CN1842110A - Echo eliminating device and method - Google Patents

Abstract

The invention discloses an echo elimination device which comprises a remote end buffer module, a near end buffer module, a self-adapt wave filter module, a first accumulator and a second accumulator.

Description

A kind of echo cancelling device and method

Technical field

The present invention relates to field of voice signal, be specifically related to a kind of echo cancelling device and method.

Background technology

Internet network technology, software engineering and Computer Multimedia Technology fast-developing and mutually combining constantly promoting the generation and the fusion of new business.Some business that need multiple proprietary hardware device to support also can realize and be deployed on the all-purpose computer by software traditionally, and the networking telephone that software is realized is exactly one of them.Compare with traditional real-time phone, also there is a certain distance in the present networking telephone on voice quality, and wherein, echo being influences one of most critical factor of its voice quality.

Fig. 1 is the echo schematic diagram in the communication network, the S among the figure _InThe signal that the expression near-end will send, S _OutThe signal that the expression far-end receives from near-end, R _InThe signal that the expression far-end will send, R _OutRemote signaling after the electric echo that the expression near-end receives is eliminated, as shown in Figure 1, echo can be divided into two kinds according to producing reason:

A kind of is " electric echo ", and the main cause that it produces is because the 2-4 line conversion that exists in the communication system.As shown in Figure 1, the local access switch of network side is with the voice signal S of near-end transmission _InOutput on the 2/4 line change-over circuit of network side, 2/4 line change-over circuit will carry out 2-4 line conversion and be about to the two-wire local loop and be connected on the 4 line main lines, in 2-4 line transfer process because of reasons such as impedance matchings, can be on the near-end RX path with some signals " leakage " on the near-end transmit path, the result causes near-end user to hear the voice of oneself.

Another kind is " acoustic echo ", and it is to be produced by the coupling of the voice between voice playing equipment and the voice capture device.As shown in Figure 1, be example with the near-end, the voice signal R that loudspeaker plays is come out from far-end _OutThrough different path one or many reflections or without any reflection, be delivered to microphone then, after being picked up by microphone, resend to far-end, remote subscriber is heard the voice of oneself as a result.

Existing echo cancellation scheme is the multipoint mode scheme, promptly all adopts special-purpose dsp chip to carry out echo elimination in end side and network side.When concrete the deployment, be difference according to the echo kind, Echo Canceller is deployed on the diverse location of system.

Because electric echo mainly produces at network side, so the electric echo arrester is deployed in network side usually, and Fig. 2 is the schematic diagram that existing realization electric echo is eliminated, S among the figure _In, R _Out, R _In, S _OutImplication identical with the implication of corresponding symbol among Fig. 1.As shown in Figure 2, particularly, the electric echo arrester is deployed between the local access switch and 2/4 line change-over circuit of network side.The detailed process that electric echo is eliminated is: because electric echo signal r _eBe owing to S on the near-end transmit path _InPortion of energy be leaked to by 2/4 line change-over circuit and produce on the near-end RX path, therefore, the electric echo arrester is according to S _OutEstimate the value r of electric echo signal _e', then from being mixed with the R of electric echo signal _InIn deduct r _e', like this, the R that finally obtains _OutThe approximate real speech signal that is equal to the far-end transmission.

Because acoustic echo mainly produces in end side, so acoustic echo canceller (AEC, Acoustic Echo Canceller) is deployed in end side usually, in microphone.Fig. 3 is the schematic diagram that existing realization acoustic echo is eliminated, S among the figure _In, R _Out, R _In, S _OutImplication identical with the implication of corresponding symbol among Fig. 1.As shown in Figure 3, be example with the near-end, because acoustic echo signal r _aThe signal that comes from loud speaker is R _OutSo AEC is according to R _OutEstimate the value r of acoustic echo signal _a', then from being mixed with the S of acoustic echo signal _InIn deduct r _a', the final like this S that obtains _OutThe approximate real speech signal that is equal to the near-end transmission.

The shortcoming of prior art is:

1, owing to need dispose Echo Canceller at network side and end side multiple spot, therefore, the echo canceller performance of disposing when network side descends or when losing efficacy, to all bringing adverse effect with the speech quality of related all terminals of this Echo Canceller in the system.

2, because in real network environment, actual network is to form by the subnet of multiple heterogeneous networks technology is interconnected, therefore, be deployed in the echo that the electric echo arrester of a certain position of network side can only solve this section network path of its deployed position, be difficult to guarantee on the whole that the echo of end user terminal eliminates effect.For example,, enter in the IP network by the IP telephony network pass, be delivered to the IP network terminal, influence speech quality but still understand some echo although the Echo Canceller that is deployed on the PSTN can remove a part of echo.

3, existing scheme is hardware solution basically, adopts special-purpose dsp chip, and cost and maintenance cost that echo is eliminated are very high.The more important thing is that echo elimination needs to preserve the reference signal that the voice signal in the past period is estimated as echo, because DSP is subjected to condition restriction such as hardware internal memory, elimination for the long streaking echo that surpasses 128ms is difficult to support, particularly in IP network, the message time delay always can not guarantee that less than 128ms this support that just causes existing echo to be eliminated for the speech data business on the IP network exists weak point.

Summary of the invention

In view of this, main purpose of the present invention is to provide a kind of echo cancelling device, dispose echo that Echo Canceller brings and eliminate the effect problem that influences each other to solve multiple spot, and the problem that can only eliminate the echo of subnetwork side, realize eliminating electric echo and acoustic echo comprehensively, and reduce complexity of disposing Echo Canceller and the cost of safeguarding Echo Canceller;

Simultaneously, main purpose of the present invention is to provide a kind of echo cancel method, realizes eliminating electric echo and acoustic echo simultaneously, and effectively eliminates the long streaking echo in the networking telephone.

For achieving the above object, technical scheme of the present invention is achieved in that

A kind of echo cancelling device, be applied to end side, this device comprises: be used to carry out far-end cache module and the first adder that acoustic echo is eliminated, be used to carry out near-end cache module and the second adder that electric echo is eliminated, be used for supporting simultaneously the two filtration modules of self adaptation of acoustic echo elimination and electric echo elimination, wherein

Remote signaling or remote end input signal after the two filtration modules of self adaptation are eliminated according to the electric echo of far-end cache module output are carried out the adaptive-filtering that acoustic echo is estimated, when the filtering result restrains, calculate the acoustic echo signal according to the remote signaling after acoustic echo filter factor and the electric echo elimination, then the acoustic echo signal is outputed to first adder; First adder calculates the near end signal after thereby the difference of near end input signal and acoustics echo signal obtains the acoustic echo elimination, and this signal is outputed to the near-end cache module, and exports to far-end;

The two filtration modules of self adaptation according to the output of near-end cache module near end signal or the near end input signal of acoustic echo after eliminating carry out the adaptive-filtering that electric echo is estimated, when the filtering result restrains, calculate the electric echo signal according to the near end signal after electric echo filter factor and the acoustic echo elimination, then the electric echo signal is outputed to second adder; Second adder calculates the remote signaling after thereby the difference of remote end input signal and electric echo signal obtains the electric echo elimination, and this signal is outputed to the far-end cache module, and exports to near-end.

Described device further comprises,

The near-end speech detection module is used to receive the near end signal after near end input signal or acoustic echo are eliminated, and when judging this signal and being voice, sends the near-end speech triggering signal to adaptive-filtering coefficient adjustment module;

The far-end speech detection module is used to receive the remote signaling after remote end input signal or electric echo are eliminated, and when judging this signal and being voice, sends the far-end speech triggering signal to adaptive-filtering coefficient adjustment module;

Adaptive-filtering coefficient adjustment module, when receiving only the near-end speech triggering signal that the near-end speech detection module sends, electric echo is finely tuned step-length send to the two filtration modules of self adaptation, the two filtration modules of self adaptation are adjusted current electric echo filter factor according to this step-length

When receiving only the far-end speech triggering signal that the far-end speech detection module sends, acoustic echo to be finely tuned step-length send to the two filtration modules of self adaptation, the two filtration modules of self adaptation are adjusted current acoustic echo filter factor according to this fine setting step-length,

When not receiving any triggering signal, whether inquire about current filtering result to the two filtration modules of self adaptation restrains, if convergence, just simultaneously electric echo is finely tuned step-length and acoustic echo fine setting step-length and send to the two filtration modules of self adaptation, the two filtration modules of self adaptation are adjusted current electric echo filter factor according to electric echo fine setting step-length, adjust current acoustic echo filter factor according to acoustic echo fine setting step-length, if do not restrain, just give self adaptation two filtration modules fast pacing length of electric echo and the fast pacing long hair of acoustic echo simultaneously, the two filtration modules of self adaptation are according to the current electric echo filter factor of the long adjustment of the fast pacing of electric echo, according to the current acoustic echo filter factor of the long adjustment of the fast pacing of acoustic echo

When receiving near-end speech triggering signal that near-end speech detection module and far-end speech detection module send and far-end speech triggering signal simultaneously, these two triggering signals are transmitted to the two filtration modules of self adaptation simultaneously do not adjust current electric echo and acoustic echo filter factor with the two filtration modules of notice self adaptation.

Described near-end speech detection module comprises:

Near-end energy computing module, be used to receive the near end signal after near end input signal or acoustic echo are eliminated, calculate the energy of this signal, and determine the near-end speech threshold value according to the energy of this signal of quiet period, the energy of this signal during with this near-end speech threshold value and formal conversation outputs to the near-end speech judge module;

The near-end speech judge module, the energy of the near end signal after near end input signal when being used to receive near-end speech threshold value and formal conversation or acoustic echo are eliminated, relatively the size of the two is to judge whether the near end signal after near end input signal or acoustic echo are eliminated is voice, if voice send the near-end speech triggering signal for adaptive-filtering coefficient adjustment module.

Described far-end speech detection module comprises:

Far-end energy computing module, be used to receive the remote signaling after remote end input signal or electric echo are eliminated, calculate the energy of this signal, and determine the far-end speech threshold value according to the energy of this signal of quiet period, the energy of this signal during with this far-end speech threshold value and formal conversation outputs to the far-end speech judge module;

The far-end speech judge module, the energy of the remote signaling after remote end input signal when being used to receive far-end speech threshold value and formal conversation or electric echo are eliminated, relatively the size of the two is to judge whether the remote signaling after remote end input signal or electric echo are eliminated is voice, if voice send the far-end speech triggering signal for adaptive-filtering coefficient adjustment module.

A kind of echo cancel method, this method comprises:

Remote signaling after near end signal after A, preservation acoustic echo are eliminated and electric echo are eliminated;

B, near end signal or near end input signal after remote signaling after eliminating according to the electric echo of having preserved respectively or remote end input signal and acoustic echo are eliminated are carried out the adaptive-filtering that acoustic echo is estimated and electric echo is estimated, and judge whether the filtering result that acoustic echo and electric echo are estimated restrains, if the filtering result that acoustic echo is estimated convergence, remote signaling and acoustic echo filter factor after eliminating according to electric echo calculate the acoustic echo signal, and the difference of then calculating current near end input signal and acoustics echo signal promptly obtains the near end signal after acoustic echo is eliminated; If the filtering result that electric echo is estimated convergence, near end signal and electric echo filter factor after eliminating according to acoustic echo calculate the electric echo signal, and the difference of then calculating current remote end input signal and electric echo signal promptly obtains the remote signaling after electric echo is eliminated; Otherwise, proceed the adaptive-filtering that acoustic echo or electric echo are estimated.

Further comprise before the described step B: judge whether the remote signaling after near end signal after near end input signal or acoustic echo are eliminated and remote end input signal or electric echo are eliminated is voice, if the two all is voice, direct execution in step B; If the near end signal after near end input signal or acoustic echo are eliminated is voice, adjust current electric echo filter factor, then execution in step B; If the remote signaling after remote end input signal or electric echo are eliminated is voice, adjust current acoustic echo filter factor, then execution in step B; If the two is not voice, adjust current electric echo filter factor and acoustic echo filter factor simultaneously, then execution in step B.

Described adjustment electric echo filter factor be by: $e_{n+1}=e_n+\frac{{\mathrm{\μ}}_e}{d_e+{\mathrm{PS}}_{\mathrm{out}}\left(n\right)}{R}_{\mathrm{out}}\left(n\right)S_{\mathrm{out}}\left(n\right)$ Realize, wherein, e _nBe current electric echo filter factor, e _N+1Be adjusted electric echo filter factor, μ _eFor adjusting step-length, d _eBe the constant between 0～1, PS _Out(n) be the energy of current near-end output signal, R _Out(n) be remote signaling after current electric echo is eliminated, S _Out(n) be near end signal after current acoustic echo is eliminated.

Described adjustment acoustic echo filter factor be by $a_{n+1}=a_n+\frac{{\mathrm{\μ}}_a}{d_a+{\mathrm{PR}}_{\mathrm{out}}\left(n\right)}{S}_{\mathrm{out}}\left(n\right)R_{\mathrm{out}}\left(n\right)$ Realize, wherein, a _nBe current acoustic echo filter factor, a _N+1Be adjusted acoustic echo filter factor, μ _aFor adjusting step-length, d _aBe the constant between 0～1, PR _Out(n) be the energy of current far-end output signal, S _Out(n) be near end signal after current acoustic echo is eliminated, R _Out(n) be remote signaling after current electric echo is eliminated.

Describedly judge that whether the near end signal after near end input signal or acoustic echo are eliminated is that the concrete steps of voice are:

Calculate the energy of the near end signal after the near end input signal of each time period in the quiet period after session is set up or acoustic echo are eliminated, with the maximum of each energy with greater than the product of 1 constant as the near-end speech threshold value; After the conversation beginning, calculate the energy of the near end signal after current near end input signal or acoustic echo are eliminated, with the energy that obtains and near-end speech threshold ratio, if the former, judges that the near end signal after current near end input signal or acoustic echo are eliminated is voice greater than the latter; Otherwise, judge that the near end signal after current near end input signal or acoustic echo are eliminated is not voice.

Describedly judge that whether the remote signaling after remote end input signal or electric echo are eliminated is that the concrete steps of voice are:

Calculate the energy of the remote signaling after the remote signaling of each time period in the quiet period after session is set up or electric echo are eliminated, with the maximum of each energy with greater than the product of 1 constant as the far-end speech threshold value; After the conversation beginning, calculate the energy of the remote signaling after current remote end input signal or electric echo are eliminated, with the energy that obtains and far-end speech threshold ratio, if the former, judges that the remote signaling after current remote end input signal or electric echo are eliminated is voice greater than the latter; Otherwise, judge that the remote signaling after current remote end input signal or electric echo are eliminated is not voice.

Compared with prior art, echo cancelling device provided by the invention both can estimate according to the signal of near-end that the electric echo signal also can estimate the acoustic echo signal, and then elimination electric echo and acoustic echo, therefore, apparatus of the present invention only need be deployed in end side, have reduced the complexity and the maintenance cost of disposing.Simultaneously, method provided by the invention can be according to the calculated signals electric echo signal or the acoustic echo signal of near-end, reach the purpose of eliminating electric echo and acoustic echo simultaneously, simultaneously, the inventive method is by utilizing the end side reasonable resources, can preserve long echo and estimate reference signal, thereby effectively eliminate the long streaking echo in the networking telephone.

Description of drawings

Fig. 1 is the echo schematic diagram in the communication network;

Fig. 2 realizes the schematic diagram that electric echo is eliminated in the prior art;

Fig. 3 realizes the schematic diagram that acoustic echo is eliminated in the prior art;

Fig. 4 is the composition frame chart that the present invention realizes the specific embodiment one that echo is eliminated;

Fig. 5 is the composition frame chart that the present invention realizes the specific embodiment two that echo is eliminated;

Fig. 6 is the composition frame chart that the present invention realizes the specific embodiment three that echo is eliminated;

Fig. 7 is the composition frame chart that the present invention realizes the specific embodiment four that echo is eliminated;

Fig. 8 is the flow chart that the present invention realizes the specific embodiment two that echo is eliminated.

Embodiment

Core concept of the present invention is: propose a kind of echo cancelling device, this device only need be deployed in end side, can eliminate acoustic echo and electric echo simultaneously, proposes a kind of echo cancel method simultaneously, only use this method, can eliminate acoustic echo and electric echo simultaneously in end side.

The present invention is further described in more detail below in conjunction with drawings and the specific embodiments.

Fig. 4 is the composition frame chart that the present invention realizes the specific embodiment one that echo is eliminated, the S among the figure _InThe initialize signal that the expression near-end will send to far-end is near end input signal, S _OutThe signal of eliminating through echo that the expression near-end outputs to far-end is the near end signal after acoustic echo is eliminated, R _InThe initialize signal that the expression far-end sends to near-end is remote end input signal, R _OutThe remote signaling through the echo elimination that the expression near-end finally receives is the remote signaling after electric echo is eliminated.As shown in Figure 4, it mainly comprises:

Far-end cache module 40: be used for receiving and preserving R _OutOr R _In, and with R _OutOr R _InOutput to the two filtration modules 42 of self adaptation.

Near-end cache module 41: be used for receiving and preserving S _OutOr S _In, and with S _OutOr S _InOutput to the two filtration modules 42 of self adaptation.

That Fig. 4 provides is the R that is input as of far-end cache module 40 _Out, near-end cache module 41 be input as S _OutSituation, far-end cache module 40 be input as R _In, near-end cache module 41 be input as S _InStructured flowchart can simply release.

The two filtration modules 42 of self adaptation: be used to preserve acoustic echo filter factor and electric echo filter factor, be used for R according to 40 outputs of far-end cache module _OutOr R _InCarry out the adaptive-filtering of acoustic echo estimation and the S that exports according to near-end cache module 41 _OutOr S _InCarry out the adaptive-filtering that electric echo is estimated, when the filtering result restrains, according to R _OutOr R _InCalculate acoustic echo signal r with the acoustic echo filter factor _a', then with r _a' output to first adder; When the filtering result restrains, according to S _OutOr and the electric echo filter factor calculate electric echo signal r _a', then the electric echo signal is outputed to second adder.

First adder: be used for according to the S that receives _InWith acoustics echo signal r _a', calculate S _InAnd r _a' difference, this difference is the near end signal after acoustic echo is eliminated, and then this difference is outputed to near-end cache module 41, simultaneously this difference is outputed to far-end.

Second adder: be used for according to the R that receives _InWith electric echo signal r _e', calculate R _InAnd r _e' difference, this difference is the remote signaling after electric echo is eliminated, and then this difference is outputed to far-end cache module 40, simultaneously this difference is outputed to near-end.

Fig. 5 is the composition frame chart that the present invention realizes the specific embodiment two that echo is eliminated, and compares with Fig. 4, and this embodiment further comprises:

Near-end speech detection module 43: whether be used to receive near end signal, detecting near end signal is voice, when being voice, sends a near-end speech triggering signal 10 for adaptive-filtering coefficient adjustment module 45.

Here, near end signal both can be S _In, also can be the S shown in the input signal of the near-end speech detection module 53 of Fig. 6 _Out

Far-end speech detection module 44: whether be used to receive remote signaling, detecting remote signaling is voice, when being voice, sends a far-end speech triggering signal 11 for adaptive-filtering coefficient adjustment module 45.

Here, remote signaling both can be R _Out, also can be the R shown in the input signal of the far-end speech detection module 54 of Fig. 6 _In

Adaptive-filtering coefficient adjustment module 45: this module functions is as follows:

1, when receiving only the near-end speech triggering signal 10 that near-end speech detection module 43 sends, electric echo is finely tuned the two filtration modules 42 of step-length notice self adaptation, and the two filtration modules 42 of self adaptation receive that the back is according to the current electric echo filter factor of this fine setting step-length adjustment;

2, when receiving only the far-end speech triggering signal 11 that far-end speech detection module 44 sends, acoustic echo is finely tuned the two filtration modules 42 of step-length notice self adaptation, and the two filtration modules 42 of self adaptation receive that the back is according to the current acoustic echo filter factor of this fine setting step-length adjustment;

3, when not receiving any triggering signal, whether current filtering result restrains to two filtration module 42 inquiries of self adaptation, if what the two filtration modules 42 of self adaptation returned is convergence, just simultaneously electric echo is finely tuned step-length and acoustic echo fine setting step-length and send to the two filtration modules 42 of self adaptation, the two filtration modules 42 of self adaptation are adjusted current electric echo filter factor according to electric echo fine setting step-length, adjust current acoustic echo filter factor according to acoustic echo fine setting step-length, if what the two filtration modules 42 of self adaptation returned is not restrain, just give self adaptation two filtration modules 42 fast pacing length of electric echo and the fast pacing long hair of acoustic echo simultaneously, the two filtration modules 42 of self adaptation are according to the current electric echo filter factor of the long adjustment of the fast pacing of electric echo, according to the current acoustic echo filter factor of the long adjustment of the fast pacing of acoustic echo;

4, when receiving near-end speech triggering signal 10 that near-end speech detection module 43 and far-end speech detection module 44 send and far-end speech triggering signal 11 simultaneously, transmit near-end speech triggering signal 10 simultaneously for the two filtration modules 42 of self adaptation and far-end speech triggering signal 11 needn't be adjusted current acoustic echo and electric echo filter factor with the two filtration modules 42 of notice self adaptation.

It is pointed out that only to have provided near-end speech detection module and the two kind positions of far-end speech detection module in the total block diagram among Fig. 5 and Fig. 6, in the practical application, because the input signal of near-end speech detection module can be S _InOr S _Out, the input signal of far-end speech detection module can be R _InOr R _Out, therefore, can release: according to the difference of input signal, near-end speech detection module and the position of far-end speech detection module in the total block diagram have four kinds.

Fig. 7 is the composition frame chart that the present invention realizes the specific embodiment four that echo is eliminated, and as shown in Figure 7, this embodiment has comprised whole modules of Fig. 5, and:

Near-end speech detection module 43 comprises:

Near-end energy computing module 430: be used to receive near end signal, calculate the energy of near end signal, maximum according to the energy of the near end signal of quiet period is determined the near-end speech threshold value, and the energy of the near end signal when this voice threshold value and near-end or far-end had the formal conversation of voice signal outputs to near-end speech judge module 431.

Quiet period refers to be conversed beginning a period of time before to formal after session is set up, and length was generally 5～60 seconds.Can learn according to session signaling whether session sets up, for example: can learn by signalling system No.7 in the PSTN network, in the IP network can by the SIP signaling or H.323 signaling learn.

Near-end speech judge module 431: the energy of the near end signal when being used to receive near-end speech threshold value and formal conversation, the two is compared judging whether near end signal is voice, when near end signal sends a near-end speech triggering signal 10 for adaptive-filtering coefficient adjustment module 45 during for voice.

Here, near end signal both can be S _In, also can be S _Out

Far-end speech detection module 44 comprises:

Far-end energy computing module 440: be used to receive remote signaling, calculate the energy of remote signaling, maximum according to the energy of the remote signaling of quiet period is determined the far-end speech threshold value, and the energy of the remote signaling during with this voice threshold value and formal conversation outputs to far-end speech judge module 441.

Far-end speech judge module 441: the energy of the remote signaling when being used to receive far-end speech threshold value and formal conversation, the two is compared judging whether remote signaling is voice, when remote signaling sends a far-end speech triggering signal 11 for adaptive-filtering coefficient adjustment module 45 during for voice.

Here, remote signaling both can be R _In, also can be R _Out

Below be the concrete steps of the specific embodiment one of realization echo provided by the invention elimination:

Step 01: far-end cache module 40 is preserved R _OutOr R _In, simultaneously, near-end cache module 41 is preserved S _OutOr S _In

Step 02: the two filtration modules 42 of self adaptation are according to the R that has preserved _OutOr R _InCarry out the adaptive-filtering that acoustic echo is estimated, according to S _OutOr S _InCarry out the adaptive-filtering that electric echo is estimated, and judge whether the filtering result restrains, if acoustic echo signal r is calculated in the filtering result that acoustic echo is estimated convergence _a', calculate S then _OutIf electric echo signal r is calculated in the filtering result that electric echo is estimated convergence _a', calculate R then _OutOtherwise, proceed the adaptive-filtering that acoustic echo or electric echo are estimated.

The two filtration modules 42 of self adaptation when the filtering result restrains, just calculate acoustic echo and electric echo ceaselessly carrying out the adaptive-filtering that acoustic echo and electric echo are estimated.

The two filtration modules 42 of self adaptation carry out the filtering algorithm that acoustic echo is estimated and electric echo is estimated and can adopt common LMS algorithm, least square method etc.

Acoustic echo calculated signals formula is:

$r_a^{\′}\left(n\right)=\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}{a}_{n-k}{R}_{\mathrm{out}}(n-k)---\left(1\right)$

Wherein, a _N-kBe the acoustic echo filter factor, n represents current moment value (implication of the n in the following formula all implication with the n of formula (1) is identical), R _Out(n-k) (k=0 ..., the N-1) R in the nearest certain hour section of expression _Out

In the error allowed band, also can R _InFor reference signal is estimated r _a'.

Electric echo calculated signals formula is:

$r_e^{\′}\left(n\right)=\underset{k=0}{\overset{N-1}{\mathrm{\Σ}}}{e}_{n-k}{S}_{\mathrm{out}}(n-k)---\left(2\right)$

Wherein, e _N-kBe electric echo filter factor, S _Out(n-k) (k=0 ..., the N-1) S in the nearest certain hour section of expression _Out

In the error allowed band, also can S _InFor reference signal is estimated r _e'.

S _OutComputing formula be:

S _out(n)＝S _in(n)-r _a′(n) (3)

R _OutComputing formula be:

R _out(n)＝R _in(n)-r _e′(n) (4)

Fig. 8 is the flow chart that the present invention realizes the specific embodiment two that echo is eliminated, and as shown in Figure 8, its concrete steps are as follows:

Step 801: far-end cache module 40 is preserved R _OutOr R _In, simultaneously, near-end cache module 41 is preserved S _OutOr S _InNear-end speech detection module 43 judges whether current near end signal is voice; Simultaneously, far-end speech detection module 44 judges whether current remote signaling is voice, if the two is not voice, and execution in step 802; If the two is voice simultaneously, execution in step 805; If the two one of be voice, execution in step 806.

Here, near end signal both can be S _In, also can be S _OutRemote signaling both can be R _In, also can be R _Out

Wherein, near-end speech detection module 43 judges that whether near end signal is that the detailed process of voice is with S _OutBe example, can for:

A1, near-end energy computing module 430 calculate the S that each time period of back quiet period is set up in session _OutEnergy PS _Out(n), get their maximum, with this maximum be multiplied by one greater than the value that obtains behind 1 the constant as the near-end speech threshold value, and this voice threshold value outputed to near-end speech judge module 431;

PS _Out(n) computing formula is:

PS _out(n)＝S _out(n) ^TS _out(n) (5)

Wherein, S _Out(n) ^TBe S _Out(n) transposed matrix.S _OutBe sampled value repeatedly, therefore, represent with matrix form.

After A2, the conversation beginning, near-end energy computing module 430 calculates current S according to formula (5) _InEnergy PS _Out(n), then this energy value is outputed to near-end speech judge module 431;

A3, near-end speech judge module 431 are judged PS _Out(n) whether greater than the near-end speech threshold value, if judge that current near end signal is voice; Otherwise, judge that current near end signal is a non-voice.

Judge S _InWhether be the detailed process of voice, only need the S in the above-mentioned steps _OutReplace with S _InGet final product.

Far-end speech detection module 44 judges that whether remote signaling is that the detailed process of voice is with R _OutBe example, can for:

B1, far-end energy computing module 440 calculate the R that each time period of back quiet period is set up in session _OutEnergy PR _Out(n), get their maximum, with this maximum be multiplied by one greater than the value that obtains behind 1 the constant as the far-end speech threshold value, and this voice threshold value outputed to far-end speech judge module 441;

PR _Out(n) computing formula is:

PR _out(n)＝R _out(n) ^TR _out(n) (6)

After B2, the conversation beginning, far-end energy computing module 440 calculates current R according to formula (6) _OutEnergy PR _Out(n), then this energy value is outputed to far-end speech judge module 441;

B3, far-end speech judge module 441 are judged PR _Out(n) whether greater than the far-end speech threshold value, if judge that current remote signaling is voice; Otherwise, judge that current remote signaling is a non-voice.

Judge R _InWhether be the detailed process of voice, only need the R in the above-mentioned steps _OutReplace with R _InGet final product.

Here, near-end speech threshold value and far-end speech threshold value also can be determined by experience.

Step 802: whether adaptive-filtering coefficient adjustment module 45 restrains to the current filtering result of two filtration module 42 inquiries of self adaptation, if return results is for being execution in step 803; Otherwise, execution in step 804.

Step 803: the fine setting step-length of adaptive-filtering coefficient adjustment module two filtration module 42 acoustic echos of 45 notice self adaptations and electric echo, the two filtration modules 42 of self adaptation are according to formula (7) and formula (8), finely tune current electric echo and acoustic echo filter factor simultaneously, execution in step 807 then.

The electric echo filter factor is adjusted formula:

$e_{n+1}=e_n+\frac{{\mathrm{\μ}}_e}{d_e+{\mathrm{PS}}_{\mathrm{out}}\left(n\right)}{R}_{\mathrm{out}}\left(n\right)S_{\mathrm{out}}\left(n\right)---\left(7\right)$

Wherein, e _nBe current electric echo filter factor, e _N+1Be adjusted electric echo filter factor, μ _eFor adjusting step-length, get the value between 0～2 usually, d _eIt is the constant between 0～1.

The acoustic echo filter factor is adjusted formula:

$a_{n+1}=a_n+\frac{{\mathrm{\μ}}_a}{d_a+{\mathrm{PR}}_{\mathrm{out}}\left(k\right)}{S}_{\mathrm{out}}\left(n\right)R_{\mathrm{out}}\left(n\right)---\left(8\right)$

Wherein, a _nBe current acoustic echo filter factor, a _N+1Be adjusted acoustic echo filter factor, μ _aFor adjusting step-length, get the value between 0～2 usually, d _aIt is the constant between 0～1.

The fine setting step-length promptly adopts less μ _eOr μ _a, μ _eOr μ _aHour, error is less, but needs long convergence time.

Step 804: the fast pacing of adaptive-filtering coefficient adjustment module two filtration module 42 acoustic echos of 45 notice self adaptations and electric echo is long, the two filtration modules 42 of self adaptation are according to formula (7) and formula (8) then, current electric echo and the acoustic echo filter factor of fast simultaneously accent, execution in step 807 then.

The bigger μ of the long i.e. employing of fast pacing _eOr μ _a, μ _eOr μ _aWhen big, error is bigger, but convergence rate is very fast.

Step 805: the two filtration modules 42 of adaptive-filtering coefficient adjustment module 45 notice self adaptations needn't be adjusted current filter factor, and execution in step 807 then.

Step 806: if near end signal is voice, the two filtration module 42 electric echos fine setting of adaptive-filtering coefficient adjustment module 45 notice self adaptations step-length, the two filtration modules 42 of self adaptation are according to the current electric echo filter factor of formula (7) fine setting then; If remote signaling is voice, the two filtration module 42 acoustic echos fine setting of adaptive-filtering coefficient adjustment module 45 notice self adaptations step-length, the two filtration modules 42 of self adaptation are according to the current acoustic echo filter factor of formula (7) fine setting then, and execution in step 807 then.

Step 807: the two filtration modules 42 of self adaptation are according to the R that has preserved _OutOr R _InCarry out the adaptive-filtering that acoustic echo is estimated, according to S _OutOr S _InCarry out the adaptive-filtering that electric echo is estimated, and judge whether the filtering result restrains, if the filtering result that acoustic echo is estimated convergence calculates acoustic echo signal r according to formula (1) _a' after, with acoustic echo signal r _a' outputing to first adder, first adder calculates S according to formula (3) then _OutIf the filtering result that electric echo is estimated convergence is calculated electric echo signal r according to formula (2) _e' after, with electric echo signal r _e' outputing to second adder, second adder calculates R far away according to formula (4) then _Out, calculating finishes returns step 801; Otherwise, return step 807.

Echo cancel method provided by the invention can be with S _OutOr S _InThe voice data that is kept at near-end sends in the buffering area, with R _OutOr R _InBe kept in the voice data play buffer,, set up the reference signal of estimating electric echo signal and acoustics echo signal based on control to these audio data buffers.Because the memory space of audio data buffer is relatively large, therefore can effectively eliminate long streaking echo.

The above only is process of the present invention and method embodiment, in order to restriction the present invention, all any modifications of being made within the spirit and principles in the present invention, is not equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (10)

1, a kind of echo cancelling device, be applied to end side, it is characterized in that, this device comprises: be used to carry out far-end cache module and the first adder that acoustic echo is eliminated, be used to carry out near-end cache module and the second adder that electric echo is eliminated, be used for supporting simultaneously the two filtration modules of self adaptation of acoustic echo elimination and electric echo elimination, wherein

Remote signaling or remote end input signal after the two filtration modules of self adaptation are eliminated according to the electric echo of far-end cache module output are carried out the adaptive-filtering that acoustic echo is estimated, when the filtering result restrains, calculate the acoustic echo signal according to the remote signaling after acoustic echo filter factor and the electric echo elimination, then the acoustic echo signal is outputed to first adder; First adder calculates the near end signal after thereby the difference of near end input signal and acoustics echo signal obtains the acoustic echo elimination, and this signal is outputed to the near-end cache module, and exports to far-end;

The two filtration modules of self adaptation according to the output of near-end cache module near end signal or the near end input signal of acoustic echo after eliminating carry out the adaptive-filtering that electric echo is estimated, when the filtering result restrains, calculate the electric echo signal according to the near end signal after electric echo filter factor and the acoustic echo elimination, then the electric echo signal is outputed to second adder; Second adder calculates the remote signaling after thereby the difference of remote end input signal and electric echo signal obtains the electric echo elimination, and this signal is outputed to the far-end cache module, and exports to near-end.

2, device as claimed in claim 1 is characterized in that, described device further comprises,

The near-end speech detection module is used to receive the near end signal after near end input signal or acoustic echo are eliminated, and when judging this signal and being voice, sends the near-end speech triggering signal to adaptive-filtering coefficient adjustment module;

The far-end speech detection module is used to receive the remote signaling after remote end input signal or electric echo are eliminated, and when judging this signal and being voice, sends the far-end speech triggering signal to adaptive-filtering coefficient adjustment module;

Adaptive-filtering coefficient adjustment module, when receiving only the near-end speech triggering signal that the near-end speech detection module sends, electric echo is finely tuned step-length send to the two filtration modules of self adaptation, the two filtration modules of self adaptation are adjusted current electric echo filter factor according to this step-length

When receiving only the far-end speech triggering signal that the far-end speech detection module sends, acoustic echo to be finely tuned step-length send to the two filtration modules of self adaptation, the two filtration modules of self adaptation are adjusted current acoustic echo filter factor according to this fine setting step-length,

When not receiving any triggering signal, whether inquire about current filtering result to the two filtration modules of self adaptation restrains, if convergence, just simultaneously electric echo is finely tuned step-length and acoustic echo fine setting step-length and send to the two filtration modules of self adaptation, the two filtration modules of self adaptation are adjusted current electric echo filter factor according to electric echo fine setting step-length, adjust current acoustic echo filter factor according to acoustic echo fine setting step-length, if do not restrain, just give self adaptation two filtration modules fast pacing length of electric echo and the fast pacing long hair of acoustic echo simultaneously, the two filtration modules of self adaptation are according to the current electric echo filter factor of the long adjustment of the fast pacing of electric echo, according to the current acoustic echo filter factor of the long adjustment of the fast pacing of acoustic echo

When receiving near-end speech triggering signal that near-end speech detection module and far-end speech detection module send and far-end speech triggering signal simultaneously, these two triggering signals are transmitted to the two filtration modules of self adaptation simultaneously do not adjust current electric echo and acoustic echo filter factor with the two filtration modules of notice self adaptation.

3, device as claimed in claim 2 is characterized in that, described near-end speech detection module comprises:

Near-end energy computing module, be used to receive the near end signal after near end input signal or acoustic echo are eliminated, calculate the energy of this signal, and determine the near-end speech threshold value according to the energy of this signal of quiet period, the energy of this signal during with this near-end speech threshold value and formal conversation outputs to the near-end speech judge module;

The near-end speech judge module, the energy of the near end signal after near end input signal when being used to receive near-end speech threshold value and formal conversation or acoustic echo are eliminated, relatively the size of the two is to judge whether the near end signal after near end input signal or acoustic echo are eliminated is voice, if voice send the near-end speech triggering signal for adaptive-filtering coefficient adjustment module.

4, device as claimed in claim 2 is characterized in that, described far-end speech detection module comprises:

Far-end energy computing module, be used to receive the remote signaling after remote end input signal or electric echo are eliminated, calculate the energy of this signal, and determine the far-end speech threshold value according to the energy of this signal of quiet period, the energy of this signal during with this far-end speech threshold value and formal conversation outputs to the far-end speech judge module;

The far-end speech judge module, the energy of the remote signaling after remote end input signal when being used to receive far-end speech threshold value and formal conversation or electric echo are eliminated, relatively the size of the two is to judge whether the remote signaling after remote end input signal or electric echo are eliminated is voice, if voice send the far-end speech triggering signal for adaptive-filtering coefficient adjustment module.

5, a kind of echo cancel method is characterized in that, this method comprises:

Remote signaling after near end signal after A, preservation acoustic echo are eliminated and electric echo are eliminated;

B, near end signal or near end input signal after remote signaling after eliminating according to the electric echo of having preserved respectively or remote end input signal and acoustic echo are eliminated are carried out the adaptive-filtering that acoustic echo is estimated and electric echo is estimated, and judge whether the filtering result that acoustic echo and electric echo are estimated restrains, if the filtering result that acoustic echo is estimated convergence, remote signaling and acoustic echo filter factor after eliminating according to electric echo calculate the acoustic echo signal, and the difference of then calculating current near end input signal and acoustics echo signal promptly obtains the near end signal after acoustic echo is eliminated; If the filtering result that electric echo is estimated convergence, near end signal and electric echo filter factor after eliminating according to acoustic echo calculate the electric echo signal, and the difference of then calculating current remote end input signal and electric echo signal promptly obtains the remote signaling after electric echo is eliminated; Otherwise, proceed the adaptive-filtering that acoustic echo or electric echo are estimated.

6, method as claimed in claim 5, it is characterized in that, further comprise before the described step B: judge whether near end signal after near end input signal or acoustic echo are eliminated and the remote signaling after remote end input signal or the electric echo elimination are voice, if the two all is voice, direct execution in step B; If the near end signal after near end input signal or acoustic echo are eliminated is voice, adjust current electric echo filter factor, then execution in step B; If the remote signaling after remote end input signal or electric echo are eliminated is voice, adjust current acoustic echo filter factor, then execution in step B; If the two is not voice, adjust current electric echo filter factor and acoustic echo filter factor simultaneously, then execution in step B.

7, method as claimed in claim 6 is characterized in that, described adjustment electric echo filter factor be by: $e_{n+1}=e_n+\frac{{\mathrm{\μ}}_e}{d_e+{\mathrm{PS}}_{\mathrm{out}}\left(n\right)}{R}_{\mathrm{out}}\left(n\right)S_{\mathrm{out}}\left(n\right)$ Realize, wherein, e _nBe current electric echo filter factor, e _N+1Be adjusted electric echo filter factor, μ _eFor adjusting step-length, d _eBe the constant between 0～1, PS _Out(n) be the energy of current near-end output signal, R _Out(n) be remote signaling after current electric echo is eliminated, S _Out(n) be near end signal after current acoustic echo is eliminated.

8, method as claimed in claim 6 is characterized in that, described adjustment acoustic echo filter factor be by $a_{n+1}=a_n+\frac{{\mathrm{\μ}}_a}{d_a+{\mathrm{PR}}_{\mathrm{out}}\left(n\right)}{S}_{\mathrm{out}}\left(n\right)R_{\mathrm{out}}\left(n\right)$ Realize, wherein, a _nBe current acoustic echo filter factor, a _N+lBe adjusted acoustic echo filter factor, μ _aFor adjusting step-length, d _aBe the constant between 0～1, PR _Out(n) be the energy of current far-end output signal, S _Out(n) be near end signal after current acoustic echo is eliminated, R _Out(n) be remote signaling after current electric echo is eliminated.

9, method as claimed in claim 6 is characterized in that, describedly judges that whether the near end signal after near end input signal or acoustic echo are eliminated is that the concrete steps of voice are:

Calculate the energy of the near end signal after the near end input signal of each time period in the quiet period after session is set up or acoustic echo are eliminated, with the maximum of each energy with greater than the product of 1 constant as the near-end speech threshold value; After the conversation beginning, calculate the energy of the near end signal after current near end input signal or acoustic echo are eliminated, with the energy that obtains and near-end speech threshold ratio, if the former, judges that the near end signal after current near end input signal or acoustic echo are eliminated is voice greater than the latter; Otherwise, judge that the near end signal after current near end input signal or acoustic echo are eliminated is not voice.

10, method as claimed in claim 6 is characterized in that, describedly judges that whether the remote signaling after remote end input signal or electric echo are eliminated is that the concrete steps of voice are:

Calculate the energy of the remote signaling after the remote signaling of each time period in the quiet period after session is set up or electric echo are eliminated, with the maximum of each energy with greater than the product of 1 constant as the far-end speech threshold value; After the conversation beginning, calculate the energy of the remote signaling after current remote end input signal or electric echo are eliminated, with the energy that obtains and far-end speech threshold ratio, if the former, judges that the remote signaling after current remote end input signal or electric echo are eliminated is voice greater than the latter; Otherwise, judge that the remote signaling after current remote end input signal or electric echo are eliminated is not voice.

Images