CN1946105A - Method and its system for eliminating stereophonic echo based on voice signal separate model - Google Patents
Method and its system for eliminating stereophonic echo based on voice signal separate model Download PDFInfo
- Publication number
- CN1946105A CN1946105A CN 200610123051 CN200610123051A CN1946105A CN 1946105 A CN1946105 A CN 1946105A CN 200610123051 CN200610123051 CN 200610123051 CN 200610123051 A CN200610123051 A CN 200610123051A CN 1946105 A CN1946105 A CN 1946105A
- Authority
- CN
- China
- Prior art keywords
- echo
- module
- voice signal
- impulse response
- microphone
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Circuit For Audible Band Transducer (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
Abstract
This invention provides a system for eliminating echo of stereos based on the separation of phone signals including an echo eliminator, a microphone and a loudspeaker, in which, said eliminator includes a periodic power detection module, a pulse response difference measurement module, a phone signal separation module, an intercommunication test module, an echo path estimation module and an adder. A method includes: estimating or renovating difference of near end room pulse response and echo path pulse response to separate single-path echo from multi-path echo when there is no intercommunication, getting the estimation of the echo path pulse response to further get the estimation of the separated single-path echo combining the source signal from a far end room, the single-path echo left from the separation of the received signal of a near end room microphone subtracts the corresponding estimation value so as to eliminate echos.
Description
Technical field
The present invention relates to seal the ECT echo cancellation technique in the voice environment, specifically be meant a kind of stereo echo removing method and system thereof based on the voice signal disjunctive model.
Background technology
Along with the development of multimedia communication technology, teleconference and car handsfree radio telephone system are universal day by day.In the closed voice environment that this class is made of loud speaker, room and microphone, exist acoustic echo to influence normally carrying out of communicating by letter, therefore the interference that echo eliminator reduces echo need be set.And,, adopt stereo being absolutely necessary in order to strengthen participant's sensation on the spot in person along with the raising of people to the fidelity requirement, and still, because increasing of the number of loud speaker and microphone produces multi-path echo thus and eliminates problem.Compare with the elimination of single channel echo, the multi-path echo that with the stereo case is representative is eliminated because the strong correlation between the input signal exists separating of echo eliminator not unique, it is very big to lack of proper care, the essential problem of impulse response that promptly can't the unique identification near-end room can not satisfy application requirements well.It all is to eliminate at the single channel echo that existing echo is eliminated patented technology many, uses traditional sef-adapting filter to obtain reasonable effect, and the conventional method that still direct situation with single channel expands to multichannel can not satisfy actual application demand.Application number is that 200610034214.6 Chinese invention patent application discloses a kind of " based on the multi-path echo eliminating system of SIMO channel blind discrimination ", this is a technology of eliminating at multi-path echo, and the main thought of this technology is blind signal separation technology to be applied to multi-path echo eliminate model; But this technology is difficult to overcome the defective that operand is huge, arithmetic speed is slower of blind signal separation technology itself, in this technology, need the initial construction matrix of using, follow-up structural matrix etc. all to take great memory space, also can cause hardware cost higher.
Summary of the invention
The objective of the invention is to overcome the shortcoming and defect of above-mentioned prior art, provide a kind of simple in structure, operand is little, realization, fast convergence rate, imbalance real-time, that be easy to hardware are little, residual echo is few, the measured stereo echo based on the voice signal disjunctive model of Speech Communication matter is eliminated system.
Another object of the present invention is to provide a kind of stereo echo removing method based on the voice signal disjunctive model.
Purpose of the present invention is achieved through the following technical solutions: a kind of stereo echo based on the voice signal disjunctive model is eliminated system, comprises echo eliminator, microphone, loud speaker; It is right that near-end microphone and far-end loudspeaker connect into one to one, and the far-end microphone also is linked to be rightly with the near-end loud speaker one to one, and each is to connecting an echo eliminator between microphone, the loud speaker; It is characterized in that: described echo eliminator comprises the cycle power detection module, impulse response difference measurement module, the voice signal separation module, the intercommunication detection module, the echo path estimation module, adder, described cycle power detection module and impulse response difference measurement module, the voice signal separation module, adder connects successively, the intercommunication detection module links to each other with microphone with the near-end loud speaker by input, be connected with the echo path estimation module with the voice signal separation module respectively by output, described voice signal separation module is connected with the near-end microphone by input, also link to each other with adder by output, described adder is connected with corresponding far-end loudspeaker by output.
Described cycle power detection module links to each other with impulse response difference measurement module, and the former exports triggering signal and gives the latter.Described impulse response difference measurement module comprises an acoustic pulse generator, the tone burst generator L520279 that provides as EHSY Lab; One of them is connected acoustic pulse generator and two near-end loud speakers; Be connected by adaptive frequency domain filter and adder between two near-end microphones.Acoustic pulse generator sends ping and gives the loud speaker that links to each other, and measures the impulse response coefficient of variation of the acoustic path between this loud speaker loca to two microphone.When acoustic pulse generator links to each other with another loud speaker, then measure the impulse response coefficient of variation of the acoustic path between another loud speaker loca to two microphone.Described adaptive frequency domain filter can realize that the value of the value impulse response coefficient of variation of filter is exported to the voice signal separation module during stable state by the frequency domain least mean square algorithm.
Described voice signal separation module comprises two Fourier transform module, four multipliers, two adders and two inverse-Fourier transform modules; Fourier transform module connects a multiplier, two-way separately again, and wherein one the tunnel links to each other with first adder, is connected to an inverse-Fourier variation module; Another road links to each other with next multiplier, links to each other with second adder again; Second half of this structure also is similar: another Fourier changes and also links to each other with a multiplier, two-way equally separately, and one the tunnel links to each other with second adder, is connected to an inverse-Fourier variation module; Another Lu Yuyi multiplier links to each other, and links to each other with first adder again.Top four multipliers mentioning all are connected with impulse response difference detection module.
Described echo path estimation module can be realized by the time-domain adaptive filter, specifically can adopt normalized least mean square algorithm.
A kind of stereo echo removing method of being realized by said system based on the voice signal disjunctive model comprises the steps:
(1) at first impulse response difference detection module when the acoustical power in detecting the room is lower than threshold value, carries out the measurement of impulse response difference parameter by triggering cycle time.
(2) the impulse response coefficient of variation is admitted to the voice signal separation module as output, prepares for next step calculates; The received signal of near-end microphone is that the echo behind the aliasing also is used as the input of voice signal separation module, thereby isolates the echo that the wall scroll echo path is produced; Be converted to the single channel echo through this step multi-path echo.
(3) far-end room microphone received signal is admitted to the echo path estimation module, and the echo that while wall scroll echo path is produced also is used as input and sends into, and finishes the elimination of single channel echo after sef-adapting filter is handled.
(4) as there being both sides' intercommunication, be that near-end and far-end room all exist the teller, then this moment voice signal separation module and echo path estimation module the computing meeting temporarily stop: both sides' intercommunication detection module is by triggering cycle time, and the correlation between the speaker output signal (being far-end room microphone received signal) by detecting near-end room and the received signal of near-end room microphone judges whether the generation of both sides' intercommunication.If the generation that both sides' intercommunication arranged then output order stops computing for the voice signal separation module.
(5) the impulse response coefficient of variation in the impulse response difference measurement module comprises Δ H
12, Δ H
21With the F parameter, Δ H
12With Δ H
21Directly measured by adaptive frequency domain filter, the F parameter is by Δ H
12With Δ H
21Calculate.Δ H
12, Δ H
21All be admitted to the voice signal separation module with the F parameter.The received signal of near-end microphone is that the echo behind the aliasing is y
1And y
2, as shown in Figure 1, its corresponding Fourier transform is Y
1And Y
2, being expressed as of the single channel echo frequency domain after the separation
With
Y through the corresponding time domain of inverse-Fourier transform
11And y
22The intercommunication detection module is finished the work that intercommunication is judged by the unit period Time Triggered.Because both sides' intercommunication is that near-end and far-end are when the someone talks simultaneously, correlation between near-end loudspeaker signal and the microphone received signal can reduce, so can judge whether the generation of both sides' intercommunication by estimating correlation between the two, the available coefficient correlation of the quantification of correlation is represented.
Described adaptive frequency domain filter, time-domain adaptive filter, Fourier transform module and inverse-Fourier transform module, intercommunication detection module, adder, multiplier, cycle power detection module, echo path estimation module etc. all can be realized by general dsp chip, as TMS320lf2407 and the TMS320C55x of Ti (Texas Instruments), the DSP56000 of Motorola Inc. and DSP96000 etc.
Action principle of the present invention is: the voice signal separation module is applied to multi-path echo eliminates, mainly comprise three processes: (1) is measured by impulse response difference detection module and is obtained near-end room impulse response coefficient of variation; Under the situation that does not have both sides' intercommunication, estimate or renewal near-end room echo path impulse response; (2) the voice signal separation module is separated the single channel echo from multi-path echo on the basis of known near-end room impulse response coefficient of variation; (3) pass to the source signal of near-end after receiving in conjunction with far-end room microphone, thereby obtain the estimation of echo path impulse response respectively, and then the estimation of the single channel echo that obtains separating; Isolated single channel echo deducts its estimated value, thereby realizes the echo elimination.
Signal in this specification is all digital signal, for simplicity, is that the variable of independent variable will be no longer specially be indicated independent variable with moment n.
The difference of room impulse response is based between the impulse response of same sound source to two microphone in the process 1 following relation
h
12＝Δh
12*h
11 (1)
h
21＝Δh
21*h
22 (2)
Δ h
IjExpression h
IiAnd h
IjMeasurement difference between these two impulse responses
(i，j＝1，2；i≠j)。
As shown in Figure 1, the microphone received signal is represented by following formula
y
1＝h
11*x
1+h
21*x
2 (3)
y
2＝h
12*x
1+h
22*x
2 (4)
For convenient, can remember
y
11＝h
11*x
1 (5)
y
22＝h
22*x
2 (6)
y
11, y
22Can think y
1, y
2In respectively by echo path h
11And h
22The echo component that produces.
By formula (1)～(6), can derive following relation:
Wherein
Y
11And Y
22Corresponding y
11And y
22Fourier change.
Process 2 is carried out by the voice signal separation module.At first to microphone received signal y
1And y
2Carry out Fourier transform, obtain Y respectively
1And Y
2, and finish the work of formula (7): the calculating of formula (8) will be placed in the impulse response difference measurement module to be carried out.
Process 3 is carried out by the echo path estimation module, utilizes normalized least mean square algorithm, and this algorithm has been widely used in the hardware realization that actual unipath echo is eliminated, and algorithm is as follows:
I=1 wherein, 2,
Be h
Ii(n) estimated value,
For
Vector representation.
The job step that intercommunication is judged is:
(1) by down trigger cycle time, the intercommunication detection module calculates near-end loudspeaker signal and wheat
Coefficient correlation ξ between the gram wind received signal (x, y);
(2) judge coefficient correlation ξ and threshold values c
1Magnitude relationship, if | ξ (x, y) |＞c
1, thinking does not have both sides' intercommunication; Otherwise think and have both sides' intercommunication.
The job step of cycle power detection module is:
(1) by triggering cycle time, the cycle power detection module is calculated the power of microphone received signal.
(2) judgement factor and threshold values c
2Magnitude relationship, if Q＜c
2, then think and to carry out new impulse response difference and detect this moment.
Basically, for two microphones that are placed in the same room, the convolution that same sound source can be used the impulse response of another microphone and the difference between the two to the impulse response of one of them is represented, thereby this relation is used to the voice signal separation module can separate the echo-signal of having mixed, the echo of therefrom isolating individual paths and being produced, thereby the echo that stereo echo elimination problem is converted into single channel is eliminated problem, the echo of single channel is eliminated problem has essence than multi-path echo elimination problem difference, and it is simple a lot, single channel echo eliminator technology is comparatively ripe through years of development, is well used in reality.
The present invention has following advantage and effect with respect to prior art:
(1) echo is eliminated effective; Utilize that the present invention can avoid fundamentally that filter imbalance in traditional stereo (multichannel) echo eliminator is big, echo disturbs many etc. problems, than traditional stereo echo arrester, the present invention can reduce imbalance 5～8dB, has voice communication quality preferably.
(2) realize easily; Operand of the present invention is little, and is real-time, need not to take bigger memory space, is easy to utilize hardware to realize that application cost is lower; And have fast convergence rate, imbalance is little, residual echo is few advantage, a practical application effect ideal.
(3) applied widely; Because eliminate and the not difference of essence of stereo echo elimination more than the multi-path echo of two-way, so the also as easy as rolling off a log situation about being generalized to more than two-way of stereo echo technology for eliminating of the present invention is used more convenient.
Description of drawings
Fig. 1 utilizes the stereo echo that the present invention is based on the voice signal disjunctive model to eliminate the principle schematic of systemic effect.
Fig. 2 is the structural representation that the present invention is based on the stereo echo elimination system of voice signal disjunctive model.
The internal structure schematic diagram of Fig. 3 impulse response difference shown in Figure 2 detection module.
The internal structure schematic diagram of Fig. 4 voice signal separation module shown in Figure 2.
Fig. 5 is the workflow diagram of the inventive method.
Embodiment
The present invention is described in further detail below in conjunction with embodiment and accompanying drawing, but of the present invention
Execution mode is not limited thereto.
Embodiment
Fig. 1～Fig. 4 shows concrete structure of the present invention, as shown in Figure 1, this stereo echo elimination system based on the voice signal disjunctive model comprises echo eliminator, microphone, loud speaker, it is right that near-end microphone and far-end loudspeaker connect into one to one, and the far-end microphone also with the near-end loud speaker be linked to be one to one right.Each is to connecting an echo eliminator between microphone, the loud speaker.
As shown in Figure 2, described echo eliminator comprises cycle power detection module, impulse response difference detection module, voice signal separation module, intercommunication detection module, echo path estimation module, adder.Described cycle power detection module is connected with impulse response difference detection module, and the latter is connected with the voice signal separation module.The intercommunication detection module links to each other with two near-end microphones with two near-end loud speakers by parallel input, links to each other with the echo path estimation module with the voice signal separation module respectively by two parallel output terminals again.Two near-end loud speakers also are connected with the echo path estimation module by output.Voice signal separation module, echo path estimation module, adder link to each other successively, and the voice signal separation module also links to each other with two adders respectively by two parallel output terminals.Two adders link to each other with two loud speakers of far-end respectively by output.
Fig. 3 shows the structure of described impulse response difference detection module; As shown in Figure 3, impulse response difference measurement module comprises an acoustic pulse generator, the tone burst generator L520279 that provides as EHSY Lab; One of them is connected acoustic pulse generator and two near-end loud speakers; Be connected by adaptive frequency domain filter and adder between two near-end microphones.
Fig. 4 shows the structure of described voice signal separation module; As shown in Figure 4, this voice signal separation module comprises two Fourier transform module, four multipliers, two adders and two inverse-Fourier transform modules; Fourier transform module connects a multiplier, two-way separately again, and wherein one the tunnel links to each other with first adder, is connected to an inverse-Fourier variation module; Another road links to each other with next multiplier, links to each other with second adder again; Second half of this structure also is similar: another Fourier changes and also links to each other with a multiplier, two-way equally separately, and one the tunnel links to each other with second adder, is connected to an inverse-Fourier variation module; Another Lu Yuyi multiplier links to each other, and links to each other with first adder again.All being connected of top four multipliers mentioning with impulse response difference detection module.
Impulse response difference detection module shown in Figure 3 is finished frequency domain Δ h
IjMeasurement, Δ h
IjExpression h
IiAnd h
IjMeasurement difference (i, j=1,2 between these two impulse responses; I ≠ j), its meaning such as formula (1) (2) expression
h
12＝Δh
12*h
11 (1)
h
21＝Δh
21*h
22 (2)
Before starting echo cancellation system, at first use adaptive frequency domain filter, adopt ping to trigger, measure Δ H respectively
12With Δ H
21, and finish following calculating
This module is output as Δ H
12, Δ H
21And F, they are placed into the memory cell of parameter correspondence respectively.
Voice signal separation module shown in Figure 4 is at first to microphone received signal y
1And y
2Carry out Fourier transform, obtain Y respectively
1And Y
2, and mainly finish the work of formula (7):
In the echo path estimation module, mainly finish the work of formula (9) and (10):
μ is the iteration step length of setting in the formula (10).In the echo path estimation module
Meaning be echo path h
IiCharacteristic estimating,
For
Corresponding vector representation.This module adopts adaptive filter method to estimate and renewal
It upgrades iterative formula is above-mentioned formula (10).E in the formula (9)
i(n) be to have eliminated the final output that the echo back echo is eliminated.
The intercommunication detection module is to detect whether intercommunication of far-end room and near-end room speaker, and its input is remote source signal x
1, x
2Union variable x and and the single channel echo-signal y that separates
1, y
2Union variable y, output is the result who judges.
As shown in Figure 4, the voice signal separation module is at first to y
1And y
2Utilization Fourier transform module FFT handles and obtains Y
1And Y
2, Y
2And Y
1Multiply each other with F, more respectively with Δ H
12, Δ H
21Multiply each other, and get negative after, add F * Y more respectively
1And F * Y
2, then obtain Y respectively
11And Y
22At last with Y
11And Y
22Carry out inverse-Fourier by the inverse-Fourier transform module and change, thereby can obtain their y in the time domain correspondence
11And y
22
H wherein
11, h
12, h
21And h
22The unit impulse response vector of representing four echo path correspondences respectively in the near-end room, y
1=h
11* x
1+ h
21* x
2, y
2=h
12* x
1+ h
22* x
2Represent that respectively microphone M1 and M2 in the near-end room examine the mixed signal of picking up.y
11=h
11* x
1, y
22=h
22* x
2Can think y
1, y
2In respectively by wall scroll echo path h
11And h
22The echo component that produces.
Eliminate based on the stereo echo of voice signal disjunctive model system method workflow as shown in Figure 5, comprise that parameter update separates with echo and eliminates, parameter update and echo are eliminated parallel carrying out, parameter update is meant under the situation that does not have both sides' intercommunication, estimates or upgrade that the difference of the impulse response that the same microphone of near-end room receives is Δ H
12, Δ H
21With the F parameter, and the near-end echo path impulse response in the self adaptation echo cancellation module of unipath is h
11And h
22,, then do not upgrade if there is intercommunication in both sides; Described echo is eliminated and is meant that in known near-end room impulse response difference be Δ h
12With Δ h
21On the basis of parameter and F parameter, from multi-path echo, earlier the single channel echo is separated the source signal x that comes in conjunction with the far-end room
1And x
2, adopt adaptive filter algorithm to estimate the estimation of echo path impulse response respectively
With
And then obtain y in the received signal of near-end room microphone respectively
11And y
22, the received signal of near-end room microphone is through y remaining after the separating treatment
11And y
22Deduct the estimation of echo-signal respectively
With
Thereby realize the echo elimination.
The course of work that echo is eliminated is:
(1) the Δ H in the paired pulses difference detection module
12, Δ H
21Carry out initialization with the F parameter, and to the estimation of the impulse response of self adaptation echo channel in the echo path estimation module
With
Carry out initialization;
(2) with Δ H
12, Δ H
21Send into the voice signal separation module with the F parameter, simultaneously far-end room microphone received signal x
1And x
2Also be transfused to echo path estimation module and intercommunication detection module, the voice signal separation module is handled the echo y that draws single channel
11And y
22, and with y
11And y
22Send the echo path estimation module to;
(3) the echo path estimation module receives x
1, x
2After, with x
1, x
2With
Convolution obtains y
11And y
22Estimated signal
With
(4)
Be sent to adder after the negate, respectively with the single channel echo-signal y of input summer
11And y
22Addition obtains corresponding e
1, e
2Eliminate clean signal afterwards for echo, realize the echo elimination;
(5) return step (2), proceed echo and eliminate.
The step of parameter update comprises:
(6) according to far-end microphone received signal (being the near-end speaker output signal) x
1, x
2With near-end microphone received signal y
1, y
2Intercommunication detection module down trigger cycle time is carried out intercommunication and is judged, and the result that output is judged gives the voice signal separation module, if there is both sides' intercommunication, then do not upgrade each parameter, and insert break period next time, if there is no both sides' intercommunication, then trigger step (7) and step (8), and insert break period next time;
(7) renewal that is subjected to the intercommunication detection module triggers, and the voice signal separation module is to the y of input
1And y
2Handle, result sends the update calculation module to;
(8) renewal that is subjected to the intercommunication detection module triggers, and the echo path estimation module is to wherein
With
Upgrade.
(9) be subjected to the triggering of cycle power detection module, impulse response difference detection module is to Δ H wherein
12, Δ H
21Upgrade with the F parameter.
The job step that intercommunication is judged is:
(1) by down trigger cycle time, the intercommunication detection module calculates x=[x
1(n), x
1(n-1) ..., x
1(n-m+1), x
2(n), x
2(n-1) ..., x
2And y=[y (n-m+1)]
1(n), y
1(n-1) ..., y
1(n-m+1) y
2(n), y
2(n-1) ..., y
2Coefficient correlation ξ (n-m+1)] (x, y);
(2) judge coefficient correlation ξ and threshold values c
1Magnitude relationship, c wherein
1Be positive number, get c
1=0.2～0.5, if | ξ (x, y) |＞c
1, thinking does not have both sides' intercommunication; Otherwise think and have both sides' intercommunication.
The job step of cycle power detection module is:
(1) by triggering cycle time, the cycle power detection module is calculated the power of microphone received signal
The pairing sampling number of voice length of the corresponding 50ms of p.
(2) judgement factor and threshold values c
2Magnitude relationship, c wherein
2Be positive number, get c
1=0.1～0.3,
If Q＜c
1, then think and to carry out new impulse response difference and detect this moment.
Disturb if want the echo of eliminating near-end room, owing to far away, near-end are relative, so its structure in like manner.
As mentioned above, just can realize the present invention preferably.
Claims (8)
1, a kind of stereo echo based on the voice signal disjunctive model is eliminated system, comprises echo eliminator, microphone, loud speaker; It is right that near-end microphone and far-end loudspeaker connect into one to one, and the far-end microphone also is linked to be rightly with the near-end loud speaker one to one, and each is to connecting an echo eliminator between microphone, the loud speaker; It is characterized in that: described echo eliminator comprises the cycle power detection module, impulse response difference measurement module, the voice signal separation module, the intercommunication detection module, the echo path estimation module, adder, described cycle power detection module is connected with impulse response difference measurement module, described impulse response difference measurement module, the voice signal separation module, adder connects successively, the intercommunication detection module is connected with the voice signal separation module, described voice signal separation module is connected with the near-end microphone respectively by parallel input, described adder, the echo path estimation module is connected with the voice signal separation module by input, is connected with corresponding far-end loudspeaker by output.
2, the stereo echo based on the voice signal disjunctive model according to claim 1 is eliminated system, it is characterized in that: described impulse response difference measurement module comprises an acoustic pulse generator, and one of them is connected acoustic pulse generator and two near-end loud speakers; Be connected by adaptive frequency domain filter and adder between two near-end microphones.
3, the stereo echo based on the voice signal disjunctive model according to claim 1 is eliminated system, and it is characterized in that: described voice signal separation module comprises two Fourier transform module, four multipliers, two adders and two inverse-Fourier transform modules; Fourier transform module connects a multiplier, two-way separately again, and wherein one the tunnel links to each other with first adder, is connected to an inverse-Fourier variation module; Another road links to each other with next multiplier, links to each other with second adder again; Another Fourier changes and also links to each other with a multiplier, two-way equally separately, and one the tunnel links to each other with second adder, is connected to an inverse-Fourier variation module; Another Lu Yuyi multiplier links to each other, and links to each other with first adder again; Described four multipliers all are connected with impulse response difference detection module.
4, the stereo echo based on the voice signal disjunctive model according to claim 1 is eliminated system, and it is characterized in that: described echo path estimation module is the time-domain adaptive filter.
5, a kind of stereo echo removing method of being realized by the described system of claim 1～4 based on the voice signal disjunctive model comprises the steps:
(1) measures acquisition near-end room impulse response coefficient of variation by impulse response difference detection module; Under the situation that does not have both sides' intercommunication, estimate or renewal near-end room echo path impulse response;
(2) the voice signal separation module is separated the single channel echo from multi-path echo on the basis of known near-end room impulse response coefficient of variation;
(3) source signal of coming in conjunction with the far-end room, thus the estimation of echo path impulse response obtained respectively, and then the estimation of the single channel echo that obtains separating; Isolated single channel echo deducts its estimated value, thereby realizes the echo elimination.
6, the stereo echo removing method based on the voice signal disjunctive model according to claim 5 is characterized in that:
(1) at first impulse response difference detection module when the acoustical power in detecting the room is lower than threshold value, carries out the measurement of impulse response difference parameter by triggering cycle time;
(2) the pulse differences coefficient is admitted to the voice signal separation module as output, prepares for next step calculates; The received signal of near-end microphone also is used as the input of voice signal separation module, thereby isolates the echo that the wall scroll echo path is produced; Be converted to the single channel echo through this step multi-path echo;
(3) far-end room microphone received signal is admitted to the echo path estimation module, and the echo that while wall scroll echo path is produced also is used as input and sends into, and finishes the elimination of single channel echo after sef-adapting filter is handled.
7, the stereo echo removing method based on the voice signal disjunctive model according to claim 6 is characterized in that: in step (2), described impulse response coefficient of variation comprises Δ H
12, Δ H
21With the F parameter, Δ H
12With Δ H
21Directly measured by adaptive frequency domain filter, the F parameter is by Δ H
12With Δ H
21Calculate.Δ H
12, Δ H
21All be admitted to the voice signal separation module with the F parameter.
8, the stereo echo removing method based on the voice signal disjunctive model according to claim 7 is characterized in that:
(1) the impulse response coefficient of variation Δ H in the paired pulses response difference detection module
12, Δ H
21Carry out initialization with the F parameter, and to the estimation of the impulse response of self adaptation echo channel in the echo path estimation module
With
Carry out initialization;
(2) with Δ H
12, Δ H
21Send into the voice signal separation module with the F parameter, simultaneously far-end room microphone received signal x
1And x
2Also be transfused to echo path estimation module and intercommunication detection module, the voice signal separation module is handled the echo y that draws single channel
11And y
22, and with y
11And y
22Send the echo path estimation module to;
(3) the echo path estimation module receives x
1, x
2After, with x
1, x
2Corresponding with
Convolution, the received signal that obtains the near-end room microphone is carried out the estimation of the signal after the separating treatment
(4)
Be sent to adder after the negate, respectively with the single channel echo-signal y of input summer
11And y
22Addition obtains corresponding e
1, e
2Eliminate clean signal afterwards for echo, realize the echo elimination.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2006101230519A CN100559817C (en) | 2006-10-27 | 2006-10-27 | Stereo echo removing method and system thereof based on the voice signal disjunctive model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2006101230519A CN100559817C (en) | 2006-10-27 | 2006-10-27 | Stereo echo removing method and system thereof based on the voice signal disjunctive model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1946105A true CN1946105A (en) | 2007-04-11 |
| CN100559817C CN100559817C (en) | 2009-11-11 |
Family
ID=38045315
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2006101230519A Expired - Fee Related CN100559817C (en) | 2006-10-27 | 2006-10-27 | Stereo echo removing method and system thereof based on the voice signal disjunctive model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100559817C (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101820302A (en) * | 2009-02-27 | 2010-09-01 | 比亚迪股份有限公司 | Device and method for canceling echo |
| CN105895114A (en) * | 2016-03-22 | 2016-08-24 | 南京大学 | Pulse-response-based room sound propagation path separation method |
| CN107005268A (en) * | 2014-12-15 | 2017-08-01 | 三菱电机株式会社 | Echo eliminator and method for echo cancellation |
| CN109756822A (en) * | 2017-11-01 | 2019-05-14 | 通用汽车环球科技运作有限责任公司 | The effective Echo cancellation estimated using transmission function |
-
2006
- 2006-10-27 CN CNB2006101230519A patent/CN100559817C/en not_active Expired - Fee Related
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101820302A (en) * | 2009-02-27 | 2010-09-01 | 比亚迪股份有限公司 | Device and method for canceling echo |
| CN101820302B (en) * | 2009-02-27 | 2013-10-30 | 比亚迪股份有限公司 | Device and method for canceling echo |
| CN107005268A (en) * | 2014-12-15 | 2017-08-01 | 三菱电机株式会社 | Echo eliminator and method for echo cancellation |
| CN107005268B (en) * | 2014-12-15 | 2020-06-16 | 三菱电机株式会社 | Echo cancellation device and echo cancellation method |
| CN105895114A (en) * | 2016-03-22 | 2016-08-24 | 南京大学 | Pulse-response-based room sound propagation path separation method |
| CN105895114B (en) * | 2016-03-22 | 2019-09-27 | 南京大学 | A kind of room acoustic propagation path separation method based on impulse response |
| CN109756822A (en) * | 2017-11-01 | 2019-05-14 | 通用汽车环球科技运作有限责任公司 | The effective Echo cancellation estimated using transmission function |
| CN109756822B (en) * | 2017-11-01 | 2021-04-06 | 通用汽车环球科技运作有限责任公司 | Efficient echo cancellation using transfer function estimation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100559817C (en) | 2009-11-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI398855B (en) | Multiple microphone voice activity detector | |
| JP5675848B2 (en) | Adaptive noise suppression by level cue | |
| CN103426433B (en) | Noise cancellation method | |
| CN102387272B (en) | Restraining method for residual echo in echo cancellation system | |
| CN105244036A (en) | Microphone speech enhancement method and microphone speech enhancement device | |
| US20080107281A1 (en) | Acoustic echo canceller system | |
| CN108447496B (en) | Speech enhancement method and device based on microphone array | |
| CN107371079B (en) | A kind of the diamylose gram noise reduction system and noise-reduction method of earphone | |
| CN1926911A (en) | Howling suppression device, program, integrated circuit, and howling suppression method | |
| CN1967658A (en) | Small scale microphone array speech enhancement system and method | |
| CN101315772A (en) | Speech reverberation eliminating method based on Wiener filtering | |
| CN109068012B (en) | Double-end call detection method for audio conference system | |
| CN105635500B (en) | The inhibition system and method for dual microphone echo and noise | |
| US9532138B1 (en) | Systems and methods for suppressing audio noise in a communication system | |
| WO2008011319A3 (en) | Method and system for near-end detection | |
| TW202117706A (en) | Apparatus and method for multiple-microphone speech enhancement | |
| KR102429152B1 (en) | Deep learning voice extraction and noise reduction method by fusion of bone vibration sensor and microphone signal | |
| CN1946105A (en) | Method and its system for eliminating stereophonic echo based on voice signal separate model | |
| JP2016506673A5 (en) | ||
| CN108962276B (en) | Voice separation method and device | |
| CN1949795A (en) | Echo canceller and microphone apparatus | |
| CN200969609Y (en) | Voice signal separation model based stereo echo canceling system | |
| CN102739286B (en) | Echo cancellation method used in communication system | |
| CN111225317B (en) | Echo cancellation method | |
| US20080288253A1 (en) | Automatic speech recognition method and apparatus, using non-linear envelope detection of signal power spectra |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20091111 Termination date: 20161027 |