respectively, calculate correlation and pass through sef-adapting filter function calculation filter weight coefficient w, add up and the rub-out signal e that exports calculates the echo signal Y of simulation by weight coefficient w, and then frequency-region signal is converted into time-domain signal y, obtain output signal e rub-out signal just after cutting echo signal y with microphone signal, constantly revise and calculating filter coefficient w double counting and study so repeatedly.Reach the effect in estimated echo path.

7〉need carry out preemphasis respectively to loudspeaker signal handles.Reduced the high fdrequency component of noise, but therefore preemphasis has improved output signal-to-noise ratio effectively to not influence of noise.

8〉after each transfer function is calculated, exported after the removal echo signal, still needing that output signal is carried out further residual echo suppresses to calculate, this part is calculated the residual echo that obtains according to sef-adapting filter and residual echo estimating part and is revealed parameter, carry out the inhibition once more of echo, this part has played the residual echo that can eliminate more than 90%, obviously reduce the correlation between the different transfer functions, thereby made the echo elimination effect of different passages reach desirable effect.

9〉all calculating of a corresponding linear regression learning parameter of the calculating of each transfer function, the calculations incorporated of the weight coefficient by this parameter and aluminium sef-adapting filter can effectively solve the renewal of the filter coefficient of both sides when speaking simultaneously and the problem of calculating.Certainly be separate between each transfer function, the influence of the duplex of each passage also is independently like this.Thereby the collision problem of the coefficient calculations of the duplex communication between the multichannel of avoiding.

10〉corresponding noise absorber of each microphone, this part is estimated to calculate according to noise, and in conjunction with gain control noise is suppressed, and plays abating the noise, and improves the effect of speech quality.

11〉in general, each separate space is placed 2 microphones and 2 loud speakers (audio amplifier) can reach near stereosonic effect.Also can place 3,4 or 5 s' microphone according to the size and the position in space, and synthesize stereo by the principle that phase place is redistributed, and and stereo sound card mutual, reach the effect of multi-faceted while collected sound signal, make telepresenc true stereo more.

12〉need be under native system and the situation that PC (sound card) is connected, if we use the microphone that surpasses more than 4 as signal input sources, then we adopt following method: 4 tunnel audio collections independently, and (the system default setting is that MIC1 is 1 to set phase place (the volume ratio of left and right acoustic channels) parameter, 0 MIC2 is 0.75, and 0.25MIC3 is 0.25, and 0.75MIC4 is 0,1), just, the n paths communicates mutual principle with PC again by being re-assigned in the binary channels.

For example, specific to a kind of embodiment (not shown) of audio system of stereophony.This device is carried out transmission and the processing of carrying out the stereophony audio frequency between place A and the place B, is applied to video conferencing system, and can be used in combination with PC.In the separate space of place A, put two loud speaker Spk1-L and Spk1-R and two microphone Mic1-L and Mic1-R.With hearer is the center, and Spk1-L is placed in the left side, and Spk1-R is placed in the right same distance.Concrete distance and space arrangement are adjusted according to the space size.Mic1-L and Mic1-R also respectively correspondence be placed in the scope of speaker distance not far (generally at 0.5 meter in 5 meters), be placed on the left side and the right-hand part in space respectively, be used for gathering the signal of left and right acoustic channels.The stereo of two-way inputs to the stereo sound card of PC by circuit, then after the PC end carries out stereo compressed encoding, with the stereo place B that sends to.The PC end of place B carries out giving left and right sides loud speaker reduction respectively with the data of left and right acoustic channels behind the stereo decoding.

In the separate space of place B, put two loud speaker Spk2-L and Spk2-R and two microphone Mic2-L and Mic2-R equally.With hearer is the center, because dialogue time both sides are aspectant, so in the B space, Spk2-L will restore the signal of Mic1-R in the A space, Spk2-R will reduce the signal of the Mic1-L in the A space.

After receiving the signal that the A point sends over, Spk2-L of ordering through B and Spk2-R are also original sound the time, Mic2-L that B is ordered and Mic2-R be the voice signal of while in gathering the B space also, gives AD/DA chip, and the AD/DA chip is given AEC digital signal again and carried out echo elimination.Stereosonic L channel need carry out 2 transfer function self adaptations and calculate, and just carries out the calculating and the elimination of twice echo path.Suppress and given PC sound card by the AD/DA chip by circuit input at the noise to L channel afterwards, same, the self adaptation that stereosonic R channel also needs to carry out 2 transfer functions is calculated, and carries out the calculating and the elimination of twice echo path.Suppress and given PC sound card by circuit input at noise afterwards, carry out compressed package sent in stereophonic signal like this and go back to space to A by the AD/DA chip to L channel.The audio signal that does not have echo and noise will be heard in the A space in this time.

The computational process of first transfer function is described below, the corresponding sound conducting apparatus Mic1-L of this computational process and a loud speaker Spk1-L. hypothesis Mic1-L corresponding audio signal are d1, Spk-L is x1 to deserved audio signal, x1 is as the reference signal so, and d1 then is the local signal that needs processing.

At first reference signal x1 is carried out preemphasis and calculate, x1 is carried out obtaining frequency domain value X1 after the fast fourier transform then;

Entering sef-adapting filter then calculates; The filter input parameter is: reference signal frequency domain value X1, rub-out signal frequency domain value E1 and frequency domain adaptive mask value P1; Filter output median is the filter weight coefficient value W1 that adds up, and obtains final output signal Y1 by frequency spectrum conjugation multiply accumulating, and Y1 is exactly the frequency domain value of the echo signal estimated.

Change the time-domain signal y1 that Y1 is become estimated echo by reverse Fourier;

Calculated error signal e1 then just eliminates the output signal after the echo, rub-out signal e1=d1-y1; Then e1 is carried out obtaining frequency domain value E1 after the fast fourier transform;

At last, so repeatedly, the E1 value is preserved input parameter as filter calculating next time, reach by the calculating of correction wave filter continuously and restrain fast and eliminate echo.

After finishing the calculating of first transfer function, we have obtained an output signal e 1 of removing one road echo path, we continue to select e1 as local signal on this basis, and adopt the audio signal x2 of remaining another loud speaker Spk-R to carry out the calculating of next transfer function as the reference signal.

At first, still x2 is carried out preemphasis and calculate, x2 is carried out obtaining frequency domain value X2 after the fast fourier transform then; E1 postemphasised calculate e1 ';

Entering sef-adapting filter then calculates; The filter input parameter is: reference signal frequency domain value X2, rub-out signal frequency domain value E2 and frequency domain adaptive mask value P2; Filter output median is the filter weight coefficient value W2 that adds up, and obtains final output signal Y2 by frequency spectrum conjugation multiply accumulating, and Y2 is exactly the frequency domain value of the echo signal estimated.

Change the time-domain signal y2 that Y2 is become estimated echo by reverse Fourier;

Calculated error signal e2 then just eliminates the output signal after the echo, rub-out signal e2=e1 '-y2; Then e2 is carried out obtaining frequency domain value E2 after the fast fourier transform;

At last, so repeatedly, the E2 value is preserved input parameter as filter calculating next time, reach by the as above calculating of correction wave filter continuously and restrain fast and eliminate echo.

Arrive this, finished, then give the noise eliminating module output signal e 2 corresponding to the echo elimination of Mic1-L.Carry out one tunnel noise eliminating.Can give the sound card of outside or PC after the noise eliminating or send to the other end and carry out sound-reducing.

The same as above process of the calculating complete class of the part of Mic1-R.Difference is the d2 signal that input signal d1 has become the Mic-R correspondence.Other computational methods are just the same.

Filter adopts MDF, and following is the matrix variables of using:

e(l)＝F[0 _1xN，e(lN)，...，e(lN+N-1) ^T

x _k(l)＝diag{F[x((l-k-1)N)，...，x((l-k+1)N-1] ^T}

X(l)＝[x ₀，x ₁，...，x _k-1]

d(l)＝F[0 _1×N，d(lN)，...，d(1N+N-1)] ^T

Following is the computing formula of matrix:

e(l)＝d(l)-y(l)

y′(l)＝G ₁X(l)h′(l)

$h\&prime;(l+1)=h^{\&prime;}\left(l\right)+{\mathrm{<figure-callout\; id="2"\; label="G"\; filenames="A200810039070E00133.png"\; state="\{\{state\}\}">G</figure-callout>}}_2\mathrm{\&mu;}\left(l\right)\&dtri;h^{\&prime;}\left(l\right)$

$\&dtri;h\&prime;\left(l\right)={\mathrm{\&phi;}}_{\mathrm{xx}}^{-1}\left(l\right)X^H\left(l\right)e\left(l\right)$

In list of references [2], detailed explanation is arranged, please refer to.

G ₁And G ₂All be constraint matrix:

${\mathrm{<figure-callout\; id="1"\; label="G"\; filenames="A200810039070E00132.png,A200810039070E00133.png"\; state="\{\{state\}\}">G</figure-callout>}}_1=F\left[\begin{array}{cc}{0}_{N\&times;N}& 0_{N\&times;N}\\ 0_{N\&times;N}& I_{N\&times;N}\end{array}\right]F^{-1}$

${\mathrm{<figure-callout\; id="2"\; label="G"\; filenames="A200810039070E00133.png"\; state="\{\{state\}\}">G</figure-callout>}}_2\&prime;=F\left[\begin{array}{cc}{I}_{N\&times;N}& 0_{N\&times;N}\\ 0_{N\&times;N}& 0_{N\&times;N}\end{array}\right]F^{-1}$

G ₂＝diag{G ₂′，G ₂′，...，G ₂′}

Need to carry out linear regression learning parameter μ then _k(l) calculating;

${\mathrm{\&mu;}}_k\left(l\right)=\min (\frac{{\left|{y\&prime;}_k\left(l\right)\right|}^2}{{\left|e_k\left(l\right)\right|}^2},{\mathrm{\&mu;}}_0)$

Noise or residual echo are eliminated and are adopted energy spectrum subtraction estimated noise also to eliminate.Calculate the yield value of present frame through the short-time energy analysis and then by gain function, at last the frequency-region signal of output in that to carry out short-time energy synthetic.

Y _i(ω)＝S _i(ω)+N _i(ω)

$\widehat{\mathrm{\&mu;}}\left(\mathrm{\&omega;}\right)=\frac{1}{M}\underset{\mathrm{M\; noise\; frames}}{\mathrm{\&Sigma;}}\left|N_i\left(\mathrm{\&omega;}\right)\right|$

${\hat{S}}_i\left(\mathrm{\&omega;}\right)=G_i\left(\mathrm{\&omega;}\right)Y_i\left(\mathrm{\&omega;}\right)$

$G_i\left(\mathrm{\&omega;}\right)=\frac{\sqrt{\mathrm{\&pi;}}}{2}\sqrt{\left(\frac{1}{1+{\mathrm{SNR}}_{\mathrm{post}}}\right)\left(\frac{{\mathrm{SNR}}_{\mathrm{prio}}}{1+{\mathrm{SNR}}_{\mathrm{prio}}}\right)}$

$\&times;M\left[(1+{\mathrm{SNR}}_{\mathrm{post}})\left(\frac{{\mathrm{SNR}}_{\mathrm{prio}}}{1+{\mathrm{SNR}}_{\mathrm{prio}}}\right)\right]$

$M\left[\mathrm{\&theta;}\right]=e^{-\frac{\mathrm{\&theta;}}{2}}[(1+\mathrm{\&theta;})I_0\left(\frac{\mathrm{\&theta;}}{2}\right)+\mathrm{\&theta;}{I}_1\left(\frac{\mathrm{\&theta;}}{2}\right)]$

${\mathrm{SNR}}_{\mathrm{post}}\left(\mathrm{\&omega;}\right)=\frac{{\left|Y_i\left(\mathrm{\&omega;}\right)\right|}^2}{\widehat{\mathrm{\&mu;}}\left(\mathrm{\&omega;}\right)}-1$

Y _i(ω) be the original audio signal energy frequency spectrum that contains noise,

Be the average noise energy frequency spectrum of estimating, S _i(ω) be the energy frequency spectrum of normal original audio signal, N _i(ω) be the energy frequency spectrum of noise, G _iIt is gain function; SNR _PostIt is rearmounted signal to noise ratio; SNR _PrioIt is the priori signal to noise ratio; After short-time analysis by frequency spectrum calculates energy frequency spectrum, calculate according to as above formula and estimate gain function, so last calculating estimates

Signal behind the most of noise of elimination of our output just.

We so analogize, and after surpassing two Mic and two Speaker, our computational methods also are duplicate.The corresponding a plurality of Speaker of each Mic will do the calculating and the last corresponding output signal of output of transfer function repeatedly.And the processing of each passage all walks abreast.Also be mutually non-interfering.Exported the signal of a plurality of Mic ' at last.This signal has been eliminated echo and noise.

If M road Mic and N road Speak are arranged, so through M road output signal d1 '---dm ' must being arranged after this system handles, because N road Speak signal is arranged, the reference signal that then every road Mic signal all need the N road is calculated the transfer function of an echo path.So need to calculate M altogether ^*N transfer function.In above-mentioned embodiment, the system of the stereophony that we propose, input has 2 road Mic and Speak signal, then needs 4 groups of MDF filters and becomes a system with echo inhibition module level joint group.Output signal is 2 road signal d1 ' and d2 ', and this two paths of signals is exactly to eliminate the L channel behind echo and the noise and the digital signal of R channel.Give AD/DA and playback equipment then.

More specifically, the invention has the advantages that:

1, is easy to realize and expand. Can be easily from the stereophonic widening to the multichannel.

2, support the echo tail of 500ms. Each passage is all supported the echo tail of the length of 500ms, mutually solely Vertical and be independent of each other.

3, support the noise of each passage to eliminate.

4, support that each channel sample rate is unrestricted, support the single channel sample rate up to 48khz.

5, the input signal of each passage and output signal frequency scope support 20hz to 20000hz. So no matter be voice Or music can both well be processed, and comprises that music gets the processing that echo and noise too can high-fidelities.

6, the enhancing of the acoustic image distribution sense of live audio playback and the raising of sound articulation.

Claims (10)

1. system that improves quality of multichannel audio call, be used for the local system that is provided with at least two audio collecting devices and at least two audio playing apparatus simultaneously, wherein, audio signal of the corresponding collection of each audio collecting device, reference signal of the corresponding output of each audio playing apparatus, it is characterized in that: described system further comprises some echo path processing modules, the mode of the reference signal employing hierarchical composition of the audio signal of each audio collecting device collection and the output of each audio playing apparatus realizes improving quality of multichannel audio call by the processing of above-mentioned echo processing module, and described echo path processing module number is to determine according to the product number of the number of audio collecting device and audio playing apparatus.

2. a kind of system that improves quality of multichannel audio call as claimed in claim 1 is characterized in that: described hierarchical composition be meant each audio signal that audio collecting device gathers successively with the processing of each reference signal of audio playing apparatus output by above-mentioned echo processing module.

3. a kind of system that improves quality of multichannel audio call as claimed in claim 1 is characterized in that: the principle that a plurality of audio signals are redistributed by phase place synthesizes the processing by above-mentioned echo processing module of the mode that adopts above-mentioned hierarchical composition after two channel stereo and realizes improving quality of multichannel audio call.

4. a kind of system that improves quality of multichannel audio call as claimed in claim 1 is characterized in that: be connected with echo cancellation module after each the echo processing module in the described system that improves quality of multichannel audio call.

5. method of improving quality of multichannel audio call is characterized in that comprising step:

1) adopt at least two microphones as signal input part, the corresponding collection playback equipment of each microphone;

2) signal that adopts at least two loudspeaker plays to export;

3) adopt a plurality of sef-adapting filter circulation carrying out calculating of echo path;

4) signal to described loudspeaker plays carries out preemphasis respectively;

5) signal that loud speaker (Spk1...Spkn) is corresponding is respectively as with reference to signal (X1...Xn), the signal Mic1 of Spk1 and microphone 1 carries out single transfer function and calculates, and will export the result at last as echo estimated value Y1, after first transfer function is calculated, obtained removing the signal Mic1_1=(Mic1-Y1) after the echo of Mic1 and Spk1; Calculate second transfer function then, use Mic1_1 and Spk2, export the result at last and be echo estimated value Y2, obtained removing the signal Mic1_2=(Mic1_1-Y2) after the echo of Mic1 and Spk2 respectively as input signal and reference signal; After the echo transfer function Yn of n reference signal Spkn and Mic1_n-1 correspondence being calculated so repeatedly always, just obtained normal signal, the final signal of first via microphone after just eliminating: Mic1_n=(Mic1_n-1-Yn) through multichannel echo corresponding to Mic1;

6) corresponding sef-adapting filter of the calculating of each transfer function, after the input time-domain signal is converted into frequency-region signal, by microphone signal and loudspeaker signal are handled respectively, calculate correlation and pass through sef-adapting filter function calculation filter weight coefficient w, add up and the rub-out signal e that exports calculates the echo signal Y of simulation by weight coefficient w, and then frequency-region signal is converted into time-domain signal y, obtain output signal e rub-out signal just after cutting echo signal y with microphone signal, constantly revise so repeatedly and calculating filter coefficient w, double counting and study reach the effect in estimated echo path;

7) be separate between each transfer function, the calculating of all corresponding linear regression learning parameter of the calculating of each transfer function, the calculations incorporated of the weight coefficient by this parameter and sef-adapting filter solve the renewal of the filter coefficient of both sides when speaking simultaneously and the problem of calculating.

6. a kind of method of improving quality of multichannel audio call as claimed in claim 4 is characterized in that: loudspeaker signal is carried out preemphasis respectively handle the raising output signal-to-noise ratio.

7. a kind of method of improving quality of multichannel audio call as claimed in claim 4, it is characterized in that: after each transfer function is calculated, output is removed after the echo signal, still needing that output signal is carried out further residual echo suppresses to calculate, this part is calculated the residual echo that obtains according to sef-adapting filter and residual echo estimating part and is revealed parameter, carries out the inhibition once more of echo.

8. a kind of method of improving quality of multichannel audio call as claimed in claim 4 is characterized in that: the corresponding noise absorber of each microphone, this part is estimated to calculate according to noise, and in conjunction with gain control noise is suppressed.

9. a kind of method of improving quality of multichannel audio call as claimed in claim 4 is characterized in that: each output signal is given the system in another space by holding wire or network delivery.

10. a kind of method of improving quality of multichannel audio call as claimed in claim 4, it is characterized in that: according to the size and the position in space, place 3,4 or 5 s' microphone, and synthesize by the principle that phase place is redistributed stereo, and and stereo sound card mutual.

Images