CN102065190B - Method and device for eliminating echo - Google Patents

Abstract

The invention discloses a method and a device for eliminating echo. The method comprises the steps of: determining the state of each subband adaptive filter respectively according to the cross correlation coefficient of an input signal on the near end of each subband and an echo estimation signal; when one end of the subband adaptive filter is in a speaking state, replacing the residual signal of the subband adaptive filter with comfortable noise of the subband, and outputting the replaced signal; and when both ends of the subband adaptive filter are in the speaking state, outputting the residual signal of the subband adaptive filter. In the invention, the signal processing characteristic of the subband can be fully utilized so that the residual echo can be suppressed more effectively. In addition, the suppression on the residual echo while both ends are in the speaking state and the protection on the voice of a person speaking on one end are enhanced, and the integral effect and the fluency of the system are improved.

Description

A kind of echo cancel method and device thereof

Technical field

The present invention relates to communication technical field, relate in particular to a kind of echo cancel method and device thereof.

Background technology

In voice communication system; after remote end input signal arrives local signal receiving equipment (for example telephone set); arrive receiver through the audio amplifier of local signal receiving equipment, room etc., in this process through regular meeting because the sound reflection echogenicity in the audio amplifier, room.And in order to eliminate echo, then need echo cancellation technology, monitoring echo signal, and echo signal is eliminated from voice signal.For a long time, echo elimination is a very challenging job, mainly due to:

(1) acoustic echo will directly or through the form with stack after the reflection of one or many enter microphone, and the result causes the last or end syllable of echo very long, and the impulse response of corresponding echo passage is also very long, generally has the hundreds of millisecond.

(2) sound spectrum of voice signal be non-flat forms with the diffusion, and adaptive algorithm commonly used is relevant with the statistical property of input signal, the diffusion of eigenwert can make that for example the self-adaptation convergence process of NLMS (Normalized Least Mean Square, normalization minimum mean-square) algorithm is slack-off in the autocorrelation matrix of voice signal.

(3) the characteristic right and wrong of acoustic echo passage stably, the movement of other people or object all can make the impulse response of acoustic echo great changes will take place in speaker or the room, the quick variation characteristic of echo passage requires AEC (Acoustic Echo Cancellation, the acoustic echo elimination) speed of convergence should be fast as far as possible, and possess well quick tracking power.

(4) in real system, because non-linear etc. the influence of audio collection, playback equipment, consequent nonlinear echo can not be eliminated by sef-adapting filter; Because influence such as neighbourhood noise, adaptive filter coefficient also may can not ideal matching with actual room impulse response after convergence, and clean residual echo is not eliminated in generation; These nonlinear echoes and linear residual echo need add non-linear post-processing module after sef-adapting filter, do further processing, suppress residual echo, promote the effect of echo cancelling system integral body.

Fig. 1 is the synoptic diagram that a kind of echo is eliminated post-processing algorithm, being described as follows of each symbol among the figure:

X: remote end input signal;

The actual echo signal that y:x forms through the room;

V: local speaker's sound and ground unrest;

D: the near end input signal of Echo Canceller;

The estimated echo that computing obtains through sef-adapting filter;

E: the residual signals of filtering output;

e _Post: through the output signal of post-processing algorithm;

H: actual room impulse response;

Adaptive filter coefficient, the i.e. estimation of h.

As shown in Figure 1, post-processing algorithm is to have added VAD (Voice Activity Detector, voice-activated detecting device), CNG (Comfort Noise Generation, comfort noise produces) module on the basis of sef-adapting filter.The VAD module is to detect the signal that the near-end microphone receives whether voice signal is arranged, and the signal that microphone receives in the system comprises the echo of passing through the room, the near-end speaker's of remote signaling sound, near-end environmental background noise etc.; By VAD detect to determine far-end, near-end all nobody speak, be that signal that microphone receives is when only containing ground unrest, estimate the spectrum signature (being LPC (Linear Prediction Coefficient, linear predictor coefficient), energy gain) of ground unrest.Since ground unrest is general change slower, the spectrum signature cycle that VAD estimates be updated to the CNG module.CNG produces one section white-noise excitation according to the background noise spectrum feature that VAD provides, and the prediction error filter of being made up of LPC coefficient and energy gain generates comfort noise; When echo cancelling system detects single-ended speaking during state by the both-end detection module, NLP (None-Linear Post-processing, non-linear aftertreatment) comfort noise that generates with CNG substitutes the residual echo signal of sef-adapting filter output, does not eliminate clean residual echo fully in order to avoid far-end is heard; Certainly when both-end is talked (near-end has the people to speak), NLP is delivered directly to far-end with the signal that contains near-end speaker sound of sef-adapting filter output.

At present, adaptive filter algorithm has based on full range band, subband, based on the post-processing algorithm of subband with compare advantage specific as follows based on the post-processing algorithm of full range band:

(1) based on the adaptive filter algorithm of subband advantages such as fast convergence rate, computation complexity be low is arranged and adopted widely, be difficult to expand to subband based on the post-processing algorithm of full range band;

(2) when both-end is talked, the signal of sef-adapting filter output comprises near-end speaker's sound and residual echo signal, and can't handle residual echo at this moment based on the post-processing algorithm of full range band, and far-end can be heard residual echo;

(3) in some actual environment, because equipment exists than severe nonlinear, or neighbourhood noise is more serious, sef-adapting filter can not well be restrained, cause residual echo apparent in view, and when the near-end speaker sound suitable with echo amplitude occurring, the Nonlinear Processing algorithm is difficult to distinguish the single-ended state of speaking and speaking with both-end, erroneous judgement can cause single-ended in a minute the time residual echo can not get suppressing preferably or both-end to cut signal to noise ratio when speaking more serious.

Based on the adaptive filter algorithm of subband, owing to fast convergence rate, the low echo elimination field that is widely used in of computation complexity.But because the existence of the residual echo of nonlinear echo and linearity needs non-linear post-processing algorithm do further processing to the signal of subband sef-adapting filter output, suppress residual echo.Simultaneously, having only far-end when speech, suppress to insert the comfort noise consistent with the near-end background noise frequency spectrum after the residual echo, when making single-ended speech owing to residual echo transition inhibition causes the interrupted problem of ground unrest to be eased.

The inventor finds that there is following problem at least in prior art in realizing process of the present invention:

In the legacy system, with the synthetic full range band signal of sub-band adaptive filtering output signal, by handling based on the post-processing algorithm of full range band, such post-processing algorithm does not take full advantage of the advantage of subband again, further improves the effect of aftertreatment.

Summary of the invention

The object of the present invention is to provide a kind of echo cancel method and device thereof, in order to realize echo elimination based on subband, for this reason, the present invention adopts following technical scheme:

A kind of echo cancel method may further comprise the steps:

According to the cross-correlation coefficient of near end input signal and the echo estimated signal of each subband, determine the residing state of each sub-band adaptive wave filter respectively;

When the subband sef-adapting filter is in single-ended speaking during state, replace the signal after the output replacement behind the residual signals of this sub-band adaptive wave filter with the comfort noise of this subband;

When the subband sef-adapting filter is in both-end and speaks state, export the residual signals of this sub-band adaptive wave filter;

Wherein, according to the cross-correlation coefficient of near end input signal and the echo estimated signal of subband, determine the residing state of sub-band adaptive wave filter, comprising:

When the cross-correlation coefficient of subband during more than or equal to first threshold, determine that the sub-band adaptive wave filter is in the single-ended state of speaking;

When the cross-correlation coefficient of subband is less than or equal to second threshold value, determine that the sub-band adaptive wave filter is in the both-end state of speaking;

For the subband of cross-correlation coefficient between second threshold value and first threshold, quantity or proportion according to the sub-band adaptive wave filter of designated state, or the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter;

Wherein, 0＜the second threshold value＜first threshold＜1.

In the said method, for the subband of cross-correlation coefficient between second threshold value and first threshold, according to the quantity of the sub-band adaptive wave filter of designated state, determine the residing state of this sub-band adaptive wave filter, be specially:

Surpass setting threshold if be in the quantity of the sub-band adaptive wave filter of the single-ended state of speaking, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps,

Surpass setting threshold if be in the speak quantity of sub-band adaptive wave filter of state of both-end, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

In the said method, for the subband of cross-correlation coefficient between second threshold value and first threshold, the ratio shared according to the sub-band adaptive wave filter of designated state determined the residing state of this sub-band adaptive wave filter, is specially:

Surpass setting threshold if be in the shared ratio of the sub-band adaptive wave filter quantity of the single-ended state of speaking, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps,

Surpass setting threshold if be in the both-end shared ratio of the sub-band adaptive wave filter quantity of state of speaking, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

In the said method, for the subband of cross-correlation coefficient between second threshold value and first threshold, according to the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter, be specially:

If the mean value of the described cross-correlation coefficient of each subband is greater than the 3rd threshold value, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking;

Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

In the said method, for the subband of cross-correlation coefficient between second threshold value and first threshold, according to the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter, be specially:

If the weighted mean value of the described cross-correlation coefficient of each subband is greater than the 3rd threshold value, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking;

Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

In the said method, cross-correlation coefficient greater than the corresponding weighted value of the subband of first threshold and cross-correlation coefficient less than the corresponding weighted value of the subband of second threshold value, greater than the weighted value of the subband of cross-correlation coefficient between second threshold value and first threshold.

In the said method, the weighted value of subband is:

${\mathrm{\λ}}_i=\frac{{\mathrm{\σ}}_{d_i}^2}{\underset{i-1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\σ}}_{d_i}^2}$

Wherein, N is number of sub-bands,

Energy for the subband near end input signal of Echo Canceller.

A kind of echo cancelling device comprises:

The state determination module is used for respectively the cross-correlation coefficient according near end input signal and the echo estimated signal of each subband, determines the residing state of each sub-band adaptive wave filter;

Output module is used for being in single-ended speaking during state when the subband sef-adapting filter, replaces the signal of exporting behind the residual signals of this sub-band adaptive wave filter after the replacement with the comfort noise of this subband; When the subband sef-adapting filter is in both-end and speaks state, export the residual signals of this sub-band adaptive wave filter;

Wherein, described state determination module specifically is used for, and when the cross-correlation coefficient of subband during more than or equal to first threshold, determines that the sub-band adaptive wave filter is in the single-ended state of speaking; When the cross-correlation coefficient of subband is less than or equal to second threshold value, determine that the sub-band adaptive wave filter is in the both-end state of speaking; For the subband of cross-correlation coefficient between second threshold value and first threshold, quantity or proportion according to the sub-band adaptive wave filter of designated state, or the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter; Wherein, 0＜the second threshold value＜first threshold＜1.

In the said apparatus, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, quantity according to the sub-band adaptive wave filter of designated state, when determining the residing state of this sub-band adaptive wave filter, surpass setting threshold if be in the quantity of the sub-band adaptive wave filter of the single-ended state of speaking, judge that then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps, surpass setting threshold if be in the speak quantity of sub-band adaptive wave filter of state of both-end, the sef-adapting filter of then judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise, judge that the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

In the said apparatus, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, the ratio shared according to the sub-band adaptive wave filter of designated state, when determining the residing state of this sub-band adaptive wave filter, surpass setting threshold if be in the shared ratio of the sub-band adaptive wave filter quantity of the single-ended state of speaking, judge that then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps, surpass setting threshold if be in the both-end shared ratio of the sub-band adaptive wave filter quantity of state of speaking, the sef-adapting filter of then judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise, judge that the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

In the said apparatus, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, mean value according to the described cross-correlation coefficient of each sub-band adaptive wave filter, when determining the residing state of this sub-band adaptive wave filter, if the mean value of the described cross-correlation coefficient of each subband, is then judged the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold greater than the 3rd threshold value and is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

In the said apparatus, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, mean value according to the described cross-correlation coefficient of each sub-band adaptive wave filter, when determining the residing state of this sub-band adaptive wave filter, if the weighted mean value of the described cross-correlation coefficient of each subband, is then judged the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold greater than the 3rd threshold value and is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

In the said apparatus, the employed cross-correlation coefficient of described state determination module greater than the corresponding weighted value of the subband of first threshold and cross-correlation coefficient less than the corresponding weighted value of the subband of second threshold value, greater than the weighted value of the subband of cross-correlation coefficient between second threshold value and first threshold.

Useful technique effect of the present invention comprises:

By respectively according to the cross-correlation coefficient of near end input signal and the echo estimated signal of each subband; determine the residing state of each sub-band adaptive wave filter; and according to the different different disposal routes of state use; namely only just carry out echo cancellation process during state single-ended speak; thereby take full advantage of the characteristic that subband signal is handled; make residual echo obtain more effective inhibition; strengthened simultaneously when both-end is spoken to the inhibition of residual echo and to the protection of local terminal speaker sound, the effect of entire system and fluency are promoted.

Description of drawings

Fig. 1 is the synoptic diagram that echo is eliminated post-processing algorithm in the prior art;

Fig. 2 eliminates schematic flow sheet for a kind of echo that the embodiment of the invention provides;

Fig. 3 eliminates schematic flow sheet for the another kind of echo that the embodiment of the invention provides;

Fig. 4 eliminates schematic flow sheet for the another kind of echo that the embodiment of the invention provides;

The structural representation of the echo cancellation devices that Fig. 5 provides for the embodiment of the invention.

Embodiment

In echo cancelling system, near end input signal d commonly used and estimated echo signal

Cross-correlation coefficient η characterize the degree of convergence of sef-adapting filter, namely when related coefficient η close to 1 the time, think the sef-adapting filter convergence better, i.e. estimated echo

Approach input signal d preferably; And when related coefficient η close to 0 the time, think that the wave filter convergence is undesirable or be in the state that both-end is spoken.

η can use following formula to represent:

$\mathrm{\η}=\frac{E\left[d\left(n\right)\hat{y}\left(n\right)\right]}{\sqrt{{\mathrm{\σ}}_d^2{\mathrm{\σ}}_{\hat{y}}^2}}\·\·\·\left[1\right]$

Wherein, the near end input signal of d (n) expression Echo Canceller,

The expression estimated echo that computing obtains through sef-adapting filter,

The energy of expression d (n),

Expression

Energy.

With

Computing formula can be as follows:

${\mathrm{\σ}}_d^2=(1-\mathrm{\β}){\mathrm{\σ}}_d^2+\mathrm{\β}|d\left(n\right){|}^2\·\·\·\left[2\right]$

${\mathrm{\σ}}_{\hat{y}}^2=(1-\mathrm{\β}){\mathrm{\σ}}_{\hat{y}}^2+\mathrm{\β}{\left|\hat{y}\left(n\right)\right|}^2\·\·\·\left[3\right]$

Based on above-mentioned near end input signal d and estimated echo signal

Cross-correlation coefficient η and the relation of the degree of convergence of sef-adapting filter, the embodiment of the invention is utilized this cross-correlation coefficient η to judge the residing state of wave filter, and this thought is generalized to subband in non-linear last handling process,, uses η that is _i(subscript i represents i subband) judges i the residing state of subband sef-adapting filter:

${\mathrm{\η}}_i=\frac{E\left[d_i^{*}\left(n\right){\hat{y}}_i\left(n\right)\right]}{\sqrt{{\mathrm{\σ}}_{d_i}^2{\mathrm{\σ}}_{{\hat{y}}_i}^2}},i=\mathrm{1,2},...,L,...,N\·\·\·\left[4\right]$

Wherein, N is number of sub-bands; d _i(n) the subband near end input signal of expression Echo Canceller,

The expression estimated echo that computing obtains through the sub-band adaptive wave filter, subband signal d _i(n) and

May be plural number; Subscript * is conjugate operation;

Expression d _i(n) energy,

Expression

Energy, the subband signal energy

With

Calculating provided by following formula:

${\mathrm{\σ}}_{d_i}^2=(1-\mathrm{\β}){\mathrm{\σ}}_{d_i}^2+\mathrm{\β}|d_i\left(n\right){|}^2\·\·\·\left[5\right]$

${\mathrm{\σ}}_{{\hat{y}}_i}^2=(1-\mathrm{\β}){\mathrm{\σ}}_{{\hat{y}}_i}^2+\mathrm{\β}{\left|{\hat{y}}_i\left(n\right)\right|}^2\·\·\·\left[6\right]$

In the embodiment of the invention, when being single-ended speaking, the state of judging the sub-band adaptive wave filter (works as related coefficient η _iClose to 1, be expressed as η _i→ 1) time, shows that the residual signals of this i subband sef-adapting filter output mainly is residual echo, need in such cases its further inhibition, as this residual signals being replaced with the comfort noise that this subband generates; When the state of judging the sub-band adaptive wave filter be both-end when speaking (as related coefficient η _iClose to 0, be expressed as η _i→ 0) time, can the residual signals of this sub-band adaptive wave filter output not handled, but directly output.

Below in conjunction with accompanying drawing the embodiment of the invention is described in detail.

Referring to Fig. 2, eliminate schematic flow sheet for a kind of echo based on subband that the embodiment of the invention provides, as shown in the figure, this flow process can comprise:

Step 201 is calculated the cross-correlation coefficient η of near end input signal and the echo estimated signal of each subband _i

Concrete, can calculate the cross-correlation coefficient of near end input signal and the echo estimated signal of each subband according to formula (4), formula (5) and formula (6).

Step 202 is according to the cross-correlation coefficient η of each subband _i, determine the residing state of each sub-band adaptive wave filter.If the sub-band adaptive wave filter is in the single-ended state of speaking, then change step 203 over to; If the sub-band adaptive wave filter is in the both-end state of speaking, then change step 204 over to.

Wherein, if the cross-correlation coefficient η of subband _i〉=T (wherein T is setting threshold, 0＜T＜1) thinks that then this sub-band adaptive wave filter is in the single-ended state of speaking; If the cross-correlation coefficient η of subband _i＜T thinks that then this sub-band adaptive wave filter is in the both-end state of speaking.Wherein the value of T can be according to the power of the nonlinear degree of system, environmental background noise, and the factors such as performance of sub-band adaptive wave filter determine, as T=0.5.

Step 203 for the sub-band adaptive wave filter that is in the single-ended state of speaking, is handled back output to its residual signals, to suppress or elimination echo.

In this step, can eliminate echo in several ways, the comfort noise that the embodiment of the invention preferably replaces with the subband residual signals this subband is then by the mode of sub-band adaptive wave filter output.

Step 204 for being in the speak sub-band adaptive wave filter of state of both-end, is directly exported the residual signals of sub-band adaptive wave filter.

By above-mentioned flow process as can be seen, to each subband of synchronization, judge that it is in different states, and do corresponding subsequent treatment respectively; η _i〉=T is judged as single-ended speaking, and residual echo is replaced η with the comfort noise of corresponding subband _i＜T is judged as both-end and speaks, and directly exports the residual signals of sub-band adaptive wave filter output, thereby suppresses residual echo targetedly, the phenomenon of having cut sound when having avoided both-end to speak to a certain extent.

When the non-linear more serious or environmental background noise of system is more intense, subband give coefficient η mutually _i→ (T ± Δ) is as η _i→ (0.5 ± 0.2), in this case, if only according to the cross-correlation coefficient η of each subband _iWith the relation of threshold value T, just be difficult to draw state accurately, and then can influence the effect that echo is eliminated.

For addressing this problem, the embodiment of the invention is further improved scheme shown in Figure 1, the comprehensive information of each subband, (η when existing the subband of setting quantity can clearly be judged as single-ended speaking _i→ 1), perhaps can clearly be judged as single-ended subband proportion of speaking when reaching preset proportion, other subbands that are in critical conditions (are η _i→ (T ± Δ) is as η _i→ (0.5 ± 0.2)) being judged to be broken into single-ended probability of speaking will be bigger; In like manner, when subband that exist to set quantity can clearly be judged as both-end and speaks, other subbands that are in critical conditions are judged to be broken into the probability that both-end speaks can be bigger.Be described in detail below in conjunction with 2 pairs of these improvement flow processs of accompanying drawing.

Referring to Fig. 3, eliminate schematic flow sheet for the another kind that the embodiment of the invention provides based on the echo of subband, as shown in the figure, this flow process can comprise:

Step 301 is calculated the cross-correlation coefficient η of near end input signal and the echo estimated signal of each subband _i

Step 302～304 are according to the cross-correlation coefficient η of each subband _i, determine the residing state of each sub-band adaptive wave filter.If the sub-band adaptive wave filter is in the single-ended state of speaking, then change step 305 over to; If the sub-band adaptive wave filter is in the both-end state of speaking, then change step 306 over to.

Wherein, if the cross-correlation coefficient η of subband _i〉=T ₁(T wherein ₁Be the first threshold of setting), think that then this sub-band adaptive wave filter is in the single-ended state of speaking; If the cross-correlation coefficient η of subband _i≤ T ₂(T wherein ₂Second threshold value of be setting), think that then this sub-band adaptive wave filter is in the both-end state of speaking; If the cross-correlation coefficient T of subband ₂＜η _i＜T ₁, need then further to determine that it still is the both-end state of speaking that this sub-filter belongs to the single-ended state of speaking.For cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope, can be according to being in speak quantity or the proportion of sub-band adaptive wave filter of state of single-ended speak state or both-end, determine that it still is the both-end state of speaking that this part sub-band adaptive wave filter belongs to the single-ended state of speaking.Wherein, 0＜T ₂＜T ₁＜1, such as, when the non-linear more serious or environmental background noise of system is more intense, the cross-correlation coefficient η of each sub-band adaptive wave filter _iDuring → (0.5 ± 0.2), can establish T ₁=0.3, T ₂=0.7.

Concrete, for cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope, can determine its residing state (only showing the wherein specific implementation of mode one among Fig. 3) in the following way:

Mode one: if cross-correlation coefficient η _i〉=T ₁The quantity of subband surpass setting threshold (n＞th_n as shown in Figure 3, wherein n be cross-correlation coefficient η _i〉=T ₁The quantity of subband, th_n is setting threshold), then think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the single-ended state of speaking; Otherwise, think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the both-end state of speaking.

Mode two: if cross-correlation coefficient η _i〉=T ₁Subband shared ratio in all subbands surpass setting threshold (as surpassing 50%), then think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the single-ended state of speaking; Otherwise, think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the both-end state of speaking.

Mode three: if cross-correlation coefficient η _i≤ T ₂The quantity of subband surpass setting threshold, then think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the both-end state of speaking, otherwise think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the single-ended state of speaking.

Mode four: if cross-correlation coefficient η _i≤ T ₂Subband shared ratio in all subbands surpass setting threshold (as surpassing 50%), then think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the both-end state of speaking, otherwise think cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the single-ended state of speaking.

Step 305 for the sub-band adaptive wave filter that is in the single-ended state of speaking, is handled back output to its residual signals, to suppress or elimination echo.Specific implementation can be with the step 203 among Fig. 2.

Step 306 for being in the speak sub-band adaptive wave filter of state of both-end, is directly exported the residual signals of sub-band adaptive wave filter.

A kind of replacement scheme of flow process shown in Figure 3 can be as shown in Figure 4, and different with flow process shown in Figure 3 is: for cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope, determine that the mode of its state of living in is: according to the cross-correlation coefficient η of all subbands _iAverage determine the average that obtains as arithmetic mean or other algorithms.As shown in Figure 4, this flow process can comprise:

Step 401 is calculated the cross-correlation coefficient η of near end input signal and the echo estimated signal of each subband _i

Step 402～404 are according to the cross-correlation coefficient η of each subband _i, determine the residing state of each sub-band adaptive wave filter.If the sub-band adaptive wave filter is in the single-ended state of speaking, then change step 405 over to; If the sub-band adaptive wave filter is in the both-end state of speaking, then change step 406 over to.

Wherein, if the cross-correlation coefficient η of subband _i〉=T ₁, think that then this sub-band adaptive wave filter is in the single-ended state of speaking; If the cross-correlation coefficient η of subband _i≤ T ₂, think that then this sub-band adaptive wave filter is in the both-end state of speaking; If the cross-correlation coefficient T of subband ₂＜η _i＜T ₁, need then further to determine that it still is the both-end state of speaking that this sub-filter belongs to the single-ended state of speaking.For cross-correlation coefficient η _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope, can be according to the cross-correlation coefficient η of all subbands _iArithmetic mean determine that it still is the both-end state of speaking that this part sub-band adaptive wave filter belongs to the single-ended state of speaking.

Concrete, if the cross-correlation coefficient η of all subbands _iArithmetic mean γ 〉=T ₃(T wherein ₃The 3rd threshold value of be setting), cross-correlation coefficient η then _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the single-ended state of speaking; If the cross-correlation coefficient η of all subbands _iArithmetic mean γ＜T ₃, cross-correlation coefficient η then _iAt (T ₂, T ₁) the sub-band adaptive wave filter of scope is in the both-end state of speaking.Wherein, 0＜T ₂＜T ₃＜T ₁＜1, as establishing T ₁=0.8, T ₂=0.2, T ₃=0.5.The cross-correlation coefficient η of all subbands _iArithmetic mean γ can be expressed as:

$\mathrm{\γ}=\frac{1}{N}\underset{i-1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\η}}_i\·\·\·\left[7\right]$

Wherein, N is number of sub-bands.

Step 405 for the sub-band adaptive wave filter that is in the single-ended state of speaking, is handled back output to its residual signals, to suppress or elimination echo.Specific implementation can be with the step 103 among Fig. 1.

Step 406 for being in the speak sub-band adaptive wave filter of state of both-end, is directly exported the residual signals of sub-band adaptive wave filter.

In the flow process shown in Figure 4, directly judge cross-correlation coefficient η by the mean value that calculates each subband related coefficient _iAt (T ₂, T ₁) state of sub-band adaptive wave filter of scope, may compare rough.For example only contain ground unrest at certain subband, or energy ingredient seldom echo or during near-end voice signals, the related coefficient η of this subband _iAlso participate in mean value calculation, when the faint subband ratio of energy more for a long time, may to final average as a result γ cause adverse influence.

For addressing this problem, another embodiment of the present invention is further improved flow process shown in Figure 4, has introduced the factor of energy and has judged cross-correlation coefficient η _iAt (T ₂, T ₁) state of sub-band adaptive wave filter of scope.Concrete, average γ carries out the energy weighted mean by the related coefficient of each subband and tries to achieve, and can give the clearer and more definite subband of the relatively large state of energy (as η _i〉=T ₁Or η _i≤ T ₂) with bigger weight, make final judgement critical conditions become relatively reasonably state.

Flow process after the improvement and flow process shown in Figure 3 are basic identical, just come computation of mean values with following formula (8) replacement formula (7):

$\mathrm{\γ}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_i{\mathrm{\η}}_i\·\·\·\left[8\right]$

Wherein, N is number of sub-bands, and the energy weight can be expressed as:

${\mathrm{\λ}}_i=\frac{{\mathrm{\σ}}_{d_i}^2}{\underset{i-1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\σ}}_{d_i}^2}\·\·\·\left[9\right]$

In the formula (9)

Can calculate by formula (2).

By above flow process as can be seen, the embodiment of the invention is judged that it is in different states, and is done corresponding subsequent treatment respectively each subband of synchronization; η _i〉=T ₁Be judged as single-ended speaking, residual echo is replaced η with the comfort noise of corresponding subband _i≤ T ₂Be judged as both-end and speak, directly export the residual signals of sub-band adaptive wave filter output; Particularly the subband for critical conditions (is cross-correlation coefficient η _iAt (T ₂, T ₁) subband of scope), by weighted mean it is classified as reasonably state, reach better inhibition residual echo, and the phenomenon of cutting sound when avoiding both-end to speak.

Based on identical technical conceive, the embodiment of the invention also provides a kind of echo cancellation devices that can be applied to the above-mentioned flow process that the embodiment of the invention provides.

As shown in Figure 5, this echo cancellation devices can comprise:

State determination module 501 is used for respectively the cross-correlation coefficient according near end input signal and the echo estimated signal of each subband, determines the residing state of each sub-band adaptive wave filter;

Output module 502 is used for being in single-ended speaking during state when the subband sef-adapting filter, replaces the signal of exporting behind the residual signals of this sub-band adaptive wave filter after the replacement with the comfort noise of this subband; When the subband sef-adapting filter is in both-end and speaks state, export the residual signals of this sub-band adaptive wave filter.

The cross-correlation coefficient that state determination module 501 is judged subbands determines that the sub-band adaptive wave filter is in the single-ended state of speaking during more than or equal to first threshold; When the cross-correlation coefficient of judging subband is less than or equal to second threshold value, determine that the sub-band adaptive wave filter is in the both-end state of speaking; For the subband of cross-correlation coefficient between second threshold value and first threshold, quantity or proportion according to the sub-band adaptive wave filter of designated state, or the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter; Wherein, 0＜the second threshold value＜first threshold＜1.

Concrete, state determination module 501 is for the subband of cross-correlation coefficient between second threshold value and first threshold, quantity according to the sub-band adaptive wave filter of designated state, when determining the residing state of this sub-band adaptive wave filter, surpass setting threshold if be in the quantity of the sub-band adaptive wave filter of the single-ended state of speaking, judge that then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps, surpass setting threshold if be in the speak quantity of sub-band adaptive wave filter of state of both-end, the sef-adapting filter of then judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise, judge that the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

Concrete, state determination module 501 is for the subband of cross-correlation coefficient between second threshold value and first threshold, the ratio shared according to the sub-band adaptive wave filter of designated state, when determining the residing state of this sub-band adaptive wave filter, surpass setting threshold if be in the shared ratio of the sub-band adaptive wave filter quantity of the single-ended state of speaking, judge that then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps, surpass setting threshold if be in the both-end shared ratio of the sub-band adaptive wave filter quantity of state of speaking, the sef-adapting filter of then judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise, judge that the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

Concrete, state determination module 501 is for the subband of cross-correlation coefficient between second threshold value and first threshold, mean value according to the described cross-correlation coefficient of each sub-band adaptive wave filter, when determining the residing state of this sub-band adaptive wave filter, if the mean value of the described cross-correlation coefficient of each subband, is then judged the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold greater than the 3rd threshold value and is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

Concrete, state determination module 501 is for the subband of cross-correlation coefficient between second threshold value and first threshold, mean value according to the described cross-correlation coefficient of each sub-band adaptive wave filter, when determining the residing state of this sub-band adaptive wave filter, if the weighted mean value of the described cross-correlation coefficient of each subband, is then judged the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold greater than the 3rd threshold value and is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

Concrete, state determination module 501 employed cross-correlation coefficients greater than the corresponding weighted value of the subband of first threshold and cross-correlation coefficient less than the corresponding weighted value of the subband of second threshold value, greater than the weighted value of the subband of cross-correlation coefficient between second threshold value and first threshold.

In sum; the embodiment of the invention is simple and be easy to real-time system and realize; taken full advantage of the characteristic that subband signal is handled; make residual echo obtain more effective inhibition; strengthened simultaneously when both-end is spoken to the inhibition of residual echo and to the protection of local terminal speaker sound, the effect of entire system and fluency are promoted.

Through the above description of the embodiments, those skilled in the art can be well understood to the present invention and can realize by the mode that software adds essential general hardware platform, can certainly pass through hardware, but the former is better embodiment under a lot of situation.Based on such understanding, the part that technical scheme of the present invention contributes to prior art in essence in other words can embody with the form of software product, this computer software product is stored in the storage medium, comprise that some instructions are with so that a station terminal equipment (can be mobile phone, personal computer, server, the perhaps network equipment etc.) carry out the described method of each embodiment of the present invention.

The above only is preferred implementation of the present invention; should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; can also make some improvements and modifications, these improvements and modifications also should be looked protection scope of the present invention.

Claims (13)

1. an echo cancel method is characterized in that, may further comprise the steps:

According to the cross-correlation coefficient of near end input signal and the echo estimated signal of each subband, determine the residing state of each sub-band adaptive wave filter respectively;

When the subband sef-adapting filter is in single-ended speaking during state, replace the signal after the output replacement behind the residual signals of this sub-band adaptive wave filter with the comfort noise of this subband;

When the subband sef-adapting filter is in both-end and speaks state, export the residual signals of this sub-band adaptive wave filter;

Wherein, according to the cross-correlation coefficient of near end input signal and the echo estimated signal of subband, determine the residing state of sub-band adaptive wave filter, comprising:

When the cross-correlation coefficient of subband during more than or equal to first threshold, determine that the sub-band adaptive wave filter is in the single-ended state of speaking;

When the cross-correlation coefficient of subband is less than or equal to second threshold value, determine that the sub-band adaptive wave filter is in the both-end state of speaking;

For the subband of cross-correlation coefficient between second threshold value and first threshold, quantity or proportion according to the sub-band adaptive wave filter of designated state, or the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter;

Wherein, 0＜the second threshold value＜first threshold＜1.

2. the method for claim 1 is characterized in that, for the subband of cross-correlation coefficient between second threshold value and first threshold, according to the quantity of the sub-band adaptive wave filter of designated state, determines the residing state of this sub-band adaptive wave filter, is specially:

Surpass setting threshold if be in the quantity of the sub-band adaptive wave filter of the single-ended state of speaking, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps,

Surpass setting threshold if be in the speak quantity of sub-band adaptive wave filter of state of both-end, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

3. the method for claim 1, it is characterized in that, for the subband of cross-correlation coefficient between second threshold value and first threshold, the ratio shared according to the sub-band adaptive wave filter of designated state, determine the residing state of this sub-band adaptive wave filter, be specially:

Surpass setting threshold if be in the shared ratio of the sub-band adaptive wave filter quantity of the single-ended state of speaking, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps,

Surpass setting threshold if be in the both-end shared ratio of the sub-band adaptive wave filter quantity of state of speaking, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

4. the method for claim 1, it is characterized in that, for the subband of cross-correlation coefficient between second threshold value and first threshold, according to the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter, be specially:

If the mean value of the described cross-correlation coefficient of each subband is greater than the 3rd threshold value, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking;

Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

5. the method for claim 1, it is characterized in that, for the subband of cross-correlation coefficient between second threshold value and first threshold, according to the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter, be specially:

If the weighted mean value of the described cross-correlation coefficient of each subband is greater than the 3rd threshold value, then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking;

Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

6. method as claimed in claim 5, it is characterized in that, cross-correlation coefficient greater than the corresponding weighted value of the subband of first threshold and cross-correlation coefficient less than the corresponding weighted value of the subband of second threshold value, greater than the weighted value of the subband of cross-correlation coefficient between second threshold value and first threshold.

7. method as claimed in claim 5 is characterized in that, the weighted value of subband is:

${\mathrm{\λ}}_i=\frac{{\mathrm{\σ}}_{d_i}^2}{\underset{i-1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\σ}}_{d_i}^2}$

Wherein, N is number of sub-bands,

Energy for the subband near end input signal of Echo Canceller.

8. an echo cancelling device is characterized in that, comprising:

The state determination module is used for respectively the cross-correlation coefficient according near end input signal and the echo estimated signal of each subband, determines the residing state of each sub-band adaptive wave filter;

Output module is used for being in single-ended speaking during state when the subband sef-adapting filter, replaces the signal of exporting behind the residual signals of this sub-band adaptive wave filter after the replacement with the comfort noise of this subband; When the subband sef-adapting filter is in both-end and speaks state, export the residual signals of this sub-band adaptive wave filter;

Wherein, described state determination module specifically is used for, and when the cross-correlation coefficient of subband during more than or equal to first threshold, determines that the sub-band adaptive wave filter is in the single-ended state of speaking; When the cross-correlation coefficient of subband is less than or equal to second threshold value, determine that the sub-band adaptive wave filter is in the both-end state of speaking; For the subband of cross-correlation coefficient between second threshold value and first threshold, quantity or proportion according to the sub-band adaptive wave filter of designated state, or the mean value of the described cross-correlation coefficient of each sub-band adaptive wave filter, determine the residing state of this sub-band adaptive wave filter; Wherein, 0＜the second threshold value＜first threshold＜1.

9. device as claimed in claim 8, it is characterized in that, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, quantity according to the sub-band adaptive wave filter of designated state, when determining the residing state of this sub-band adaptive wave filter, surpass setting threshold if be in the quantity of the sub-band adaptive wave filter of the single-ended state of speaking, judge that then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps, surpass setting threshold if be in the speak quantity of sub-band adaptive wave filter of state of both-end, the sef-adapting filter of then judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise, judge that the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

10. device as claimed in claim 8, it is characterized in that, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, the ratio shared according to the sub-band adaptive wave filter of designated state, when determining the residing state of this sub-band adaptive wave filter, surpass setting threshold if be in the shared ratio of the sub-band adaptive wave filter quantity of the single-ended state of speaking, judge that then the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Perhaps, surpass setting threshold if be in the both-end shared ratio of the sub-band adaptive wave filter quantity of state of speaking, the sef-adapting filter of then judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Otherwise, judge that the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold is in the single-ended state of speaking.

11. device as claimed in claim 8, it is characterized in that, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, mean value according to the described cross-correlation coefficient of each sub-band adaptive wave filter, when determining the residing state of this sub-band adaptive wave filter, if the mean value of the described cross-correlation coefficient of each subband, is then judged the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold greater than the 3rd threshold value and is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

12. device as claimed in claim 8, it is characterized in that, described state determination module specifically is used for, for the subband of cross-correlation coefficient between second threshold value and first threshold, mean value according to the described cross-correlation coefficient of each sub-band adaptive wave filter, when determining the residing state of this sub-band adaptive wave filter, if the weighted mean value of the described cross-correlation coefficient of each subband, is then judged the sef-adapting filter of each subband of cross-correlation coefficient between second threshold value and first threshold greater than the 3rd threshold value and is in the single-ended state of speaking; Otherwise the sef-adapting filter of judging each subband of cross-correlation coefficient between second threshold value and first threshold is in the both-end state of speaking; Wherein, 0＜the second threshold value＜the 3rd threshold value＜first threshold＜1.

13. device as claimed in claim 12, it is characterized in that, the employed cross-correlation coefficient of described state determination module greater than the corresponding weighted value of the subband of first threshold and cross-correlation coefficient less than the corresponding weighted value of the subband of second threshold value, greater than the weighted value of the subband of cross-correlation coefficient between second threshold value and first threshold.

Images