CN104954595A - Cancellation method and device of residual echo - Google Patents

Abstract

The invention provides a cancellation method and device of residual echo. The method includes the steps: a microphone signal is filtered by a filter to obtain a first estimated value of near-end voice; the filter is updated with the first estimated value and the microphone signal to obtain an updated one; the updated filter filters the microphone signal to obtain a second estimated value of the near-end voice. By introducing the first estimated value of the near-end voice to the adaptively filtered input signal, adjustment of the filter always follows the echo component in the microphone signal, near-end voice estimation is less distorted, and better voice recognition and audio communication quality are can be acquired.

Description

Residual echo removing method and device

Technical field

The present invention relates to voice processing technology field, particularly relate to a kind of residual echo removing method and device.

Background technology

Acoustic echo eliminates (Acoustic Echo Chancellor; Hereinafter referred to as: AEC) technology is intelligent sound call and an indispensable part in voice interactive system.In voice call, particularly under hands-free mode, again by microphone typing and then be transmitted to the other side after the sound air-borne transmission that loud speaker plays back, user hears that the echo of oneself has uncomfortable sensation again and again.In the intelligent speech interactive systems such as vehicle mounted guidance, user wish by voice control navigate " search destination " or " report weather " etc., navigate in reciprocal process from Text To Speech (Text to Speech; Hereinafter referred to as: the TTS) order of meeting interference user, if TTS is taken as instruction to identify the misoperation that more can cause navigation, makes intelligent navigation no longer intelligent.Thus the performance of AEC directly has influence on the experience of speech production.

It is the method generally adopted in current AEC system that adaptive-filtering adds post filtering, but existing method is to the poor-performing of speech recognition, and near-end speech distortion is serious, and voice sound intermittent, and user experience is bad.

Summary of the invention

Object of the present invention is intended to solve one of technical problem in correlation technique at least to a certain extent.

For this reason, first object of the present invention is to propose a kind of residual echo removing method, the method by introducing the first time estimated value of near-end speech in the input signal of adaptive-filtering, the adjustment of filter is made always to follow to the echo composition in microphone signal, decrease the distortion that near-end speech is estimated, thus better speech recognition performance and voice communication quality can be obtained.

Second object of the present invention is to propose a kind of residual echo cancellation element.

To achieve these goals, the residual echo removing method of first aspect present invention embodiment, comprising: carry out filtering by filter to microphone signal, obtains the first time estimated value of near-end speech; Utilize described first time estimated value and described microphone signal described filter is upgraded, obtain the filter after upgrading; By the filter after described renewal, filtering is carried out to described microphone signal, obtain the second time estimated value of described near-end speech.

The residual echo removing method of the embodiment of the present invention, by filter, filtering is carried out to microphone signal, obtain the first time estimated value of near-end speech, then utilization first time estimated value and microphone signal upgrade above-mentioned filter, obtain the filter after upgrading, finally by the filter after upgrading, filtering is carried out to above-mentioned microphone signal, obtain the second time estimated value of near-end speech.Said method by introducing the first time estimated value of near-end speech in the input signal of adaptive-filtering, the adjustment of filter is made always to follow to the echo composition in microphone signal, decrease the distortion that near-end speech is estimated, thus better speech recognition performance and voice communication quality can be obtained.

To achieve these goals, the residual echo cancellation element of second aspect present invention embodiment, comprising: filtration module, for carrying out filtering by filter to microphone signal, obtains the first time estimated value of near-end speech; Update module, obtain for utilizing described filtration module first time estimated value and described microphone signal described filter is upgraded, obtain upgrade after filter; Described filtration module, also for carrying out filtering by the filter after described update module renewal to described microphone signal, obtains the second time estimated value of described near-end speech.

The residual echo cancellation element of the embodiment of the present invention, filtration module carries out filtering by filter to microphone signal, obtain the first time estimated value of near-end speech, then update module utilizes estimated value and microphone signal for the first time to upgrade above-mentioned filter, obtain the filter after upgrading, last filtration module carries out filtering by the filter after renewal to above-mentioned microphone signal, obtains the second time estimated value of near-end speech.Said apparatus by introducing the first time estimated value of near-end speech in the input signal of adaptive-filtering, the adjustment of filter is made always to follow to the echo composition in microphone signal, decrease the distortion that near-end speech is estimated, thus better speech recognition performance and voice communication quality can be obtained.

The aspect that the present invention adds and advantage will part provide in the following description, and part will become obvious from the following description, or be recognized by practice of the present invention.

Accompanying drawing explanation

The present invention above-mentioned and/or additional aspect and advantage will become obvious and easy understand from the following description of the accompanying drawings of embodiments, wherein:

Fig. 1 is the flow chart of a residual echo removing method of the present invention embodiment;

Fig. 2 be another embodiment of residual echo removing method of the present invention realize block diagram;

Fig. 3 is the structural representation of a residual echo cancellation element of the present invention embodiment.

Embodiment

Be described below in detail embodiments of the invention, the example of described embodiment is shown in the drawings, and wherein same or similar label represents same or similar element or has element that is identical or similar functions from start to finish.Being exemplary below by the embodiment be described with reference to the drawings, only for explaining the present invention, and can not limitation of the present invention being interpreted as.On the contrary, embodiments of the invention comprise fall into attached claims spirit and intension within the scope of all changes, amendment and equivalent.

Fig. 1 is the flow chart of a residual echo removing method of the present invention embodiment, and as shown in Figure 1, this residual echo removing method can comprise:

Step 101, carries out filtering by filter to microphone signal, obtains the first time estimated value of near-end speech.

Step 102, utilizes estimated value and above-mentioned microphone signal for the first time to upgrade above-mentioned filter, obtains the filter after upgrading.

Particularly, estimated value and above-mentioned microphone signal is for the first time utilized to upgrade above-mentioned filter, obtain upgrade after filter can be: using above-mentioned microphone signal with first time estimated value difference as input signal, utilize echo signal to upgrade above-mentioned filter as with reference to signal, obtain the filter after upgrading.Filter after above-mentioned renewal can be:

wherein, ε (K)=[X (K)-S ' (K)]-H _t' ^*(K) R (K); (1)

In formula (1), H ' _t+1(K) be the filter after renewal, H ' _t(K) be the filter before renewal, R (K) is echo signal, and X (K) is microphone signal, and S ' (K) is first time estimated value, and μ is for upgrading step-length; " * " represents conjugate operation.

Step 103, carries out filtering by the filter after upgrading to above-mentioned microphone signal, obtains the second time estimated value of above-mentioned near-end speech.

Particularly, by the filter after upgrading, filtering is carried out to above-mentioned microphone signal, the second time estimated value obtaining above-mentioned near-end speech can be: calculate residual echo according to the filter after above-mentioned renewal, and secondary filtering is carried out to above-mentioned microphone signal, obtain the second time estimated value of near-end speech.The second time estimated value of above-mentioned near-end speech can be:

$S^{\′\′}\left(K\right)=W_2^{*}\left(K\right)X\left(K\right),$ Wherein, $W_2\left(K\right)=R_{\mathrm{xx}}^{-1}\left(K\right)R_{\mathrm{sx}}^{\′}\left(K\right),$ R _xx(K)＝E{X ^*(K)X(K)}， $R_{\mathrm{sx}}^{\′}\left(K\right)=E\left\{{[X\left(K\right)-H_{t+1}^{\′*}\left(K\right)R\left(K\right)]}^{*}X\left(K\right)\right\};---\left(2\right)$

In formula (2), S " the second time estimated value that (K) is near-end speech, H ' _t+1(K) be the filter after renewal, for residual echo, R (K) is echo signal, and X (K) is described microphone signal, and " * " represents conjugate operation, and E{} represents mathematic expectaion.

Filter in the present embodiment can be postfilter, such as Weiner filter, and the present embodiment is not construed as limiting this.

Above-mentioned residual echo removing method, by filter, filtering is carried out to microphone signal, obtain the first time estimated value of near-end speech, then utilization first time estimated value and microphone signal upgrade above-mentioned filter, obtain the filter after upgrading, finally by the filter after upgrading, filtering is carried out to above-mentioned microphone signal, obtain the second time estimated value of near-end speech.Said method by introducing the first time estimated value of near-end speech in the input signal of adaptive-filtering, the adjustment of filter is made always to follow to the echo composition in microphone signal, decrease the distortion that near-end speech is estimated, thus better speech recognition performance and voice communication quality can be obtained.

The present invention obtain residual echo transfer function process in, using near-end speech first time estimated value as least mean square algorithm (Least Mean Square Algorithm; Hereinafter referred to as: a LMS) input signal of adaptive-filtering, can reach the object of protection near-end speech.Now the transfer function restrained is brought into second Weiner filter and just can estimate near-end speech.

Fig. 2 be another embodiment of residual echo removing method of the present invention realize block diagram, Fig. 2 is described signal model for frequency-region signal.In Fig. 2, S (K) is near-end speech, i.e. desired signal; R (K) is echo signal, i.e. reference signal; X (K) is microphone signal, i.e. observation signal; H _t(K) for loud speaker is to the path transfer function of microphone.

" (K) is respectively the first time estimated value and second time estimated value of S (K), H ' for S ' (K) and S _t+1(K) be H ' _t(K) the once renewal after the first time estimated value S ' (K) introducing near-end speech.

Microphone signal is near-end speech and the superposing of residual echo, and can be expressed as:

X(K)＝S(K)+H ^*(K)R(K)

(3)

Wherein, " * " represents conjugate operation.So, the first time estimated value of near-end speech can be obtained according to the computing formula of Wiener filtering:

$S^{\′}\left(K\right)=W_1^{*}\left(K\right)X\left(K\right),$ Wherein, $W_1\left(K\right)=R_{\mathrm{xx}}^{-1}\left(K\right)R_{\mathrm{sx}}\left(K\right),$ R _xx(K)＝E{X ^*(K)X(K)}， $R_{\mathrm{sx}}\left(K\right)=E\left\{{[X\left(K\right)-H_t^{*}\left(K\right)R\left(K\right)]}^{*}X\left(K\right)\right\}.---\left(4\right)$

In formula (4), R _xx(K) be the covariance function of microphone signal, R _sx(K) be the cross covariance function of microphone signal and near-end speech.

H _t(K) and between S (K) can mutually restrict, H _t(K) estimation will definitely not cause being deteriorated to the estimation of S (K), and the estimation of S (K) can affect H again conversely _t(K) constringency performance.If using the difference of microphone signal and near-end speech as input signal, echo signal R (K) is as reference signal, echo is followed the tracks of by the method for LMS adaptive-filtering, will reduce during residual echo is estimated the near-end speech comprised to leak, so secondary Wiener filtering just can obtain near-end speech estimated value more reliably.So, first time estimated value S ' (K) can be brought into LMS sef-adapting filter, the H ' that a constringency performance improves can be obtained thus _t+1(K), then by H ' _t+1(K) the second time estimated value S " (K) of near-end speech just can be obtained for second Weiner filter.Wherein, the LMS filter after renewal such as formula shown in (1), can not repeat them here.

H ' _t+1(K) be a better filter of constringency performance, the residual echo in microphone signal can be traced into preferably.At acquisition H ' _t+1(K) after, then by H ' _t+1(K) just can obtain the second time estimated value S of near-end speech " (K), shown in (2) for second Weiner filter.

Residual echo removing method provided by the invention realizes in conjunction with twice Wiener filtering and a LMS adaptive-filtering, by introducing the first time estimated value of near-end speech in the input signal of LMS filter, the adjustment of filter is made always to follow to the echo composition in microphone signal, decrease the distortion that near-end speech is estimated, thus better speech recognition performance and voice communication quality can be obtained.And after eliminating near-end speech, the adjustment process of LMS filter reduces says (Double Talk to two; Hereinafter referred to as: the DT) dependence of testing result, and then enhance the robustness of non-linear reprocessing.

Fig. 3 is the structural representation of a residual echo cancellation element of the present invention embodiment, residual echo cancellation element in the present embodiment can realize the present invention's flow process embodiment illustrated in fig. 1, as shown in Figure 3, above-mentioned residual echo cancellation element can comprise: filtration module 31 and update module 32;

Wherein, filtration module 31, for carrying out filtering by filter to microphone signal, obtains the first time estimated value of near-end speech;

Update module 32, obtain for utilizing filtration module 31 first time estimated value and above-mentioned microphone signal above-mentioned filter is upgraded, obtain upgrade after filter;

Filtration module 31, also for carrying out filtering by the filter after update module 32 renewal to above-mentioned microphone signal, obtains the second time estimated value of near-end speech.

In the present embodiment, update module 32, specifically for using above-mentioned microphone signal with first time estimated value difference as input signal, utilizing echo signal as upgrading above-mentioned filter with reference to signal, obtaining the filter after upgrading.Wherein, update module 32 obtain renewal after filter be: wherein, ε (K)=[X (K)-S ' (K)]-H ' _t ^*(K) R (K);

Wherein, H ' _t+1(K) be the filter after renewal, H ' _t(K) be the filter before renewal, R (K) is echo signal, and X (K) is microphone signal, and S ' (K) is first time estimated value, and μ is for upgrading step-length; " * " represents conjugate operation.

In the present embodiment, filtration module 31, specifically for calculating residual echo according to the filter after above-mentioned renewal, and carrying out secondary filtering to above-mentioned microphone signal, obtaining the second time estimated value of above-mentioned near-end speech.Wherein, the second time estimated value of the near-end speech of filtration module 31 acquisition is: $S^{\′\′}\left(K\right)=W_2^{*}\left(K\right)X\left(K\right),$ Wherein, $W_2\left(K\right)=R_{\mathrm{xx}}^{-1}\left(K\right)R_{\mathrm{sx}}^{\′}\left(K\right),$ R _xx(K)＝E{X ^*(K)X(K)}， $R_{\mathrm{sx}}^{\′}\left(K\right)=E\left\{{[X\left(K\right)-H_{t+1}^{\′*}\left(K\right)R\left(K\right)]}^{*}X\left(K\right)\right\};$ Wherein, S " the second time estimated value that (K) is above-mentioned near-end speech, H ' _t+1(K) be the filter after renewal, for residual echo, R (K) is echo signal, and X (K) is microphone signal, and " * " represents conjugate operation, and E{} represents mathematic expectaion.

Above-mentioned residual echo cancellation element, filtration module 31 carries out filtering by filter to microphone signal, obtain the first time estimated value of near-end speech, then update module 32 utilizes estimated value and microphone signal for the first time to upgrade above-mentioned filter, obtain the filter after upgrading, last filtration module 31 carries out filtering by the filter after renewal to above-mentioned microphone signal, obtains the second time estimated value of near-end speech.Said apparatus by introducing the first time estimated value of near-end speech in the input signal of adaptive-filtering, the adjustment of filter is made always to follow to the echo composition in microphone signal, decrease the distortion that near-end speech is estimated, thus better speech recognition performance and voice communication quality can be obtained.

It should be noted that, in describing the invention, term " first ", " second " etc. only for describing object, and can not be interpreted as instruction or hint relative importance.In addition, in describing the invention, except as otherwise noted, the implication of " multiple " is two or more.

Describe and can be understood in flow chart or in this any process otherwise described or method, represent and comprise one or more for realizing the module of the code of the executable instruction of the step of specific logical function or process, fragment or part, and the scope of the preferred embodiment of the present invention comprises other realization, wherein can not according to order that is shown or that discuss, comprise according to involved function by the mode while of basic or by contrary order, carry out n-back test, this should understand by embodiments of the invention person of ordinary skill in the field.

Should be appreciated that each several part of the present invention can realize with hardware, software, firmware or their combination.In the above-described embodiment, multiple step or method can with to store in memory and the software performed by suitable instruction execution system or firmware realize.Such as, if realized with hardware, the same in another embodiment, can realize by any one in following technology well known in the art or their combination: the discrete logic with the logic gates for realizing logic function to data-signal, there is the application-specific integrated circuit (ASIC) of suitable combinational logic gate circuit, programmable gate array (Programmable Gate Array; Hereinafter referred to as: PGA), field programmable gate array (Field Programmable Gate Array; Hereinafter referred to as: FPGA) etc.

Those skilled in the art are appreciated that realizing all or part of step that above-described embodiment method carries is that the hardware that can carry out instruction relevant by program completes, described program can be stored in a kind of computer-readable recording medium, this program perform time, step comprising embodiment of the method one or a combination set of.

In addition, each functional module in each embodiment of the present invention can be integrated in a processing module, also can be that the independent physics of modules exists, also can two or more module integrations in a module.Above-mentioned integrated module both can adopt the form of hardware to realize, and the form of software function module also can be adopted to realize.If described integrated module using the form of software function module realize and as independently production marketing or use time, also can be stored in a computer read/write memory medium.

The above-mentioned storage medium mentioned can be read-only memory, disk or CD etc.

In the description of this specification, specific features, structure, material or feature that the description of reference term " embodiment ", " some embodiments ", " example ", " concrete example " or " some examples " etc. means to describe in conjunction with this embodiment or example are contained at least one embodiment of the present invention or example.In this manual, identical embodiment or example are not necessarily referred to the schematic representation of above-mentioned term.And the specific features of description, structure, material or feature can combine in an appropriate manner in any one or more embodiment or example.

Although illustrate and describe embodiments of the invention above, be understandable that, above-described embodiment is exemplary, can not be interpreted as limitation of the present invention, and those of ordinary skill in the art can change above-described embodiment within the scope of the invention, revises, replace and modification.

Claims (10)

1. a residual echo removing method, is characterized in that, comprising:

By filter, filtering is carried out to microphone signal, obtain the first time estimated value of near-end speech;

Utilize described first time estimated value and described microphone signal described filter is upgraded, obtain the filter after upgrading;

By the filter after described renewal, filtering is carried out to described microphone signal, obtain the second time estimated value of described near-end speech.

2. method according to claim 1, is characterized in that, described utilize described first time estimated value and described microphone signal described filter is upgraded, the filter obtained after upgrading comprises:

Using described microphone signal and described first time estimated value difference as input signal, utilize echo signal to upgrade described filter as with reference to signal, obtain the filter after upgrading.

3. method according to claim 2, is characterized in that,

Filter after described renewal is: wherein, $\mathrm{\ϵ}\left(K\right)=[X\left(K\right)-S^{\′}\left(K\right)]-H_t^{\′*}\left(K\right)R\left(K\right);$

Wherein, H ' _t+1(K) be the filter after renewal, H ' _t(K) be the filter before upgrading, R (K) is echo signal, and X (K) is described microphone signal, and S ' (K) is described first time estimated value, and μ is for upgrading step-length; " * " represents conjugate operation.

4. the method according to claim 1-3 any one, is characterized in that, describedly carries out filtering by the filter after described renewal to described microphone signal, and the second time estimated value obtaining described near-end speech comprises:

Calculate residual echo according to the filter after described renewal, and secondary filtering is carried out to described microphone signal, obtain the second time estimated value of described near-end speech.

5. method according to claim 4, is characterized in that,

The second time estimated value of described near-end speech is: $S^{\′\′}\left(K\right)=W_2^{*}\left(K\right)X\left(K\right),$ Wherein, $W_2\left(K\right)=R_{\mathrm{xx}}^{-1}\left(K\right)R_{\mathrm{sx}}^{\′}\left(K\right),$ R _xx(K)＝E{X ^*(K)X(K)}， $R_{\mathrm{sx}}^{\′}\left(K\right)=E\left\{{[X\left(K\right)-H_{t+1}^{\′*}\left(K\right)R\left(K\right)]}^{*}X\left(K\right)\right\};$

Wherein, S " the second time estimated value that (K) is described near-end speech, H ' _t+1(K) be the filter after renewal, for residual echo, R (K) is echo signal, and X (K) is described microphone signal, and " * " represents conjugate operation, and E{} represents mathematic expectaion.

6. a residual echo cancellation element, is characterized in that, comprising:

Filtration module, for carrying out filtering by filter to microphone signal, obtains the first time estimated value of near-end speech;

Update module, obtain for utilizing described filtration module first time estimated value and described microphone signal described filter is upgraded, obtain upgrade after filter;

Described filtration module, also for carrying out filtering by the filter after described update module renewal to described microphone signal, obtains the second time estimated value of described near-end speech.

7. device according to claim 6, is characterized in that,

Described update module, specifically for using described microphone signal and described first time estimated value difference as input signal, utilize echo signal to upgrade described filter as with reference to signal, obtain the filter after upgrading.

8. device according to claim 7, is characterized in that,

Filter after the renewal that described update module obtains is: wherein, $\mathrm{\ϵ}\left(K\right)=[X\left(K\right)-S^{\′}\left(K\right)]-H_t^{\′*}\left(K\right)R\left(K\right);$

Wherein, H ' _t+1(K) be the filter after renewal, H ' _t(K) be the filter before upgrading, R (K) is echo signal, and X (K) is described microphone signal, and S ' (K) is described first time estimated value, and μ is for upgrading step-length; " * " represents conjugate operation.

9. the device according to claim 6-8 any one, is characterized in that,

Described filtration module, specifically for calculating residual echo according to the filter after described renewal, and carrying out secondary filtering to described microphone signal, obtaining the second time estimated value of described near-end speech.

10. device according to claim 9, is characterized in that,

The second time estimated value of the near-end speech that described filtration module obtains is: wherein, $W_2\left(K\right)=R_{\mathrm{xx}}^{-1}\left(K\right)R_{\mathrm{sx}}^{\′}\left(K\right),$ R _xx(K)＝E{X ^*(K)X(K)}， $R_{\mathrm{sx}}^{\′}\left(K\right)=E\left\{{[X\left(K\right)-H_{t+1}^{\′*}\left(K\right)R\left(K\right)]}^{*}X\left(K\right)\right\};$

Wherein, S " the second time estimated value that (K) is described near-end speech, H ' _t+1(K) be the filter after renewal, for residual echo, R (K) is echo signal, and X (K) is described microphone signal, and " * " represents conjugate operation, and E{} represents mathematic expectaion.