CN102447795B - Echo cancellation method and device - Google Patents
Echo cancellation method and device Download PDFInfo
- Publication number
- CN102447795B CN102447795B CN 201010505213 CN201010505213A CN102447795B CN 102447795 B CN102447795 B CN 102447795B CN 201010505213 CN201010505213 CN 201010505213 CN 201010505213 A CN201010505213 A CN 201010505213A CN 102447795 B CN102447795 B CN 102447795B
- Authority
- CN
- China
- Prior art keywords
- time window
- echo
- window
- screened
- screening
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
The invention provides an echo cancellation method and a device. In the invention, N time windows are arranged on a self-adaption filter, the length 1 of each time window is sufficient to cover a primary echo signal, expect the determined time windows in the N time windows, the rest time windows are screened, the time window in the time windows which are screened, which has the largest influence to echo, is determined, and the filtering coefficient of the self-adaption filter corresponding to the currently determined time windows is computed; and whether the condition that a residual signal produced in the currently-determined time windows is smaller a preset residual signal or the total number of the currently-determined time windows meets the requirement of the echo cancellation level is met is judged, if SO, the screening process of the time windows is ended, self-adpation echo cancellation processing is carried out in the determined time windows by utilizing the finally determinedfiltering coefficient, otherwise, the next screening is carried out to the time windows. According to the invention, when the multilevel echo is cancelled, the computing quantity is reduced and the convergence rate is improved.
Description
Technical field
The present invention relates to the signal processing technology field, particularly a kind of method and apparatus of echo elimination.
Background technology
Echo is voice signal when transmission, because impedance does not match and produces, for example 2/4 line conversion hybrid coil imbalance can cause electric current to leak, thereby causes a part of signal energy to be reflected back toward signal source and then produce echo.If voice signal then can produce multi-stage echo through the unmatched 2/4 line conversion of impedance repeatedly hybrid coil.Owing to have IP network storage forward delay, coding delay and decoding delay, echo delay is bigger usually in VOIP, and when echo delay was worked energetically 100ms, the user can obviously feel and causes speech quality to descend by echo.
As shown in Figure 1, when side a and b is conversed, the voice signal of A in transferring to the process of B end since 2/4 line conversion hybrid coil do not match can cause the part speech signal energy to be reflected can the A end, the A end subscriber has just been heard the echo-signal of own voice like this.Because to transmit through IP in the middle of the voice signal, Code And Decode could arrive the B end, and the echo-signal that produces need be transmitted through IP equally, Code And Decode could arrive A and hold, it is very big that this process often postpones, and causes A to hold to feel tangible echo.If as shown in fig. 1, hybrid coil impedance between voice gateways 2 and the PBX does not match, then can produce the one-level echo-signal, if the hybrid coil impedance between PBX and the B end phone does not match, then can produce the one-level echo-signal again, be referred to as the generation multi-stage echo in this situation that will produce the above echo-signal of two-stage.
Fig. 2 is the schematic diagram that concerns of multi-stage echo signal, as shown in Figure 2, the voice-activated signal that a is ordered successively b point, c point ..., the z point produces the multi-stage echo signal, concrete progression is determined by the topological structure of communication link.Time interval Delay1, Delay2 between voice signal and the echo-signal ..., DelayN determines by the concrete transmission characteristic of communication link.The multi-stage echo signal that produces at b point, c point, z point again elapsed time interval D elay1, Delay2 ..., DelayN transmit back the voice signal input, and in A point generation acting in conjunction result, as shown in Figure 3.As the above analysis, the echo-signal of particular point in time is by the coefficient result of voice signal of some of special time period less zone time of sections at interval with it.
The method for echo cancellation that carries out in the prior art is in long time window voice signal of far-end to be gathered, the length of this time window guarantees to cover the echoes at different levels of far-end speech signal, the filter factor of sef-adapting filter is determined and upgraded to the voice signal of the far-end that utilization collects, usually the training process that the definite and renewal process of filter factor is called filter factor, sef-adapting filter behind the filter factor of employing renewal afterwards carries out the filtering processing to the input signal of near-end, thereby eliminates echo.
Yet because single long window is when retardation ratio is big between multi-stage echo especially echo delay echoes bigger or at different levels, length of window must be very long, sampled point quantity is many, strengthened amount of calculation, many irrelevant numerical value have also been participated in the training process of filter factor, and convergence rate is very slow.
Summary of the invention
The invention provides the method and apparatus that a kind of echo is eliminated, so that when eliminating multi-stage echo, reduce amount of calculation and improve convergence rate.
The method that a kind of echo is eliminated, for sef-adapting filter arranges N time window, the length l of each time window is enough to cover the one-level echo-signal, and this method comprises:
A, after the filter factor training process starts, the fixed time window of original records be sky;
B, each time window except all fixed time windows in the described N time window is carried out epicycle screening, the filtering result of on all fixed time window bases, each screened time window being carried out according to sef-adapting filter, determine the time window that in the screened time window of epicycle echo had the greatest impact, calculate the filter factor of the sef-adapting filter of current fixed time window correspondence;
C, judge whether to satisfy: the residual signals that produces at current fixed time window satisfies echo less than default residual signals or current fixed time window total number and eliminates the progression requirement, if, execution in step D; Otherwise go to execution in step B and carry out the next round screening;
D, finish the screening process to time window, sef-adapting filter utilizes last to take turns definite filter factor to carry out the self adaptation echo eliminate and handle in each time window of determining.
The device that a kind of echo is eliminated, for sef-adapting filter arranges N time window, the length l of each time window is enough to cover the one-level echo-signal, and this device comprises:
The training start unit is used for after the filter factor training process starts, and the fixed time window of original records is empty, and notification window screening unit begins the screening process to time window;
Window screening unit, be used for a described N time window each time window except all fixed time windows is screened, according to the filtering result that sef-adapting filter carries out each screened time window, determine the time window that in the screened time window of epicycle echo is had the greatest impact on all fixed time window bases; After receiving end notification, finish the screening process to time window, make described sef-adapting filter utilize last to take turns definite filter factor and in each time window of determining, carry out self adaptation echo elimination processing;
Coefficient update unit when being used for determining the time window that the epicycle echo is had the greatest impact in described window screening unit, is calculated the filter factor of the sef-adapting filter of current fixed time window correspondence;
The condition judgment unit, when being used for determining the time window that echo is had the greatest impact in window screening unit, judge whether to satisfy: the residual signals that produces at current fixed time window satisfies the requirement of echo elimination progression less than default residual signals or current fixed time window total number, if, send end notification to described window screening unit, otherwise, notify described window screening unit to begin each time window except all fixed time windows in the described N time window is carried out the next round screening.
As can be seen from the above technical solutions, the invention provides and adopt the alternative long-time window of a plurality of short time windows to carry out the training of filter factor, satisfy the implementation that echo is eliminated the filter factor that requires to find out, because the quantity of sampled point descends greatly in each short time window, amount of calculation also descends thereupon greatly.Meanwhile, because each short time window has been removed a large amount of irrelevant voice signals, reduced the training of filter factor and disturbed, convergence rate is improved.
Description of drawings
Fig. 1 is the principle schematic that echo produces in the VOIP network;
Fig. 2 is the schematic diagram that concerns of multi-stage echo signal;
Fig. 3 is the acting in conjunction result schematic diagram of multi-stage echo shown in Figure 2;
Fig. 4 eliminates principle schematic for echo;
Fig. 5 is main method flow chart provided by the invention;
Fig. 6 is the specific implementation flow chart of step 502 among Fig. 5;
Fig. 7 is apparatus structure schematic diagram provided by the invention;
Fig. 8 is prior art and contrast schematic diagram of the present invention.
Embodiment
In order to make the purpose, technical solutions and advantages of the present invention clearer, describe the present invention below in conjunction with the drawings and specific embodiments.
The understanding of the present invention is at first simply introduced algorithm filter and the principle of available technology adopting for convenience.
The principle that echo is eliminated is: utilize remote signaling to estimate the echo that remote signaling produces, adopt sef-adapting filter near end signal to be leached the echo of estimation at near-end, thereby reach the purpose that echo is eliminated, wherein, sef-adapting filter can be adjusted its filter factor automatically according to the feature of remote signaling.Fig. 4 eliminates principle schematic for echo, and as shown in Figure 4, at current time n, remote end input signal is x (n), and the echo signal that remote end input signal produces is r (n), and the echo signal of estimation is
The voice signal of near-end is y (n), that is to say, near end input signal is y (n)+r (n).Through the residual signals that produces after the sef-adapting filter filtering
Wherein, the filter factor W=[w of the sef-adapting filter of adaptive filter algorithm generation
1, w
2... w
M]
T, remote end input signal is: X (n)=[x
1(n), x
2(n) ... x
M(n)]
T, so,
Wherein M is the exponent number of sef-adapting filter.Adaptive filter algorithm commonly used at present is normalization minimum mean-square poor (NLMS) algorithm, and its filter factor recurrence formula is:
Wherein: μ is the iteration step length constant of filter factor correction, and δ is default positive number constant.As shown from the above formula, the amount of calculation of each iteration is relevant with M, and M more intensive is more big, and the value of M becomes original n doubly, and amount of calculation then also becomes original n doubly.
What echo filtering of the prior art was adopted is single long window, in this long window, gather remote end input signal and residual signals, use above-mentioned filter factor recurrence formula finally to determine the filter factor of sef-adapting filter, adopt this filter factor that the near end input signal in the communication process is carried out filtering.
The inventor of this patent is the remote end input signal of several fixedly short period sections by what analyze to find certain some echo-signal is exerted an influence, and in communication process, the number of slices of remote end input signal that influences these echo-signals is constant, and and echo-signal at different levels between time difference constant.Therefore, can consider to adopt a plurality of short windows to substitute long window.
Fig. 5 is main method flow chart provided by the invention, for sef-adapting filter arranges N time window, wherein the length l of each time window is enough to cover the one-level echo-signal, for the long-time window L that needs to guarantee to cover echoes at different levels in the prior art, l＜L among the present invention, even can l＜＜L.Usually can be set to 2 times of one-level echo-signal, for example, the impulse response duration is 1ms, and each of its generation grade echo-signal is 1ms, and the length l that time window can be set so is 2ms.
As shown in Figure 5, this method can may further comprise the steps:
Step 501: after the filter factor training process started, the fixed time window of initialization was empty.
Because the present invention will screen the filter factor that obtains the sef-adapting filter optimum by at least one wheel the to time window, taking turns screening at each all needs to determine this to take turns in the screening window (being the optimal time window) that echo has the greatest impact is to have determined time window, therefore, when starting the filter factor training process at every turn, initial condition that need fixed time window is set to sky, so as follow-up by training process adopt determine each take turns in optimum time window.
Step 502: each time window except all fixed time windows in the above-mentioned N time window is carried out the epicycle screening, the filtering result of on all fixed time window bases, each screened time window being carried out according to sef-adapting filter, determine the time window that in the screened time window echo had the greatest impact, calculate and record the filter factor of current fixed time window correspondence.
Wherein, the filtering that sef-adapting filter carries out each screened time window on all fixed time window bases is treated to: the remote signaling that sef-adapting filter utilizes fixed time window and current screened time window to sample carries out filtering to be handled, and is being fixed time window corresponding sampling points and current screened time window corresponding sampling points sum to each screened time window place sampled point namely.
The filter factor that calculates current definite time window correspondence is: begin in current all fixed time windows the filter factor that the iterative formula according to formula (1) calculates with default filter factor initial value.
Step 503: judge whether to satisfy: the residual signals that produces at current fixed time window satisfies the requirement of echo elimination progression less than default target residual error or current fixed window number, if, execution in step 504; Otherwise, go to execution in step 502 and carry out the next round screening.
Step 504: finish the screening process to time window, sef-adapting filter utilizes last to take turns definite filter factor to carry out the self adaptation echo eliminate and handle in each time window of determining.
In this step, suppose the time window of determining be designated T1, T2 ..., Tn, last takes turns definite filter factor is W, wherein each time window corresponding part in matrix W be respectively W1, W2 ..., Wn, when then carrying out echo elimination processing, utilize the interior remote signaling of fixed time window and last to take turns the part that each time window is corresponding in definite filter factor and be weighted processing back formation echo estimate signal respectively
Namely adopt W1 as the weight coefficient of remote signaling in the T1 time window, W2 is as the weight coefficient of remote signaling in the T2 time window, and the rest may be inferred, and Wn is as the weight coefficient of remote signaling in the Tn time window.
Said process provided by the invention screens each time window in fact exactly, filters out the definite time window of time window conduct that echo is had the greatest impact, and calculates and record the filter factor of current fixed time window correspondence; If do not satisfy at least one in two conditions in the step 503, then on fixed time window basis, remaining time window is proceeded screening, at least one in satisfying above-mentioned two conditions.Utilize the filter factor of final each time window correspondence of determining to carry out follow-up echo elimination processing.
Remote end input signal can had only, and do not have voice signal when input of near-end, start the filter factor training process, in the step 502 of this process, determine that the time window that in the screened time window echo is had the greatest impact can be specially: when the voice signal input that does not have near-end only has the voice signal input of far-end, in the filtering result of determining on the time window basis each screened time window to be carried out, determine the time window of time window for echo is had the greatest impact of residual signals minimum according to sef-adapting filter.
Said method of the present invention makes echo eliminate and handles the processing that comprises three aspects: window screening process, filtering parameter training process and filtering.Wherein, the window screening process that focuses on of the present invention, this window screening process is included in the filtering parameter training process, is accompanied by the filtering of sef-adapting filter simultaneously in the window screening process.After determining the filter factor of sef-adapting filter by the window screening, the window screening process finishes, but filtering parameter training process and filtering still can continue in subsequent process, and namely sef-adapting filter still can carry out adaptive parameter adjustment and filtering processing.
Specific implementation process to step 502 in the said method is described below, and as shown in Figure 6, this process can specifically may further comprise the steps:
Step 601: begin screening from first time window.
Step 602: judge current screened time window whether with all fixed time windows exist overlapping, if, execution in step 603; Otherwise, execution in step 604.
Step 603: begin the screening of next time window, go to step 602.
Step 604: in the filtering result of determining on the time window basis current screened time window to be carried out, judge that whether the residual signals that produces is less than default target residual error, if then execution in step 605 according to sef-adapting filter; Otherwise, execution in step 606.
Step 605: the direct time window that has the greatest impact that current screened time window is determined as epicycle, utilize and currently determined that time window calculates the filter factor of sef-adapting filter, directly go to the step 504 in the execution graph 5.
Carry out in the process of filtering processing at fixed time window and current screened time window at sef-adapting filter, in case the residual signals of determining to produce is less than default target residual error, explanation determined that before epicycle the filter factor of the sef-adapting filter that time window is corresponding with this screened time window can satisfy final echo and eliminate requirement, can directly jump out the screening process of time window, directly execution in step 504.
Need to prove, a time window domestic demand otherwise broken needle carries out iterative computation to sef-adapting filter, surpass the greatest iteration step number of setting until the iteration step number, then finish to handle in the filtering of current time window.
Step 606: judge the filtering result of the time window whether filtering result that sef-adapting filter carries out current screened time window is better than recording on all fixed time window bases, if, execution in step 607; Otherwise, execution in step 608.
Need to prove, take turns in the screening process at each that the screening that each time window is carried out is actually all to be carried out filtering at fixed time window and current screened time window and handle.Give one example, suppose in first round screening process, the window ID of determining that echo is had the greatest impact is T1, in second screening process of taking turns, the screening of each time window is actually sampled point in the basis of time window T1 utilizes each screened time window carries out filtering and handle, namely the sampled point in each screened time window becomes: the sampled point in the sampled point in the time window T1 and the current screened time window.If determine that according to the filtering result of each screened time window second window ID that echo is had the greatest impact in taking turns is T2, then in the screening process of third round, the screening of each time window is actually sampled point in each the screened time window of record of the basis of time window T1 and T2 carries out filtering and handle, namely the sampled point in each screened time window becomes: the sampled point in the sampled point in the time window T1, the sampled point in the time window T2 and the current screened time window.
This step is just judged on fixed time window basis the residual signals whether residual signals that current screened time window is produced produces less than the time window that has recorded.
Step 607: the filtering result of utilizing current screened time window is the filtering result of the time window of new record more.
Step 608: judge whether to have screened all time windows, if not, execution in step 609; If, execution in step 610.
Step 609: begin the screening of next time window, go to execution in step 602.
More preferably, the original position of adjacent two time windows can be set in the present invention to the length of multi-compartment one-level echo-signal.Length with interval one-level echo-signal is example, when the screening of the next time window of beginning, can move the length of an echo-signal from current screened time window backward, for example, when echo-signal length is 1ms, when time window length is 2ms, can begin to move backward initial as next time window of half time window from current screened time window.
Step 610: the time window of current record is defined as the time window that in the screened time window echo had the greatest impact, calculates the filter factor of the sef-adapting filter of current definite time window correspondence.
So far, the flow process of step 502 finishes.
In the present invention, can the voice signal of far-end and near-end input be detected, after the conversation of far-end and near-end is set up, if far-end input speech signal and near-end do not have the voice signal input, then start above-mentioned filtering parameter training process, and the filter factor that adopts the filtering parameter training process to produce carries out echo elimination processing near end input signal in this communication process.If near-end input speech signal and far-end does not have voice signal does not then carry out echo and eliminates to handle and do not carry out the filtering parameter training yet.If near-end and far-end have the voice signal input, then only near end input signal is carried out echo and eliminate processing and do not carry out the filtering parameter training.
In addition, need to prove, in the present invention, it is adjustable that echo is eliminated progression, and namely the user can be according to concrete topological structure and electrical conditions, echo is set flexibly eliminates the progression requirement, for example, when only needing to carry out the elimination of one-level echo, it is one-level that echo elimination progression is set, then only carry out taking turns the window screening process and can determine the filter factor that meets the demands, accelerate convergence process.The ifs circuit topology causes multi-stage echo, the echo that meets the physical circuit topology then is set eliminates the progression requirement, and the effect that makes echo eliminate reaches optimization effect as far as possible.
Above-mentionedly determine that by window screening echo eliminates the process of employed filter factor and can set up the startup that be triggered when the back start the filter factor training process for the first time in conversation, also can eliminate the startup that is triggered when the progression requirement changes or the target residual error changes at echo, the mode that can certainly exist other to be triggered and to start, for example triggered startup etc. by hand by the user, concrete triggering mode the present invention is not limited.
More than be the description that method provided by the present invention is carried out, below device provided by the present invention be described in detail.Fig. 7 is apparatus structure schematic diagram provided by the invention, and as shown in Figure 7, this device comprises: training start unit 700, window screening unit 710, coefficient update unit 720 and condition judgment unit 730.
In addition, this device can also comprise: speech detection unit 740, be used for after far-end and near-end are set up voice call, near-end does not have the voice signal input if detect that far-end has voice signal input, then triggering training start unit 700 starts the filter factor training process, and triggers sef-adapting filter near end input signal is carried out echo elimination processing; Near-end has the voice signal input if detect that far-end does not have voice signal input, then forbids training start unit 700 to start the filter factor training process, and forbids that sef-adapting filter carries out echo near end input signal and eliminates and handle; If detect far-end and near-end the voice signal input is arranged, then forbid training start unit 700 to start the filter factor training process, trigger sef-adapting filter near end input signal is carried out echo elimination processing.
Wherein, near end input signal comprises: the voice signal of the echo-signal that the voice signal of far-end input produces and near-end input.
Particularly, voice signal input is arranged and near-end when not having the voice signal input at far-end, the filtering result that window screening unit 710 carries out each screened time window on all fixed time window bases according to sef-adapting filter determines that the time window of generation residual signals minimum is the time window that in the screened time window of epicycle echo is had the greatest impact.
Concrete structure to window screening unit 710 is introduced below, and this window screening unit 710 can comprise: window screening subelement 711, first judgment sub-unit 712, second judgment sub-unit 713, the 3rd judgment sub-unit 714, the first record subelement 715 and the second record subelement 716.
First judgment sub-unit 712, be used for judging current screened time window whether with the time window of the first record subelement, 715 records exist overlapping, if send displacement to window screening subelement 711 and notify; Otherwise, send first to second judgment sub-unit 713 and judge notice.
Second judgment sub-unit 713, after being used for receiving the first judgement notice, judge whether the filtering result that sef-adapting filter carries out current screened time window is better than the filtering result that the second record subelement 716 records on all fixed time window bases, if, filtering result and the current screened time window of current screened time window are sent to the second record subelement 716, and send second to the 3rd judgment sub-unit 714 and judge notice.
The 3rd judgment sub-unit 714, be used for receiving second judge notice after, judge whether window screening subelement 711 has screened all time windows, if not, send to be shifted to window screening subelement 711 and notify; If the time window of the current second record subelement record is defined as the time window that in the screened time window of epicycle echo is had the greatest impact.
The first record subelement 715 is used for the time window that record the 3rd judgment sub-unit 714 is determined.
The second record subelement 716, the filtering result and the time window that be used for to utilize the filtering result that receives and time window to upgrade current record.
More preferably, this window screening unit 710 can also comprise: the 4th judgment sub-unit 717 between first judgment sub-unit 712 and second judgment sub-unit 713, be used for receiving the first judgement notice that first judgment sub-unit 712 sends, the filtering result of on all fixed time window bases, current screened time window being carried out according to sef-adapting filter, whether the judgement residual signals that current screened time window produces on all fixed time window bases is less than default target residual error, if, send update notification to coefficient update unit 720, and send end notification to window screening unit 710; Otherwise, judge that with first notice sends to second judgment sub-unit 713.
On the said apparatus basis, sef-adapting filter utilizes the remote signaling in fixed each time window of this device to take turns after each fixed time window is corresponding in definite filter factor part is weighted processing respectively the formation echo estimate signal with last; Near end input signal is filtered out the echo estimate signal of corresponding sampled point.
The involved sef-adapting filter of method and apparatus of the present invention can adopt the NLMS self adaptation to consider the ripple device.
Fig. 8 is the contrast schematic diagram of prior art and method provided by the present invention, in the prior art, in the long window of all progression echo-signals of covering, carry out the training of filter factor, and will long window in method provided by the present invention be divided into the time window of N weak point, wherein, exist overlapping between the time window, as shown in Figure 8, per two unit interval sections constitute a time window, for example unit interval section 1 and 2 constitutes window 1, unit interval section 2 and 3 constitutes window 2, and unit interval section 3 and 4 constitutes window 3, and the rest may be inferred.Find out the time window that can satisfy filtering requirements by the window screening process and carry out the training of filter factor, be example to filter out three time windows among Fig. 8, utilize the remote signaling of these three time windows samplings and filter factor that these three time windows calculate to carry out echo and eliminate and handle.
By above description as can be seen, method and apparatus provided by the invention can possess following advantage:
1) the invention provides the alternative long-time window of a plurality of short time windows of employing and carry out the training of filter factor, satisfy the implementation that echo is eliminated the filter factor that requires to find out, because the quantity of sampled point descends greatly in each short time window, amount of calculation also descends thereupon greatly.Meanwhile, because each short time window has been removed a large amount of irrelevant voice signals, reduced the training of filter factor and disturbed, convergence rate is improved.
2) by the present invention, can the progression that echo is eliminated be set flexibly according to topological structure and the electric characteristic of near-end analog circuit, echo is eliminated can be adapted to various application scenarioss well.
The above only is preferred embodiment of the present invention, and is in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of making, is equal to replacement, improvement etc., all should be included within the scope of protection of the invention.
Claims (13)
1. the method that echo is eliminated is characterized in that for sef-adapting filter arranges N time window, the length l of each time window is enough to cover the one-level echo-signal, and this method comprises:
A, after the filter factor training process starts, the fixed time window of original records be sky;
B, each time window except all fixed time windows in the described N time window is carried out epicycle screening, the filtering result of on all fixed time window bases, each screened time window being carried out according to sef-adapting filter, determine the time window that in the screened time window of epicycle echo had the greatest impact, calculate the filter factor of the sef-adapting filter of current fixed time window correspondence;
C, judge whether to satisfy: the residual signals that produces at current fixed time window satisfies echo less than default residual signals or current fixed time window total number and eliminates the progression requirement, if, execution in step D; Otherwise go to execution in step B and carry out the next round screening;
D, finish the screening process to time window, sef-adapting filter utilizes last to take turns definite filter factor to carry out the self adaptation echo eliminate and handle in each time window of determining.
2. method according to claim 1 is characterized in that, after far-end and near-end were set up voice call, far-end had the voice signal input and near-end when not having the voice signal input, and described filter factor training process starts;
When far-end does not have voice signal input and near-end when the voice signal input is arranged, then neither carry out described filter factor training process and also near end input signal is not carried out echo and eliminate and handle;
When far-end and near-end have the voice signal input, only near end input signal is carried out echo and eliminate and handle and do not carry out the filtering parameter training process;
Wherein, described near end input signal comprises: the voice signal of the echo-signal that the voice signal of far-end input produces and near-end input.
3. method according to claim 1, it is characterized in that, the time window that in the screened time window of described definite epicycle echo is had the greatest impact is: the voice signal input is arranged and near-end when not having the voice signal input at far-end, according to the filtering result that sef-adapting filter carries out each screened time window, determine that the time window of generation residual signals minimum is the time window that in the screened time window of epicycle echo is had the greatest impact on all fixed time window bases.
4. according to claim 1,2 or 3 described methods, it is characterized in that described step B specifically comprises:
B1, from first time window begin the screening;
B2, judge current screened time window whether with all fixed time windows exist overlapping, if, execution in step B3; Otherwise, execution in step B4;
B3, begin the screening of next time window, go to step B2;
B4, judge the filtering result of the time window whether filtering result that sef-adapting filter carries out current screened time window is better than recording on all fixed time window bases, if, execution in step B5; Otherwise execution in step B6;
B5, the filtering result filtering result of the time window of new record more of utilizing current screened time window;
B6, judge whether to have screened all time windows, if not, execution in step B7; If, execution in step B8;
The screening of B7, the next time window of beginning goes to execution in step B2;
B8, the time window of current record is defined as the time window that in the screened time window of epicycle echo had the greatest impact, calculates the filter factor of the sef-adapting filter of current fixed time window correspondence.
5. method according to claim 4 is characterized in that,, and before execution in step B4, also comprises for not in the judged result of step B2:
B9, the filtering result of on all fixed time window bases, current screened time window being carried out according to sef-adapting filter, whether the judgement residual signals that current screened time window produces on all fixed time window bases is less than default target residual error, if, execution in step B10 then; Otherwise, continue execution in step B4;
B10, directly determine that current screened time window is the time window that epicycle is determined, calculate the filter factor of current fixed time window correspondence, directly go to step D.
6. according to claim 1,2 or 3 described methods, it is characterized in that the original position of adjacent two time windows is to the length of multi-compartment one-level echo-signal.
7. according to claim 1,2 or 3 described methods, it is characterized in that, utilize last to take turns definite filter factor among the step D and in time window that each is determined, carry out echo and eliminate to handle and specifically comprise:
Utilize the remote signaling in fixed each time window to take turns after each fixed each time window is corresponding in definite filter factor part is weighted processing respectively the formation echo estimate signal with last; Near end input signal is filtered out the echo estimate signal of corresponding sampled point.
8. the device that echo is eliminated is characterized in that for sef-adapting filter arranges N time window, the length l of each time window is enough to cover the one-level echo-signal, and this device comprises:
The training start unit is used for after the filter factor training process starts, and the fixed time window of original records is empty, and notification window screening unit begins the screening process to time window;
Window screening unit, be used for a described N time window each time window except all fixed time windows is screened, according to the filtering result that sef-adapting filter carries out each screened time window, determine the time window that in the screened time window of epicycle echo is had the greatest impact on all fixed time window bases; After receiving end notification, finish the screening process to time window, make described sef-adapting filter utilize last to take turns definite filter factor and in each time window of determining, carry out self adaptation echo elimination processing;
Coefficient update unit when being used for determining in described window screening unit the time window that epicycle has the greatest impact to echo, is calculated the filter factor of the sef-adapting filter of current fixed time window correspondence;
The condition judgment unit, when being used for determining the time window that echo is had the greatest impact in window screening unit, judge whether to satisfy: the residual signals that produces at current fixed time window satisfies the requirement of echo elimination progression less than default residual signals or current fixed time window total number, if, send end notification to described window screening unit, otherwise, notify described window screening unit to begin each time window except all fixed time windows in the described N time window is carried out the next round screening.
9. device according to claim 8, it is characterized in that, this device also comprises: the speech detection unit, be used for after far-end and near-end are set up voice call, near-end does not have the voice signal input if detect that far-end has voice signal input, then triggers described training start unit to start the filter factor training process; Near-end has the voice signal input if detect that far-end does not have voice signal input, forbids that then described training start unit starts the filter factor training process, and forbids that described sef-adapting filter carries out echo near end input signal and eliminates and handle; If detect far-end and near-end the voice signal input is arranged, forbid that then described training start unit starts the filter factor training process, trigger described sef-adapting filter near end input signal is carried out echo elimination processing;
Wherein, described near end input signal comprises: the voice signal of the echo-signal that the voice signal of far-end input produces and near-end input.
10. device according to claim 8, it is characterized in that, voice signal input is arranged and near-end when not having the voice signal input at far-end, the filtering result that described window screening unit carries out each screened time window on all fixed time window bases according to sef-adapting filter determines that the time window of generation residual signals minimum is the time window that in the screened time window of epicycle echo is had the greatest impact.
11. according to Claim 8,9 or 10 described devices, it is characterized in that described window screening unit specifically comprises:
Window screening subelement is used for beginning screening from first time window; After receiving the displacement notice, begin the screening of next time window;
First judgment sub-unit, be used for judging current screened time window whether with the time window of the first record subelement record exist overlapping, if send displacement to described window screening subelement and notify; Otherwise, send first to second judgment sub-unit and judge notice;
Second judgment sub-unit, after being used for receiving the described first judgement notice, judge whether the filtering result that sef-adapting filter carries out current screened time window is better than the filtering result that the second record subelement records on all fixed time window bases, if, filtering result and the current screened time window of current screened time window are sent to the described second record subelement, and send second to the 3rd judgment sub-unit and judge notice;
The 3rd judgment sub-unit, be used for receiving described second judge notice after, judge whether described window screening subelement has screened all time windows, if not, send displacement to described window screening subelement and notify; If the time window of the current second record subelement record is defined as the time window that in the screened time window of epicycle echo is had the greatest impact;
The first record subelement is used for the time window that record the 3rd judgment sub-unit is determined;
The second record subelement is used for utilizing the filtering result that receives to upgrade the filtering result of current record.
12. device according to claim 11 is characterized in that, described window screening unit also comprises:
The 4th judgment sub-unit between described first judgment sub-unit and described second judgment sub-unit, be used for receiving the first judgement notice that described first judgment sub-unit sends, the filtering result of on all fixed time window bases, current screened time window being carried out according to sef-adapting filter, whether the judgement residual signals that current screened time window produces on all fixed time window bases is less than default target residual error, if, send update notification to described coefficient update unit, and send end notification to described window screening unit; Otherwise, judge that with described first notice sends to described second judgment sub-unit;
Described coefficient update unit, after being used for receiving described update notification, determine that directly current screened time window is the time window that epicycle is determined, calculate the filter factor of current fixed time window correspondence, send end notification to described window screening unit.
13. 9 or 10 described devices according to Claim 8,, it is characterized in that, sef-adapting filter utilizes the remote signaling in fixed each time window to take turns after each fixed each time window is corresponding in definite filter factor part is weighted processing respectively the formation echo estimate signal with last; Near end input signal is filtered out the echo estimate signal of corresponding sampled point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201010505213 CN102447795B (en) | 2010-10-09 | 2010-10-09 | Echo cancellation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201010505213 CN102447795B (en) | 2010-10-09 | 2010-10-09 | Echo cancellation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102447795A CN102447795A (en) | 2012-05-09 |
| CN102447795B true CN102447795B (en) | 2013-09-11 |
Family
ID=46009866
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201010505213 Active CN102447795B (en) | 2010-10-09 | 2010-10-09 | Echo cancellation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102447795B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106597402B (en) * | 2016-11-23 | 2019-03-19 | 天津津航计算技术研究所 | A kind of adaptive Moveing target indication method based on statistical law |
| CN109451195A (en) * | 2018-09-18 | 2019-03-08 | 北京佳讯飞鸿电气股份有限公司 | A kind of echo cancel method and system of adaptive double-end monitor |
| CN109862200B (en) * | 2019-02-22 | 2021-02-12 | 北京达佳互联信息技术有限公司 | Voice processing method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6751313B2 (en) * | 2001-10-31 | 2004-06-15 | Conexant Systems, Inc. | Method and apparatus for an improved echo canceller |
| EP1827002A1 (en) * | 2006-02-22 | 2007-08-29 | Alcatel Lucent | Method of controlling an adaptation of a filter |
| CN101179294A (en) * | 2006-11-09 | 2008-05-14 | 爱普拉斯通信技术（北京）有限公司 | Self-adaptive echo eliminator and echo eliminating method thereof |
-
2010
- 2010-10-09 CN CN 201010505213 patent/CN102447795B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6751313B2 (en) * | 2001-10-31 | 2004-06-15 | Conexant Systems, Inc. | Method and apparatus for an improved echo canceller |
| EP1827002A1 (en) * | 2006-02-22 | 2007-08-29 | Alcatel Lucent | Method of controlling an adaptation of a filter |
| CN101179294A (en) * | 2006-11-09 | 2008-05-14 | 爱普拉斯通信技术（北京）有限公司 | Self-adaptive echo eliminator and echo eliminating method thereof |
Non-Patent Citations (2)
| Title |
|---|
| 王权梅，李灿平，李梅娜.《VOIP的相关技术-回声消除技术的研究》.《电脑与信息技术》.2003,(第3期),全文. * |
| 蒙淑艳,赵晓晖,顾海军.《一种自适应回声抵消和噪声消除算法》.《吉林大学学报(工学版)》.2004,第34卷(第4期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102447795A (en) | 2012-05-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101026659B (en) | Method for realizing echo time delay positioning | |
| CN101087320B (en) | Improved method and apparatus for detecting echo path changes in an acoustic echo canceller | |
| CN101179294B (en) | Self-adaptive echo eliminator and echo eliminating method thereof | |
| KR100269836B1 (en) | Apparatus and method for noise reduction for a full-duplex speakerphone or the like | |
| CN1115795C (en) | Echo canceller with silence detection | |
| CN103391381B (en) | Method and device for canceling echo | |
| EP1883213B1 (en) | System and method for calibrating a hands-free system | |
| CN1118956C (en) | Echo path delay estimation | |
| JP2007179073A (en) | Voice activity detecting device, mobile station, and voice activity detecting method | |
| US8290141B2 (en) | Techniques for comfort noise generation in a communication system | |
| Heitkamper | An adaptation control for acoustic echo cancellers | |
| CN1286862A (en) | Method and apparatus for providing comfort noise in communications system | |
| CN102447795B (en) | Echo cancellation method and device | |
| EP1208689A2 (en) | Acoustical echo cancellation device | |
| RU2009136562A (en) | METHOD AND DEVICE FOR CONTROLING SMOOTHING OF STATIONARY BACKGROUND NOISE | |
| CN101026392A (en) | Method of controlling an adaptation of a filter | |
| CN110995951B (en) | Echo cancellation method, device and system based on double-end sounding detection | |
| CN1157000C (en) | Adaptive step length of echo canceller and method for carrying out same | |
| CN105391879A (en) | Echo residue-free double-end communication robust acoustic echo elimination method | |
| RU2180984C2 (en) | Convergence measurement technique for adaptive filters | |
| WO2019028115A1 (en) | Mitigating impact of double talk for residual suppressors | |
| CN105409191A (en) | Echo removal | |
| CN1650537A (en) | Echo cancellation | |
| CN104616658A (en) | Echo canceling implementing method supporting a plurality of voice coding systems | |
| US20090304177A1 (en) | Acoustic Echo Canceller |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CP03 | Change of name, title or address |
Address after: 310052 Binjiang District Changhe Road, Zhejiang, China, No. 466, No. Patentee after: Xinhua three Technology Co., Ltd. Address before: 310053 Hangzhou hi tech Industrial Development Zone, Zhejiang province science and Technology Industrial Park, No. 310 and No. six road, HUAWEI, Hangzhou production base Patentee before: Huasan Communication Technology Co., Ltd. |
|
| CP03 | Change of name, title or address |