CN101179294B - Self-adaptive echo eliminator and echo eliminating method thereof - Google Patents

Self-adaptive echo eliminator and echo eliminating method thereof Download PDF

Info

Publication number
CN101179294B
CN101179294B CN2006101144195A CN200610114419A CN101179294B CN 101179294 B CN101179294 B CN 101179294B CN 2006101144195 A CN2006101144195 A CN 2006101144195A CN 200610114419 A CN200610114419 A CN 200610114419A CN 101179294 B CN101179294 B CN 101179294B
Authority
CN
China
Prior art keywords
sliding window
fir filter
window fir
buffering area
sample
Prior art date
Application number
CN2006101144195A
Other languages
Chinese (zh)
Other versions
CN101179294A (en
Inventor
庞潼川
章东湖
万新
周丽丽
马志军
李立锋
王义锋
Original Assignee
黄山好视达通信技术有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 黄山好视达通信技术有限公司 filed Critical 黄山好视达通信技术有限公司
Priority to CN2006101144195A priority Critical patent/CN101179294B/en
Publication of CN101179294A publication Critical patent/CN101179294A/en
Application granted granted Critical
Publication of CN101179294B publication Critical patent/CN101179294B/en

Links

Abstract

The invention provides an adaptive echo eliminator and the echo eliminating method thereof; wherein the device consists essentially of a voice state detector, an NLMS (energy normalized least mean square error) controller and a sliding window FIR filter. The method mainly includes the following steps: sampling both a far-end voice signal and a near-end voice signal, and confirming the speaking status information of the current network according to the estimated values of the short-time energy of the sampled far-end speech signal sample and the near-end speech signal sample. Then, the coefficient of a sliding window FIR filter is configured according to the speaking status information of the current network, and the sliding window FIR filter step-by-step filters the near-end speech signal and the far-end speech signal filled into the buffer area according to the mobile length set by the configured coefficient. Utilizing the invention, the echo of the digital hands-free speaking system can be effectively eliminated, thereby effectively eliminating the echo of the digital communication system based on the embedded system.

Description

Self-adaptive echo eliminator and echo cancel method thereof

Technical field

The present invention relates to the communications field, relate in particular to a kind of self-adaptive echo eliminator and echo cancel method thereof.

Background technology

Time-delay and echo being influence the principal element of communication quality, and long time-delay can make echo problem more outstanding, cause communication normally to carry out the most at last.Therefore, in communication link,, just need carry out echo elimination as long as the delay of one-way voice transmission reaches more than the 10ms.At present, adopting the end-to-end delay of the digital voice communication system of compression algorithm based on the packing transmission all is to surpass the 10ms thresholding, and above-mentioned digital voice system all need utilize Echo Canceller to carry out echo elimination.

At present, International Telecommunication Association has issued a series of international standards with regard to echo cancellation technology, as: G.165, G.167, G.168 wait.But these international standards have only proposed general designing requirement to Echo Canceller; Function, implementation method, performance index and the means of testing thereof of Echo Canceller are described; Do not stipulate the specific algorithm and the design details of Echo Canceller, need the developer to design the structure and the implementation algorithm of Echo Canceller voluntarily.Also defined relevant objective examination's project in the above-mentioned international standard, can under variety of network conditions, provide enough echoes to eliminate ability to guarantee Echo Canceller.If Echo Canceller can reach these objective examination's requirements, so just can accomplish MIN echo cancellation performance.At present, the specific algorithm of multiple Echo Canceller has been arranged, adopted different algorithms to different application scenario, choosing of concrete scheme should be according to actual requirement, to the consideration of trading off of complexity, robustness and the convergence rate of algorithm.

At present, echo cancellation technology commonly used is the adaptive echo technology for eliminating.This echo cancellation technology is according to G.168 standard-required design.The basic thought of this technology is the characteristic parameter in estimated echo path, to produce the echo signal of a simulation, from the signal that receives, deducts this echo signal, thereby realizes echo elimination.Since echo path normally unknown with the time become, so the general sef-adapting filter that adopts comes the analog echo path as Echo Canceller in this technology.Main target is when estimating the echo path characteristic parameter exactly, promptly follows the tracks of the variation of echo path.

The basic structure of the Echo Canceller of the above-mentioned self-adaptive echo counteracting technology of common employing is as shown in Figure 1; In Fig. 1; Part is the model of echo path in the frame of broken lines of the left side, and it comes emulation through a digital filter, and this filter has detailed introduction in the ITU-TG.168 standard; According to different time delays and network characteristic, 8 kinds of different types are arranged.Part is the composition of Echo Canceller in the frame of broken lines of the right, and main modular has: speech detector, adaptive algorithm controller, sef-adapting filter and NLP (nonlinear processor).The concrete course of work of this Echo Canceller can simply be expressed as:

Near end signal S InWith remote signaling R InWhile is as the input of speech detector; Speech detector is judged the talking state of current network according to the signal power size of these several signal time windows; Give the adaptive algorithm controller with the state information notification that obtains simultaneously, and then the work of control sef-adapting filter.

If it is far-end mode (speaking in the opposite end of promptly conversing) that speech detector is judged the talking state that obtains; So; The adaptive algorithm controller upgrades the coefficient of sef-adapting filter according to the size of signal power in previous moment and the current time filter and error signal e (n), and the renewal result with coefficient outputs in the sef-adapting filter simultaneously.

The main effect of above-mentioned sef-adapting filter is to predict the echo signal that produces at near-end through remote signaling, and its common structure is horizontal FIR filter.The input of sef-adapting filter is a remote signaling and by the update coefficients (if the state of Echo Canceller is a far-end mode) that obtains of output in the adaptive algorithm controller; Its effect be with remote signaling as a reference signal, and come the echogenicity estimated signal according to this reference signal.

After the process hybrid network, the near end signal S that imports in the Echo Canceller InIn comprised remote signaling R InEcho, with this near end signal S InWith the input of the echo estimated signal of sef-adapting filter output, near end signal S as subtracter InIn deduct the echo estimated signal, thereby reach the purpose of eliminating echo.

The output of subtracter is handled through Nonlinear Processing (NLP), in NLP, existed one to suppress thresholding, when the signal level of input during less than this threshold value, the signal of input will be suppressed, otherwise, the signal that makes input is passed through.

In embedded system, adaptive algorithm commonly used in the Echo Canceller is the algorithm crowd based on steepest descent method.Being represented as of this adaptive algorithm: LMS (least mean-square error) algorithm, it minimizes criterion is root-mean-square error.The amount of calculation of this adaptive algorithm is little, strong robustness, be easy to realize, in practice by extensive employing.Its shortcoming is: convergence rate is slow, and constringency performance is very sensitive to the energy variation of input signal.NLMS (energy normalized least mean-square error) algorithm is the improvement algorithm of LMS algorithm, and it has overcome the shortcoming of LMS algorithm to the input signal energy-sensitive.NLMS algorithm and its various improved forms are the adaptive filter algorithm in the Echo Canceller of main employing at present.

In the prior art, a kind of structured flowchart of the Echo Canceller that NLMS algorithm and GEIGEL algorithm realize that adopts is as shown in Figure 2.Comprise like lower module:

The voice status detector: be used to judge that the talking state of current network is far-end mode or near-end pattern, in the near-end pattern, this end subscriber is being spoken.Speech detector passes to the NLMS controller with the call status information of detected current network.

Speech detector is mainly used in the appearance of judging the near-end pattern, and detection method commonly used is the power diagnostic method, and the specific descriptions of this algorithm are following:

As long as when echo satisfies following formula 1, just be judged to be the near-end pattern that detects.

| S In(n) |>1/2*max{|R In(n) |, | R In(n-1) | ... | R In(n-N+1) | } (formula 1)

Wherein, N is the time segment value of setting, that is to say the exponent number of sef-adapting filter, shown in formula 3.

It is simple efficient adopting power diagnostic method biggest advantage, and cost performance is high.

NLMS controller: be used to obtain the call status information of the current network that speech detector passes over, then, determine whether to upgrade the sef-adapting filter in the Echo Canceller operation of current filter coefficient.

If the adaptive-filtering coefficient update algorithm when when the near-end pattern, continuing to continue to use far-end mode, sef-adapting filter can be used as cancelling out echo to the speech of near-end, finally causes sef-adapting filter to be dispersed; Echo Canceller can't normally move; Therefore, need the renewal of rejects trap coefficient this moment, only carry out Filtering Processing; After treating that the near-end pattern finishes, restart the renewal of filter coefficient again.

Therefore, the NLMS controller just stops the automatic renewal of filter coefficient after obtaining the appearance of near-end pattern, only carries out Filtering Processing.After obtaining the far-end mode appearance, then should carry out Filtering Processing and upgrade the filter coefficient processing again.To the algorithm use NLMS algorithm that the coefficient of sef-adapting filter upgrades, the specific descriptions of this algorithm are following:

W ( n + 1 ) = W ( n ) + μ ( n ) e ( n ) R In ( n ) α + R In T ( n ) R In ( n ) (formula 2)

Wherein, W (n) is illustrated in the filter coefficient vector of current time n, and being expressed as vector form is W (n)=[w 0(n), w 1(n) ..., w N-2(n), w N-1(n)] T

R In(n) be illustrated in the vector of the remote signaling in each delayer in the filter of current time n, being expressed as vector form is R In(n)=[r In(n), r In(n-1) ..., r In(n-N+2), r In(n-N+1)] T

E (n)=S In(n)-and y (n), be a scalar;

μ (n) is the variable with time correlation, import in the NLMS controller be stationary signal the time, its value μ can not change; Choosing of μ value is most important, crosses conference and causes filter divergence, and too small meeting makes convergence rate slow excessively, and choosing of μ value must be satisfied 0<μ<2.

α is the parameter of fixing, and its effect is that to avoid when the input signal vector is too small issuable denominator be zero numerical computations problem, and it is a very little positive number, can value be 0.001.

Sef-adapting filter: the structure of the horizontal sef-adapting filter that this scheme adopts is as shown in Figure 3, and wherein, the input of filter is remote signaling R In(n), output signal

Filter tap is counted delay δ (ms) decision of N by echo

N=δ * 8 (formula 3)

For example: when postponing for 90ms, tap number is taken as 720, changes over corresponding exponent number to filter tap to demands of different and just can adapt to the various network environment.

Nonlinear processor (NLP): through after the processing of above-mentioned sef-adapting filter, can not eliminate the influence of echo fully, also need be for further processing to signal through NLP.In NLP, set a certain threshold value, if the signal that NLP receives greater than this threshold value, then signal can pass through fully, output e (n); Otherwise NLP can stop the signal that receives to pass through, and the output comfort noise.

Simultaneously, when the talking state that ITU-T G168 standard-required detects current network at the voice status detector was the near-end pattern, NLP should be changed to inefficacy, in order to avoid cause the voice distortion.Only when the talking state of current network was far-end mode, it was effective just to put NLP.

The shortcoming of the implementation of the Echo Canceller of above-mentioned prior art is:

1, this scheme adopts the power diagnostic method in the voice status detection module; If the result of adjacent 2 power diagnostic methods judgement of signal is opposite, will be judged to different patterns, thereby causes the filter convergence rate to slow down; The frequent updating coefficient can't operate as normal.

2, in above-mentioned (formula 2); Make and claim that D is the energy normalized factor of NLMS algorithm, it is the quadratic sum of N sample point.It is thus clear that this scheme W of every calculating (n) need recomputate D one time, carry out N power and add operation, consuming time very big, will take the embedded system Limited resources in a large number.

3, sef-adapting filter reads in the FIR register with N input sample when work, carries out computing, whenever handles a sampling point, and all sample points are moved forward one, reads in the position, end that a new sample point deposits register in.In embedded system, so large-scale shifting function is quite consuming time.

4, when echo delay is big, the exponent number N of sef-adapting filter is with corresponding increase, and convergence rate descends, thereby causes convergence time can't satisfy real-time needs and increase operand.

5, in NLP, generate the computing of comfort noise, also increased operand and complexity to a certain extent.

In view of above problem, though above-mentioned existing echo cancellation process system possible in theory can't satisfy the requirement of embedded system for real-time, also can't adapt to the restriction of embedded system limited system resource, can't genuine productsization.

Summary of the invention

The purpose of this invention is to provide a kind of self-adaptive echo eliminator and echo cancel method thereof, thereby can eliminate echo effectively based on the digital communication system of embedded system.

The objective of the invention is to realize through following technical scheme:

A kind of self-adaptive echo eliminator comprises: voice status detector, energy normalized least mean-square error NLMS controller and sliding window FIR filter, wherein,

Voice status detector: be used for far-end speech signal and near-end voice signals are sampled; According to the far-end speech signal sample of sampling acquisition and the short-time energy estimated value of near-end voice signals sample; Confirm the call status information of current network, this call status information is passed to energy normalized least mean-square error NLMS controller;

Energy normalized least mean-square error NLMS controller: the call status information of the current network that is used for passing over according to speech detector, sliding window FIR filter is carried out the coefficient configuration;

Sliding window FIR filter: be used for coefficient according to the configuration of energy normalized least mean-square error NLMS controller, long according to the mobile position of setting, the remote signaling of inserting in the buffering area is carried out exporting after the Filtering Processing;

Said sliding window FIR filter specifically comprises:

Buffer configuration module: be used for buffering area being set, in this buffering area, insert the remote signaling sample that need carry out Filtering Processing at sliding window FIR filter;

Packing time delay configuration module: be used for confirming packing time delay sample number, dispose the fixed delay district sample length in the said buffering area according to this packing time delay sample number according to the packing time delay of the used voice compression algorithm of said remote signaling sample;

Filtering Processing module: the head that is used for the sliding window FIR register at the input remote signaling sample of FIR filter place is arranged on said buffering area; Coefficient according to the configuration of NLMS controller carries out Filtering Processing to the remote signaling sample of inserting in the said buffering area; After finishing Filtering Processing, coefficient updating operation; The afterbody of sliding window FIR register to buffering area slided one; Said remote signaling sample up to the intact frame of Filtering Processing; The FIR register is reset to the head of buffering area, with the said remote signaling sample output after the Filtering Processing;

Tap configuration module: the tap number that is used for disposing sliding window FIR filter according to the time delay size except the packing time delay that voice compression algorithm causes.

Said self-adaptive echo eliminator also comprises:

Nonlinear processor: be used for according to predefined threshold value, the residual echo signal that sliding window FIR filter is exported carries out Nonlinear Processing, further eliminates echo.

Said voice status detector specifically comprises:

Sampling processing module: be used to receive far-end speech signal, near-end voice signals, said far-end speech signal, near-end voice signals are carried out sampling processing according to the time period of setting;

The talking state judge module: if in the short-time energy estimated value of short-time energy estimated value greater than the far-end speech signal in the said setting-up time section of the sample point of the said near-end voice signals that the sampling processing module obtains during peaked setting multiple, the talking state of confirming current network is the near-end pattern; Otherwise the talking state of confirming current network is a far-end mode.

Said energy normalized least mean-square error NLMS controller specifically comprises:

Normalization factor configuration module: the initial value that the normalization factor of sliding window FIR filter input signal energy is set; Utilize this initial value to come progressively to upgrade the normalization factor of sliding window FIR filter input signal energy, the normalization factor that obtains is passed to coefficient updating module;

Coefficient updating module: the input sample value that passes in normalization factor that passes over according to the normalization factor configuration module and the sliding window FIR filter; Progressively upgrade the coefficient of sliding window FIR filter, the coefficient after this renewal is configured sliding window FIR filter.

Said Nonlinear Processing implement body comprises:

The voice output module: when the voice status detector judges that the current talking state information is the near-end pattern, former state output to the received signal;

Quiet processing module: when the voice status detector judges that the current talking state information is far-end mode, carry out quiet processing to the received signal, directly export 0 value.

Said self-adaptive echo eliminator is applicable to digital voice communication system.

A kind of adaptive echo cancellation method comprises step:

A, far-end speech signal and near-end voice signals are sampled, the far-end speech signal sample that obtains according to sampling and the short-time energy estimated value of near-end voice signals sample are confirmed the call status information of current network;

B, according to the call status information of said current network, sliding window FIR filter is carried out the coefficient configuration; Said sliding window FIR filter is long according to the mobile position of setting according to the coefficient of said configuration, and the far-end speech signal of inserting in the buffering area is carried out Filtering Processing;

Described step B specifically comprises:

B1, the initial normalization factor of sliding window FIR filter input signal energy is set, utilizes this initial normalization factor progressively to upgrade the normalization factor of sliding window FIR filter input signal energy, and then progressively upgrade the coefficient of sliding window FIR filter;

B2, in sliding window FIR filter, buffering area is set; The sliding window FIR register at the input remote signaling sample of sliding window FIR filter place is arranged on the head of said buffering area; Coefficient according to the configuration of NLMS controller carries out Filtering Processing to the remote signaling sample of inserting in the said buffering area; After finishing Filtering Processing, coefficient updating operation; The afterbody of sliding window FIR register to buffering area slided one; Said remote signaling sample up to the intact frame of Filtering Processing; The FIR register is reset to the head of buffering area, with the said remote signaling sample output after the Filtering Processing.

Described step B2 specifically comprises:

B21, according to the packing time delay of voice compression algorithm, confirm the length in fixed delay district in the buffering area; According to the tap number of the time delay size configure sliding window FIR filter except the packing time delay, confirm the length of sliding window FIR filter section in the buffering area;

B22, according to said fixed delay district, the length of sliding window FIR filter section and the frame length of speech frame, confirm buffer length, before carrying out the Filtering Processing first time, in said buffering area, insert a buffer length remote signaling sample;

B23, sliding window FIR register is arranged on the head of said buffering area, sliding window FIR filter carries out Filtering Processing according to the coefficient of configuration to the remote signaling sample of inserting in the said register; After finishing Filtering Processing, coefficient updating operation; The afterbody of sliding window FIR register to buffering area slided one; Said remote signaling sample up to the intact frame of Filtering Processing; The FIR register is reset to the head of buffering area, with the said remote signaling sample output after the Filtering Processing.

Technical scheme by the invention described above provides can be found out; The present invention is through being provided with the fixed delay buffering area in FIR (finite impulse response numeral) filter; The time delay of will packing is considered separately; Tap is not set separately, thereby can reduces the tap number of FIR filter, make the imbalance of FIR filter stable state reduce for it; Use sliding window structure FIR filter, reduce multiplication consuming time in the self-adaptive echo eliminator and shifting function significantly; Use more effective voice status detection method, improve the accuracy that the self-adaptive echo eliminator voice status detects.Thereby can eliminate echo effectively based on the digital communication system of embedded system.

Self-adaptive echo eliminator according to the invention; On identical configured hardware platform; About 1/5 of the existing self-adaptive echo eliminator that is about consuming time, CPU usage is about about 1/4 of existing self-adaptive echo eliminator, and test performance satisfies the ITU-TG.168 standard-required.When raising the efficiency, kept the high-performance of system.

Description of drawings

Fig. 1 is the basic structure sketch map of the Echo Canceller of available technology adopting self-adaptive echo counteracting technology realization;

Fig. 2 is the structural representation of the Echo Canceller of available technology adopting NLMS algorithm and the realization of GEIGEL algorithm;

Fig. 3 is the structural representation of horizontal sef-adapting filter in the prior art;

Fig. 4 is the structural representation of the embodiment of the said self-adaptive echo eliminator of this method;

Fig. 5 is the structural representation of the embodiment of sliding window FIR filter according to the invention;

Fig. 6 is the process chart of embodiment of the echo cancel method of self-adaptive echo eliminator according to the invention;

Fig. 7 is the annexation sketch map of self-adaptive echo eliminator according to the invention in digital voice communication system;

Fig. 8 is the concrete handling process sketch map of the embodiment 1 of the method for the invention.

Embodiment

The invention provides a kind of self-adaptive echo eliminator and echo cancel method thereof; Major technique characteristics of the present invention are: in the FIR filter, buffering area is set; The time delay of will packing is considered separately; For it tap is not set separately, sliding window FIR filter is long according to the mobile position of setting, and the remote signaling of inserting in the buffering area is carried out progressively Filtering Processing.

Describe the present invention in detail below in conjunction with accompanying drawing, the structure of the embodiment of the said self-adaptive echo eliminator of this method is as shown in Figure 4, comprises like lower module:

The voice status detector: the power diagnostic method after adopt improving is carried out voice status and is detected, and the talking state that detects current network is far-end mode or near-end pattern, and the call status information of detected current network is passed to the NLMS controller.Comprise: sampling processing module and talking state judge module.

Wherein, sampling processing module: be used to receive far-end speech signal, near-end voice signals, said far-end speech signal, near-end voice signals are carried out sampling processing according to the time period of setting.

Wherein, The talking state judge module: if in the short-time energy estimated value of short-time energy estimated value greater than far-end speech signal in the said setting-up time section of the sample point of the said near-end voice signals that the sampling processing module obtains during peaked setting multiple, the talking state of confirming current network is the near-end pattern; Otherwise the talking state of confirming current network is a far-end mode.

NLMS controller: be used to obtain the call status information of the current network that speech detector passes over, then, determine whether to upgrade the sef-adapting filter in the Echo Canceller operation of current filter coefficient.

When the call status information of current network is the near-end pattern, then do not carry out the coefficient updating operation of sliding window FIR filter, only control sliding window FIR filter and carry out Filtering Processing.When the call status information of current network is far-end mode, then carry out the coefficient updating operation of sliding window FIR filter, control sliding window FIR filter simultaneously and carry out Filtering Processing.NLMS algorithm after the algorithm use that the coefficient of sliding window FIR filter is upgraded is improved.Comprise: normalization factor configuration module, coefficient updating module.

Wherein, Normalization factor configuration module: the initial value that the normalization factor of filter input signal energy is set; Utilize this initial value to come progressively to upgrade the normalization factor of filter input signal energy, the normalization factor after upgrading is passed to coefficient updating module.

Wherein, coefficient updating module: the normalization factor after the renewal that passes over according to the normalization factor configuration module, progressively upgrade the coefficient of sliding window FIR filter, the coefficient after this renewal is configured sliding window FIR filter.

Sliding window FIR filter: the structure of the embodiment of the sliding window FIR filter that the present invention proposes is as shown in Figure 5.This sliding window FIR filter is used for the coefficients such as normalization factor according to the configuration of NLMS controller, the remote signaling that receives is carried out adaptive-filtering handle, and the remote signaling after handling is exported to subtracter.

The exponent number of sliding window FIR filter wherein comprises the packing time delay that is caused by voice compression coding by the decision of time delay size, and this part time delay is pure time delay, and the present invention is not provided with tap separately for it with its independent consideration, greatly saves resource.Sliding window FIR filter comprises: buffer configuration module, packing time delay configuration module, tap configuration module and Filtering Processing module.

Wherein, Buffer configuration module: be used for a buffering area being set at sliding window FIR filter; The length of this buffering area be sliding window FIR filter tap number, fixed delay sample number and speech samples frame length and, be used to insert the remote signaling sample that need carry out Filtering Processing.Before beginning to carry out the Filtering Processing first time, need treat that the remote signaling sample point fills up this buffering area.

Wherein, packing time delay configuration module: be used for packing time delay, confirm the length of fixed delay sample number in the said buffering area according to voice compression algorithm.

Wherein, tap configuration module: the tap number (being exponent number) that is used for disposing sliding window FIR filter according to the time delay size except the packing time delay.

Wherein, the Filtering Processing module: be used to control the slip of sliding window FIR register, long according to the coefficient of NLMS controller configuration according to the mobile position of setting, the remote signaling of inserting in the buffering area is carried out progressively exporting after the Filtering Processing.

Before Filtering Processing begins, sliding window FIR register is arranged at the head of buffering area, after finishing Filtering Processing, coefficient updating operation, the afterbody of sliding window FIR register to buffering area slided one.Then, after finishing next time Filtering Processing, coefficient updating operation, sliding window FIR register is slided one to the afterbody of buffering area again, up to handling a frame sample.The FIR register is reset to the head of buffering area.

Nonlinear processor: be used for according to the detected current talking state of voice status detector, further eliminate echo processing to having carried out the signal after the Filtering Processing through sliding window FIR filter.Comprise: voice output module and quiet processing module.

Wherein, the voice output module: when the voice status detector judges that the current talking state information is the near-end pattern, former state output to the received signal.

Wherein, quiet processing module: the voice status detector is judged when the current talking state information is far-end mode, carries out quiet processing to the received signal, directly exports 0 value.

The handling process of the embodiment of the echo cancel method of the said self-adaptive echo eliminator of the invention described above is as shown in Figure 6, comprises the steps:

Step 6-1: the voice status monitor detects the call status information of current network according to the short-time energy estimated value information of signal.

After the voice status detector receives far-end speech signal, near-end voice signals; This far-end speech signal, near-end voice signals are carried out sampling processing according to the time period of setting; If during peaked setting multiple, the talking state of confirming current network is the near-end pattern in the short-time energy mean value of short-time energy mean value greater than far-end speech signal in the setting-up time section of the sample point of the near-end voice signals that sampling obtains; Otherwise the talking state of confirming current network is a far-end mode.

In practical application, in case the talking state of current network has got into the near-end pattern, will keep the dual end communication of a period of time, can not switch to far-end mode at once.Therefore; Too frequent in order to overcome the filter coefficient update that exists in the existing algorithm; Cause convergence rate problem slowly; The present invention introduces the average factor of short-time energy and improves the power diagnostic method, reaches the purpose of protecting filter, enhancing robustness, and the average factor of this short-time energy is shown in following (formula 4).

S In ^ ( n ) = ( 1 - α ) * S In ^ ( n - 1 ) + α * S In ( n ) (formula 4)

Wherein, α is called forgetting factor, satisfies 0<α<1.In like manner can obtain substitution above-mentioned (formula 1), the power diagnostic method after the present invention improves is shown in following (formula 5):

| S In ( n ) ^ | > 1 2 * Max { | R In ( n ) ^ | , | R In ( n - 1 ) ^ | , . . . | R In ( n - N + 1 ) ^ | } (formula 5)

The concrete processing procedure of voice status detector is following:

Far-end sequence R in input InIn get a frame N sampled point R In(n) R In(n-1) ... R In(n-N+1), N is a sef-adapting filter length, and each sampling point in this frame is calculated its short-time energy estimated value R wherein In(n) be the sampled value of sample point, Be the short-time energy estimated value of previous sample point, α is the average factor of short-time energy, obtains Afterwards, the short-time energy estimated value of this frame sampling point is got maximum, obtain max_N.Again to the sampled point S of near-end speech current time In(n) obtain short-time energy mean value as stated above If the short-time energy of near-end input sample is average Satisfy Judge that then the current time near-end has voice, the talking state of current network is the near-end pattern; Otherwise, judge that the call status information of current network is a far-end mode.

Then, the voice status detector passes to the NLMS controller with the call status information of detected current network.

Step 6-2, NLMS controller are according to the call status information of the current network that obtains, and the normalization factor of configuration is carried out coefficient update to sliding window FIR filter.

The NLMS controller determines whether to upgrade the sef-adapting filter in the Echo Canceller operation of current filter coefficient after having obtained the call status information of the current network that speech detector passes over.

When the call status information of current network is the near-end pattern, then do not carry out the coefficient updating operation of sliding window FIR filter, only control sliding window FIR filter and carry out Filtering Processing.When the call status information of current network is far-end mode, then carry out the coefficient updating operation of sliding window FIR filter, control sliding window FIR filter simultaneously and carry out Filtering Processing.NLMS algorithm after the algorithm use that the coefficient of sliding window FIR filter is upgraded is improved.

The key of the NLMS algorithm after the above-mentioned improvement is a normalization factor of having introduced the filter input signal energy.The NLMS controller at first is provided with the normalization factor initial value, utilize this initial value control sliding window FIR filter carry out Filtering Processing, the adjustment filter coefficient after, thereby obtain new normalization factor value.According to new normalization factor value, control sliding window FIR filter carries out Filtering Processing, adjustment filter coefficient again.Afterwards, the value of current normalization factor is deducted in the current FIR register behind first sample point value, the FIR register slides backward one again, adds the new normalization factor value that square obtains of last sample value in the back FIR register that slides.Repeat said process, finish up to the Filtering Processing process.

Step 6-3: sliding window FIR filter carries out progressively Filtering Processing according to the step-length of setting near end signal, the remote signaling of inserting in the buffering area.

The present invention is provided with a buffering area in sliding window FIR filter, the length of this buffering area be fixed delay sample number, sliding window FIR filter tap number and speech frame frame length and, be used to insert the remote signaling sample that need carry out Filtering Processing.Before beginning to carry out Filtering Processing, need to wait for the remote signaling sample, this buffering area is filled up.

The tap of FIR filter among the present invention (being exponent number) is to dispose according to the time delay size except the packing time delay.

The detailed process process of sliding window FIR filter is following: divide frame with far-end speech signal according to the compressed encoding agreement, a buffering area is set, according to the length of packing time delay size, sliding window FIR filter order and speech frame frame length configuration buffering area.Beginning to carry out before the Filtering Processing remote signaling sample to be filled up buffering area the first time.Sliding window FIR register is arranged at the head of buffering area.

Then, the sample point in the sliding window FIR register is sent into the FIR filter and do Filtering Processing for the first time.After buffering; The fixed delay that far, causes owing to packing in the time delay spacing between near-end current time sample point eliminates; Be no longer necessary for the packing time delay in the filter tap is set, thereby reduced the exponent number of filter, reduce multiply operation consuming time significantly.

The step that above-mentioned sliding window FIR filter, NLMS controller are handled a frame sample is following:

1, the NLMS controller calculates initial normalization factor D.

2, according to initial normalization factor D, sliding window FIR filter carries out filtering operation, the adjustment filter coefficient.

3, the NLMS controller with the value of initial normalization factor deduct first sample value in the FIR register square.

4, the FIR register slides one to the afterbody of buffering area.

5, the NLMS controller adds the value of normalization factor the new normalization factor value that square obtains of last sample value in the current FIR register.

6, repeat 2) 3) 4) operate until handling a frame sample, data remaining in the buffering area are moved the buffering area head, simultaneously the FIR register is reset to the buffering area head.

7, read a frame new samples and insert the buffering area afterbody, repeat said process

So a large amount of shifting functions and power operation have been saved in operation, increase substantially execution efficient.

Remote signaling value after sliding window FIR filter will be handled, promptly the echo estimated value is seen off to subtracter, sees off to nonlinear processor after subtracting each other with the near end signal sample.

The talking state that step 6-4, nonlinear processor are judged according to the voice status detector is further eliminated echo processing to the received signal.

The talking state that nonlinear processor is judged according to the voice status detector is further eliminated echo processing to having carried out the signal after the Filtering Processing through sliding window FIR filter.

Compression algorithm at voice; As G.723.1 wait in the code decode algorithm, all there is one 0 input value is encoded to the processing capacity of comfort noise frame, therefore; The present invention utilizes this point in Echo Canceller; No longer repeat to generate the step of comfort noise,, directly replace output to get final product with 0 value for needing repressed sample point under the far-end mode.The concrete processing procedure of nonlinear processor is following:

When the voice status detector judges that the call mode of current network is the near-end pattern, nonlinear processor former state output to the received signal; When the voice status detector judge the call mode of current network be far-end mode constantly, then carry out quiet processing to the received signal, directly export 0 value.

The said apparatus and method of the invention described above are applicable to digital voice communication system, and the annexation sketch map of self-adaptive echo eliminator according to the invention in digital voice communication system is as shown in Figure 7.

The present invention also provides an embodiment 1 an of the method for the invention, and this embodiment 1 explains that with the echo of handling G.723.1 code encoding/decoding mode generation concrete handling process is as shown in Figure 8, comprises the steps:

1,, starts a calling with the buffering area zero clearing.The coefficients w initial value is 0.

2, G723.1 packing time delay is about 30ms (240 sample points), and frame length is 30ms (240 sample point).Put into a big buffering area to the far-end reference signal, its length is FIR tap number N+480; Before carrying out the Filtering Processing first time, need to wait for that the remote signaling sample fills up buffering area, sample in the buffering area is followed successively by x from left to right 0X N+479During beginning, sliding window FIR register-bit is x in the leftmost side 0X N-1, length is N, its content is the input of NLMS controller---and N remote signaling sampled value.

The step of handling a frame sample is following:

The initial value D=x of the calculating filter input signal energy normalized factor 0 2+ ... + x N-1 2, put into register, use during in order to the renewal filter coefficient.

3, read in the sampled value S of near end signal from the input signal of current microphone In(n).

4, calculate S In(n) short-time energy mean P Sin(i)=SIZE2*P Sin(i-1)+SIZE1*Sin (i-1) and x 1X 240The short-time energy mean P RinMaximum max (P Rin), P Rin(i)=SIZE2*P Rin(i-1)+SIZE1*Rin (i-1).Calculating filter is output as y, error of calculation sampling en=S In(i)-y.

If 5 P Sin>1/2*max (| P Rin|), the call mode of then adjudicating current network is the near-end pattern, execution in step 7; Otherwise the call mode of judgement current network is a far-end mode, execution in step 6.

6,, upgrade filter coefficient, and w is write register confession use next time according to w=w+u*en*x/ (a+D).Carry out Nonlinear Processing then.

7, calculating the output signal is out (i)=en, as the output valve of nonlinear processor.

8, the FIR register is slided one to the right, the interior content of register this moment is x 1X N

9, upgrade filter input signal energy normalized factor D:

10, repeating step 4-step 9, a frame data processing in left side finishes in buffering area, and this moment, the FIR register slid into the 240~N+479 position in the buffering area, with the head of its 240 bit recoveries to buffering area that moves to left.

11, read a frame new samples, it is inserted the N+240~N+479 position of buffering area, frame sample process circulation leaves it at that, and next again 30ms repeated said process and can handle continuously in the cycle.

The above; Be merely the preferable embodiment of the present invention, but protection scope of the present invention is not limited thereto, any technical staff who is familiar with the present technique field is in the technical scope that the present invention discloses; The variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection range of claim.

Claims (7)

1. a self-adaptive echo eliminator is characterized in that, comprising: voice status detector, energy normalized least mean-square error NLMS controller and sliding window FIR filter, wherein,
The voice status detector: being used for far-end speech signal sampled obtains the remote signaling sample; Near-end voice signals sampled obtain the near end signal sample; Short-time energy estimated value according to said remote signaling sample and near end signal sample; Confirm the call status information of current network, this call status information is passed to energy normalized least mean-square error NLMS controller;
Energy normalized least mean-square error NLMS controller: the call status information of the current network that is used for passing over according to speech detector, sliding window FIR filter is carried out the coefficient configuration;
Sliding window FIR filter: be used for coefficient according to the configuration of energy normalized least mean-square error NLMS controller, long according to the mobile position of setting, the remote signaling sample of inserting in the buffering area is carried out exporting after the Filtering Processing;
Said sliding window FIR filter specifically comprises:
Buffer configuration module: be used for buffering area being set, in this buffering area, insert the remote signaling sample that need carry out Filtering Processing at sliding window FIR filter;
Packing time delay configuration module: be used for confirming packing time delay sample number, dispose the fixed delay district sample length in the said buffering area according to this packing time delay sample number according to the packing time delay of the used voice compression algorithm of said remote signaling sample;
The Filtering Processing module: be used for sliding window FIR register is arranged on the head of said buffering area, the coefficient according to the configuration of energy normalized least mean-square error NLMS controller carries out Filtering Processing to the remote signaling sample of inserting in the said buffering area; After finishing Filtering Processing, coefficient updating operation; The afterbody of sliding window FIR register to buffering area slided one; Said remote signaling sample up to the intact frame of Filtering Processing; Sliding window FIR register is reset to the head of buffering area, with the said remote signaling sample output after the Filtering Processing;
Tap configuration module: the tap number that is used for disposing sliding window FIR filter according to the time delay size except the packing time delay that voice compression algorithm causes.
2. self-adaptive echo eliminator according to claim 1 is characterized in that, said self-adaptive echo eliminator also comprises:
Nonlinear processor: be used for according to predefined threshold value, the residual echo signal that sliding window FIR filter is exported carries out Nonlinear Processing, further eliminates echo.
3. self-adaptive echo eliminator according to claim 1 is characterized in that, said voice status detector specifically comprises:
Sampling processing module: be used to receive far-end speech signal, near-end voice signals, said far-end speech signal, near-end voice signals are carried out sampling processing according to the time period of setting;
The talking state judge module: if in the short-time energy estimated value of short-time energy estimated value greater than the far-end speech signal in the said setting-up time section of the sample point of the said near-end voice signals that the sampling processing module obtains during peaked setting multiple, the talking state of confirming current network is the near-end pattern; Otherwise the talking state of confirming current network is a far-end mode.
4. self-adaptive echo eliminator according to claim 1 is characterized in that, said energy normalized least mean-square error NLMS controller specifically comprises:
Normalization factor configuration module: the initial value that the normalization factor of sliding window FIR filter input signal energy is set; Utilize this initial value to come progressively to upgrade the normalization factor of sliding window FIR filter input signal energy, the normalization factor that obtains is passed to coefficient updating module;
Coefficient updating module: the normalization factor that passes over according to the normalization factor configuration module is progressively upgraded the coefficient of sliding window FIR filter, and the coefficient after this renewal is configured sliding window FIR filter.
5. self-adaptive echo eliminator according to claim 2 is characterized in that, said Nonlinear Processing implement body comprises:
The voice output module: when the voice status detector judges that the current talking state information is the near-end pattern, former state output to the received signal;
Quiet processing module: when the voice status detector judges that the current talking state information is far-end mode, carry out quiet processing to the received signal, directly export 0 value.
6. according to claim 1,2,3 or 4 described self-adaptive echo eliminators, it is characterized in that said self-adaptive echo eliminator is applicable to digital voice communication system.
7. an adaptive echo cancellation method is characterized in that, comprises step:
A, far-end speech signal sampled obtains the remote signaling sample; Near-end voice signals sampled obtain the near end signal sample; According to the short-time energy estimated value of said remote signaling sample and near end signal sample, confirm the call status information of current network;
B, according to the call status information of said current network, sliding window FIR filter is carried out the coefficient configuration; Said sliding window FIR filter is long according to the mobile position of setting according to the coefficient of said configuration, and the remote signaling sample of inserting in the buffering area is carried out Filtering Processing;
Described step B specifically comprises:
B1, the initial normalization factor of sliding window FIR filter input signal energy is set, utilizes this initial normalization factor progressively to upgrade the normalization factor of sliding window FIR filter input signal energy, and then progressively upgrade the coefficient of sliding window FIR filter;
B2, in sliding window FIR filter, buffering area is set; Sliding window FIR register is arranged on the head of said buffering area; Coefficient according to the configuration of energy normalized least mean-square error NLMS controller carries out Filtering Processing to the remote signaling sample of inserting in the said buffering area; After finishing Filtering Processing, coefficient updating operation; The afterbody of sliding window FIR register to buffering area slided one; Said remote signaling sample up to the intact frame of Filtering Processing; Sliding window FIR register is reset to the head of buffering area, with the said remote signaling sample output after the Filtering Processing;
Described step B2 specifically comprises:
B21, according to the packing time delay of the used voice compression algorithm of said remote signaling sample, confirm the length in fixed delay district in the buffering area; According to the tap number of the time delay size configure sliding window FIR filter except the packing time delay, confirm the length of sliding window FIR filter section in the buffering area;
B22, according to said fixed delay district, the length of sliding window FIR filter section and the frame length of speech frame, confirm buffer length, before carrying out the Filtering Processing first time, in said buffering area, insert a buffer length remote signaling sample;
B23, sliding window FIR register is arranged on the head of said buffering area, sliding window FIR filter carries out Filtering Processing according to the coefficient of configuration to the remote signaling sample of inserting in the said register; After finishing Filtering Processing, coefficient updating operation; The afterbody of sliding window FIR register to buffering area slided one; Said remote signaling sample up to the intact frame of Filtering Processing; Sliding window FIR register is reset to the head of buffering area, with the said remote signaling sample output after the Filtering Processing.
CN2006101144195A 2006-11-09 2006-11-09 Self-adaptive echo eliminator and echo eliminating method thereof CN101179294B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2006101144195A CN101179294B (en) 2006-11-09 2006-11-09 Self-adaptive echo eliminator and echo eliminating method thereof

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2006101144195A CN101179294B (en) 2006-11-09 2006-11-09 Self-adaptive echo eliminator and echo eliminating method thereof

Publications (2)

Publication Number Publication Date
CN101179294A CN101179294A (en) 2008-05-14
CN101179294B true CN101179294B (en) 2012-07-04

Family

ID=39405409

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2006101144195A CN101179294B (en) 2006-11-09 2006-11-09 Self-adaptive echo eliminator and echo eliminating method thereof

Country Status (1)

Country Link
CN (1) CN101179294B (en)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103152499B (en) * 2008-06-11 2014-12-10 三菱电机株式会社 Echo canceler
CN102118530B (en) * 2010-01-06 2013-08-14 杭州华三通信技术有限公司 Method and device for echo cancellation
CN102316231B (en) * 2010-07-08 2014-03-12 杭州华三通信技术有限公司 Echo cancellation method and device
CN101917527B (en) * 2010-09-02 2013-07-03 杭州华三通信技术有限公司 Method and device of echo elimination
CN102447795B (en) * 2010-10-09 2013-09-11 杭州华三通信技术有限公司 Echo cancellation method and device
CN102855881B (en) * 2011-06-27 2014-12-03 华为技术有限公司 Echo suppression method and echo suppression device
CN102300014A (en) * 2011-09-09 2011-12-28 南京大学 Double-talk detection method applied to acoustic echo cancellation system in noise environment
CN103391381B (en) * 2012-05-10 2015-05-20 中兴通讯股份有限公司 Method and device for canceling echo
US9288578B2 (en) * 2012-10-23 2016-03-15 Electronics And Telecommunications Research Institute Method and apparatus for canceling acoustic echo, and method for updating adaptive filter coefficient by the apparatus
CN103501375B (en) 2013-09-16 2017-04-19 华为终端有限公司 Method and device for controlling sound effect
CN104778950B (en) * 2014-01-15 2018-03-27 华平信息技术股份有限公司 A kind of microphone signal delay compensation control method based on echo cancellor
CN104539816B (en) * 2014-12-25 2017-08-01 广州华多网络科技有限公司 The intelligent sound mixing method and device of a kind of multipartite voice call
CN104616658A (en) * 2015-01-14 2015-05-13 重庆金美通信有限责任公司 Echo canceling implementing method supporting a plurality of voice coding systems
CN106571147A (en) * 2016-11-13 2017-04-19 南京汉隆科技有限公司 Method for suppressing acoustic echo of network telephone
CN108111704B (en) * 2016-11-24 2019-12-24 中国科学院声学研究所 Non-contact self-adaptive echo cancellation device and echo cancellation method
CN107040832B (en) * 2017-04-22 2019-06-11 广州市比丽普电子有限公司 A kind of handheld terminal formula microphone system based on Bluetooth technology
CN107071196B (en) * 2017-05-04 2019-09-03 重庆第二师范学院 A kind of adaptive echo cancellation method
CN107360530A (en) * 2017-07-03 2017-11-17 苏州科达科技股份有限公司 The method of testing and device of a kind of echo cancellor
CN107785026A (en) * 2017-10-18 2018-03-09 会听声学科技(北京)有限公司 A kind of delay time estimation method eliminated for set top box indoor echo

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5463618A (en) * 1993-05-28 1995-10-31 Matsushita Electric Industrial Co., Ltd. Echo canceller
CN1691716A (en) * 2004-04-23 2005-11-02 北京三星通信技术研究有限公司 Echo eliminator
CN1805011A (en) * 2005-12-23 2006-07-19 北京中星微电子有限公司 Adaptive filter method and apparatus for improving speech quality of mobile communication apparatus

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5463618A (en) * 1993-05-28 1995-10-31 Matsushita Electric Industrial Co., Ltd. Echo canceller
CN1691716A (en) * 2004-04-23 2005-11-02 北京三星通信技术研究有限公司 Echo eliminator
CN1805011A (en) * 2005-12-23 2006-07-19 北京中星微电子有限公司 Adaptive filter method and apparatus for improving speech quality of mobile communication apparatus

Also Published As

Publication number Publication date
CN101179294A (en) 2008-05-14

Similar Documents

Publication Publication Date Title
JP5450567B2 (en) Method and system for clear signal acquisition
EP0727109B1 (en) A variable block size adaptation algorithm for noise-robust acoustic echo cancellation
US6526140B1 (en) Consolidated voice activity detection and noise estimation
CN101964670B (en) Echo suppression method and apparatus thereof
CN106782593B (en) A kind of more band structure sef-adapting filter switching methods eliminated for acoustic echo
EP2235927B1 (en) Apparatus, method and program for computing control information for an echo suppression filter
Douglas Introduction to adaptive filters
US7539615B2 (en) Audio signal quality enhancement in a digital network
US8280037B2 (en) Echo canceller having its effective filter taps adaptively controlled with echo cancellation amount monitored
US7817797B2 (en) Method and apparatus for detecting echo path changes in an acoustic echo canceller
DE69728262T2 (en) Echounterprinter and nonlineine processor of an echo compensator
EP2091193B1 (en) A signal processing system, chip, external card, filter device and signal processing method
JP5122042B2 (en) System and method for near-end speaker detection by spectral analysis
US5818945A (en) Subband echo cancellation method using projection algorithm
US8098812B2 (en) Method of controlling an adaptation of a filter
EP1113588B1 (en) Echo canceling apparatus
JP4457639B2 (en) Echo canceller
JP2683490B2 (en) Adaptive noise eliminator
US6768796B2 (en) System and method for echo cancellation
US8014519B2 (en) Cross-correlation based echo canceller controllers
EP0518383B1 (en) Method and arrangement of echo elimination in digital telecommunications system
CN1914819B (en) Echo canceller with interference-level controlled step size
ES2398091T3 (en) Network Echo Compensator
EP1097562B1 (en) Adaptive filter and acoustic echo canceller using the same
EP1927192B1 (en) Method for measuring speech intelligibility

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C41 Transfer of patent application or patent right or utility model
TA01 Transfer of patent application right

Effective date of registration: 20080425

Address after: Anhui Huangshan City Mount Huangshan District Jinqiao Yuping Road South raw water association compound

Applicant after: Huangshan Haoshida Communication Technology Co., Ltd.

Address before: Beijing Chaoyang District City Anxiang Beili 11 building C

Applicant before: Plath communication technology (Beijing) Co., Ltd.

C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
EXPY Termination of patent right or utility model
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20120704

Termination date: 20141109