WO2013097357A1 - 一种回波抵消器及回波抵消方法 - Google Patents

一种回波抵消器及回波抵消方法 Download PDF

Info

Publication number
WO2013097357A1
WO2013097357A1 PCT/CN2012/072395 CN2012072395W WO2013097357A1 WO 2013097357 A1 WO2013097357 A1 WO 2013097357A1 CN 2012072395 W CN2012072395 W CN 2012072395W WO 2013097357 A1 WO2013097357 A1 WO 2013097357A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal
adaptive filter
echo
end speech
threshold
Prior art date
Application number
PCT/CN2012/072395
Other languages
English (en)
French (fr)
Inventor
薛涛
孙焘
刘冬梅
王进军
张琦
王霞
Original Assignee
中兴通讯股份有限公司
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 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Priority to US14/368,953 priority Critical patent/US9282195B2/en
Priority to ES12863916.8T priority patent/ES2644579T3/es
Priority to EP12863916.8A priority patent/EP2785032B1/en
Publication of WO2013097357A1 publication Critical patent/WO2013097357A1/zh

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04MTELEPHONIC COMMUNICATION
    • H04M9/00Arrangements for interconnection not involving centralised switching
    • H04M9/08Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic
    • H04M9/082Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic using echo cancellers
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10KSOUND-PRODUCING DEVICES; METHODS OR DEVICES FOR PROTECTING AGAINST, OR FOR DAMPING, NOISE OR OTHER ACOUSTIC WAVES IN GENERAL; ACOUSTICS NOT OTHERWISE PROVIDED FOR
    • G10K11/00Methods or devices for transmitting, conducting or directing sound in general; Methods or devices for protecting against, or for damping, noise or other acoustic waves in general
    • G10K11/16Methods or devices for protecting against, or for damping, noise or other acoustic waves in general

Definitions

  • the present invention relates to a method for acoustic echo cancellation in the field of mobile communications, an echo cancellation device and an echo cancellation method.
  • the echo canceller generally uses an adaptive echo cancellation method.
  • the adaptive filter generates the same signal as the echo by recognizing the impulse response of the acoustic feedback channel, and subtracts the echo signal from the mixed signal of the near-end voice and echo to achieve the purpose of echo cancellation. This not only ensures the minimum impact of the voice quality, but also minimizes the echo.
  • the most important feature of adaptive echo cancellation technology is that it does not limit the space where the acoustic feedback channel is located. Regardless of the internal space size, regardless of its internal settings, and regardless of the position of the speaker, it automatically tracks the acoustic characteristics of the room. Change, to minimize the echo and even howling caused by acoustic feedback. Therefore, the rapid automatic identification and tracking of the acoustic feedback characteristics of the driving or indoor LRM system channel is the key to the acoustic echo cancellation.
  • the traditional double-ended detection module generally uses the Geigel method and the correlation detection method.
  • the Geigel method is low in complexity and easy to implement, but the threshold is very difficult to determine and the effect is poor in noisy environments.
  • the correlation detection method mainly relies on the detection of near-end and far-end speech. When the noise is large or the path is abrupt. Causes the performance of the filter to deteriorate.
  • the traditional path detection module uses a master-slave filter structure.
  • the auxiliary filter is generally a 128 ms window, which covers the entire echo path. However, the cost and complexity of this method are large. At the same time, after the path is changed, the echo also changes, so that the double-ended detection method with good performance deteriorates due to the threshold failure performance.
  • the technical problem to be solved by the present invention is to provide an echo canceller and an echo canceling method, which can achieve a good convergence effect and a double-talk effect in the initialization phase.
  • An echo canceller includes: an adaptive filter, a voice signal detecting portion, and a path change detecting portion, and the far-end voice signal propagates through the speaker in an echo path , picked up by the microphone to form an echo signal, where:
  • the adaptive filter is configured to: receive a far-end speech signal as a training signal to simulate an echo path, and perform echo signal cancellation on the near-end signal;
  • the voice signal detecting unit is configured to: detect a call state, control the adaptive filter according to a call state, and control activation of the path change detecting unit according to a call state;
  • the path change detecting unit is configured to: detect whether the echo path changes, and control the adaptive filter according to whether the echo path changes.
  • the initial coefficient of the adaptive filter is 0.
  • the echo canceler further includes a random sequence transmitting unit, wherein:
  • the random sequence transmitting unit is configured to: when establishing a call link, send a random sequence to the speaker, the random sequence forms an echo signal through an echo path and a microphone, and transmits the signal to the adaptive filter,
  • the adaptive filter is initialized;
  • the adaptive filter is configured to: train the echo signals formed by the random sequence, and store the trained coefficients as initial coefficients.
  • the random sequence is not related to the voice signal.
  • the voice signal detecting unit includes a detection threshold initialization unit and a near-end voice judgment list. a meta-distal speech determination unit and a control signal transmission unit, wherein:
  • the detection threshold initialization unit is configured to: store the near-end signal when establishing a call link, and estimate an initial value of the voice detection threshold by using the stored near-end signal as a near-end speech determination unit An initial value of the end speech detection threshold and an initial value of the far end speech detection threshold of the far end speech judging unit;
  • the near-end speech determining unit is configured to: determine whether a near-end speech signal exists, and input a result of the near-end speech determination into the control signal transmitting unit;
  • the far-end speech judging unit is configured to: determine whether there is a far-end speech signal, and input the result of the far-end speech judgment into the control signal sending unit;
  • the control signal transmitting unit is configured to: after receiving the result of the near-end speech judgment and the result of the far-end speech judgment, issue a control signal to control the adaptive filter section and the path change detecting section.
  • the control signal transmitting unit is configured to issue a control signal to control the adaptive filter unit and the path change detecting unit in the following manner:
  • the adaptive filter is locked, ie the adaptive filter coefficients are not updated:
  • the path change detecting unit is activated.
  • the detection threshold initialization unit is configured to estimate an initial value of the voice detection threshold in the following manner:
  • the near-end speech determining unit is configured to determine whether there is a near-end speech signal in the following manner:
  • dl (n) is a convolution of a first L-order coefficient of the adaptive filter and a far-end speech signal
  • Threshold is a first near-end speech detection threshold
  • Threshold2 is the second near-end speech detection threshold, which is the length of the sliding window
  • M M + S
  • M is the starting point of the sliding window
  • S is the sliding window sliding length
  • the counter is 2 (/) 2 ⁇ 73 ⁇ 4 ⁇ 0 C(n)
  • d(n) is the near-end signal
  • is the frame length
  • Threshold is the first near-end speech detection threshold
  • Threshold2 is the second near-end speech detection threshold
  • M M + S
  • M M + S
  • S is the sliding window sliding length
  • the far-end speech judging unit is configured to determine whether there is a far-end speech signal according to the following manner:
  • Threshhold1 is a first far-end speech detection threshold
  • Threshold3 is a second far-end speech detection threshold
  • is a sliding window length
  • the first near-end speech detection threshold is equal to the first far-end speech detection threshold
  • the second near-end speech detection threshold is equal to the second far-end speech detection threshold.
  • the voice signal detecting unit further includes a correlation calculating unit and a threshold change control unit, where:
  • the correlation calculation unit is configured to: receive a path change detection result sent by the path change detection unit, and calculate a correlation between the near end signal and the far end speech signal when the echo path changes, when the correlation is greater than the correlation
  • receive a path change detection result sent by the path change detection unit and calculate a correlation between the near end signal and the far end speech signal when the echo path changes, when the correlation is greater than the correlation
  • comparing the thresholds determining that there is no near-end speech signal in the near-end signal, and checking a judgment result of the near-end speech judging unit, determining that the first near-end speech detection is performed when the judging result is that there is a near-end speech signal Threshold failure
  • the threshold change control unit is configured to: when the first near-end speech detection threshold fails, recalculate the detection threshold of the near-end speech, and send a control signal to the near-end speech judging unit to change the detection threshold of the near-end speech.
  • the threshold change control unit is configured to recalculate the detection threshold of the near-end speech according to the following manner: determining whether there is r C( ) > Threshold!
  • Threshold ⁇ ( ⁇ 2 ) is determined, where Threshold is the first near-end speech detection threshold, Thresholds is the correlation comparison threshold, is the signal variance, and 7, is the speech threshold estimation gain.
  • the path change detecting unit includes: a second adaptive filter, an adaptive filter performance calculating unit, a second adaptive filter performance calculating unit, and a performance comparing unit, where:
  • the input signal of the second adaptive filter is a far-end speech signal
  • the expected signal is a value obtained by convolving the first L-order coefficient of the adaptive filter and the far-end speech signal
  • the error signal is the convolution
  • the error signal is set to: the first L coefficients of the analog echo path;
  • the adaptive filter performance calculation unit is configured to: calculate performance of the adaptive filter, and send the result to the performance comparison unit;
  • the second adaptive filter performance calculation unit is configured to: calculate performance of the second adaptive filter, and send the result to the performance comparison unit;
  • the performance comparison unit is configured to: after receiving the results sent by the adaptive filter performance calculation unit and the second adaptive filter performance calculation unit, determine whether a path mutation has occurred, and restart the determining when determining a path mutation
  • the adaptive filter simultaneously turns on the correlation calculation unit of the speech signal detecting unit; when it is determined that the path mutation has not occurred, the adaptive filter is normally updated.
  • ERLE2 is the adaptive filter performance value
  • d() is the near-end signal
  • e(n) is the error signal of the near-end signal
  • Dl dl(n) calculates the performance of the second adaptive filter, where ERLE1 is the second adaptive filter performance value, dl(n) is the first L-order coefficient and the far of the second adaptive filter Convolution of the end speech signal, el(n) is the convolutional error signal;
  • the performance comparison unit is configured to determine whether a path mutation has occurred in the following manner: receiving the adaptive filter performance calculation unit and the The second adaptive filter performance calculation unit calculates the performance value, and determines whether ERLE1-ERLE2> 73 ⁇ 4r ⁇ 1 ⁇ 2 4 is present, wherein Threshold4 is a path mutation threshold, and if present, determines an echo path mutation.
  • An echo 4 method including:
  • the far-end speech signal propagates through the speaker in the echo path and is picked up by the microphone to form an echo signal;
  • the adaptive filter receives the far-end speech signal as a training signal to simulate an echo path, and performs echo signal cancellation on the near-end signal;
  • the voice signal detecting unit detects the call state, controls the adaptive filter according to the call state, and controls the path to change the start of the detecting unit according to the call state;
  • the path change detecting unit detects whether or not the echo path changes, and controls the adaptive filter according to whether or not the echo path changes.
  • the echo cancellation method further includes: setting an initial coefficient of the adaptive filter to zero.
  • the echo cancellation method further includes:
  • the random sequence transmitting unit sends a random sequence to the speaker when the call link is established, and the random sequence forms an echo signal through the echo path and the microphone, and transmits the signal to the adaptive filter, and the adaptive filtering is performed.
  • the adaptive filter ⁇ trains with the echo signal formed by the random sequence, and stores the trained coefficients as initial coefficients.
  • the step of detecting the call state by the voice signal detecting unit, controlling the adaptive filter according to the call state, and controlling the start of the path change detecting unit according to the call state includes:
  • the voice signal detecting unit stores the near-end signal when establishing a call link, and estimates an initial value of the voice detection threshold by using the stored near-end signal as an initial value of the near-end voice detection threshold and a far-end voice.
  • the initial value of the detection threshold
  • the speech signal detecting unit determines whether there is a near-end speech signal, and determines whether there is a far-end speech signal, and issues a control signal to control the adaptive filter unit and the path change detecting unit.
  • the step of the voice signal detecting unit emitting a control signal to control the adaptive filter unit and the path change detecting unit includes:
  • the adaptive filter is locked, ie the adaptive filter coefficients are not updated:
  • the path change detecting unit is activated.
  • the voice signal detecting unit estimates the voice detection threshold by using the stored near-end signal.
  • dl(n) is a convolution of the first L-order coefficient of the adaptive filter and the far-end speech signal
  • Threshold is the first near-end speech detection threshold
  • Threshold2 is The second near-end speech detection threshold is the length of the sliding window
  • M M + S
  • M is the starting point of the sliding window.
  • Signal its
  • d(n) is the near-end signal
  • is the frame length
  • Threshold is the first near-end speech detection threshold
  • Threshold2 is the second near-end speech detection threshold
  • M M + S
  • M is the sliding The starting point of the window
  • S is the sliding length of the sliding window
  • the step of the voice signal detecting unit determining whether there is a far-end voice signal includes:
  • Threshhold1 is the first far-end voice detection threshold
  • Threshold3 is the second far-end voice detection threshold
  • is the sliding window length. among them:
  • the first near-end speech detection threshold is equal to the first far-end speech detection threshold
  • the second near-end speech detection threshold is equal to the second far-end speech detection threshold
  • the voice signal detecting unit receives the path change detection result sent by the path change detecting unit, and calculates a correlation between the near-end signal and the far-end voice signal when the echo path changes, when the correlation is greater than the correlation comparison threshold. Determining that there is no near-end speech signal in the near-end signal, and checking a judgment result of the near-end speech judging unit, when the judging result is that there is a near-end speech signal, The first near-end voice detection threshold is invalid;
  • the voice signal detecting unit recalculates the detection threshold of the near-end voice when the first near-end voice detection threshold fails, and issues a control signal to change the detection threshold of the near-end voice.
  • Threshold ⁇ ( ⁇ 2 ) is determined, where Threshold is the first near-end speech detection threshold, Thresholds is the correlation comparison threshold, is the signal variance, and 7, is the speech threshold estimation gain.
  • the echo cancellation method further includes:
  • the step of adapting the performance of the filter comprises: calculating the performance of the second adaptive filter, wherein ERLE1 is
  • dl(n) is a convolution of a front L-th order coefficient of the second adaptive filter and a far-end speech signal
  • el(n) is the convolved error signal
  • the step of determining whether a path mutation has occurred includes: determining whether ERLE1 - ERLE2 > Threshold ⁇ exists, wherein Threshold4 is a path mutation threshold, and if present, determining an echo path mutation.
  • the present invention can achieve a good reception in the initialization phase after using the above technical solution.
  • the convergence effect has certain stability in double-talking; the random sequence has the function of prompting sound, which has good practicability; reduces the mutual influence between double-talk detection and path mutation detection; improves echo suppression The echo cancellation performance of the double talk and path mutation.
  • Embodiment 1 is a configuration diagram of an echo canceler according to Embodiment 1 of the present invention.
  • FIG. 2 is a configuration diagram of a speech signal detecting portion in the echo canceller according to Embodiment 1 of the present invention
  • FIG. 3 is a second adaptive filtering in the path change detecting portion in the echo canceller according to Embodiment 1 of the present invention
  • FIG. 4 is a configuration diagram of a path change detecting unit in the echo canceller according to Embodiment 1 of the present invention
  • FIG. 5 is a flowchart of an echo canceller according to Embodiment 1 of the present invention
  • Fig. 6 is a configuration diagram of an echo canceller according to a second embodiment of the present invention.
  • the conventional echo canceller does not consider the initialization problem of the adaptive filter, so that the performance of the adaptive filter is not good in the initial stage of echo suppression.
  • the short-term muting resource is established by using the communication link, the random sequence is used to initialize the adaptive filter, and the ambient noise is extracted to estimate the initial threshold of the voice detection.
  • the initial coefficients of the adaptive filter are not set to zero vector but are replaced by echo path analog values, so that the adaptive filter in the initialization phase converges faster and the steady-state error is smaller.
  • Sending a random sequence that is uncorrelated with the speech signal makes the adaptive filter itself robust to double talk. At the same time, the random sequence can remind the user that the voice channel has been connected, which has good practicability.
  • the echo cancellation device of the present embodiment includes: an adaptive filter, a random sequence transmitting unit, a voice signal detecting unit, and a path change detecting unit, wherein:
  • the adaptive filter is configured to: cancel the echo signal from the near-end signal;
  • the random sequence transmitting unit is configured to: send a random sequence to initialize the adaptive filter;
  • the voice signal detecting unit is configured to: determine the call state by detecting the near-end voice and the far-end voice, respectively, and control the adaptive filter according to the call state;
  • the path change detecting unit is configured to: compare performance between the adaptive filter and the second adaptive filter, determine whether a path change occurs, thereby controlling the adaptive filter, and simultaneously updating the voice detection threshold in the voice detecting unit.
  • Step 1 While the call link is established, the random sequence transmitting unit sends a random sequence to train the adaptive filter, and the adaptive filter stores the obtained coefficient as the initial coefficient of the adaptive filter;
  • the second step the voice signal detecting unit stores the near-end signal, and estimates the voice detection threshold as the initial threshold according to the stored near-end signal, and performs voice detection;
  • Step 3 The voice signal detecting unit controls the adaptive filter according to different results of the voice detection, including: turning off the adaptive filter when the near-end call or muting, and locking the adaptive filter when the double-talk is in use;
  • the path change detecting unit detects whether a path mutation occurs, and when the path is abrupt, the adaptive filter is restarted, and the near-end speech detection threshold is adjusted by the speech signal detecting unit; when no path mutation occurs, the adaptive filter is normally updated.
  • the speech detection method based on the speech energy estimation uses the desired signal of the second adaptive filter to perform double-talk detection.
  • the second adaptive filter is easier to distinguish between echo and near-end speech because of its smaller order, so it can obtain more accurate detection results than adaptive filters. It is judged as a double-talking when the far-end call is at the same time as the near-end call.
  • the present embodiment adds a sliding window after determining whether there is a near-end speech at the single point. When the near-end speech point in the window exceeds the threshold, it is judged that there is near-end speech, which improves the accuracy of speech detection.
  • the speech signal detecting unit can set the initial threshold of the speech detection based on the preliminary estimation of the noise.
  • the path change detecting unit judges the performance difference between the second adaptive filter and the adaptive filter Whether a path change occurs, wherein the desired signal of the second adaptive filter is derived from the adaptive filter coefficients and the far-end speech signal. Since the second adaptive filter simulates the first few coefficients of the echo path, a good convergence effect can be achieved, and the path change can be relatively clearly determined when the path is changed.
  • the near-end speech detection threshold is automatically adjusted after the path is changed: when the correlation between the received signal and the far-end speech signal is greater than the threshold, that is, there is no near correlation detection.
  • the voice is terminated, and the threshold update can be detected when the near-end voice is detected according to the original threshold.
  • a new threshold is obtained by estimating the energy of the near-end signal and the set gain until the two test results are consistent.
  • FIG. 1 is a configuration diagram of an echo canceller of the present embodiment, and the echo canceller includes a speaker 101, a microphone 102, an adaptive filter 103, a voice signal detecting section 104, and a path change detecting section 105.
  • x(n) is the far-end speech signal
  • d(n) is the received signal (near-end signal)
  • e(n) is the error signal
  • n is the time.
  • the received signal contains one or more of a near-end speech signal, an echo signal, and ambient noise.
  • the far end speech signal x(n) propagates through the speaker 101 in the echo path 100 and is picked up by the microphone 102 to form an echo signal.
  • the adaptive filter section 103 receives the far-end speech signal x(n) as a training signal to simulate the echo path 100, and cancels the echo signal in the received signal.
  • the variable step size normalized Least Mean Square (VSS-NLMS) algorithm is usually used, and the order is generally selected from 512 to 1024.
  • a near-end speech signal (double talk) may occur in the echo path. Since the near-end speech signal adversely affects the adaptive filter 103, the speech signal detecting unit 104 detects whether the near-end signal contains a near-end speech signal. In this way, the next step of the adaptive filter 103 is controlled.
  • the echo path 100 may change as the user moves, and the path change detecting unit 105 detects whether the echo path 100 changes.
  • an adaptive filter is required.
  • the 103 can be changed quickly, and the path change detecting section 105 is also used to control the operation of the adaptive filter 103. Since there is an interaction between the double talk and the path change, the path change detecting section 105 issues a control signal 107 to control the threshold update of the voice signal detecting section 104 after the path change.
  • Fig. 2 is a structural diagram of the speech signal detecting section 104 in the echo canceller of the present embodiment, assuming that the first L coefficients of the echo path 100 are simulated by the second adaptive filter.
  • the order of the second adaptive filter is L (generally 32-128), and dl(n) is obtained by convolving the first L-order coefficient of the adaptive filter 103 and the far-end speech signal x(n), that is, the second The desired signal of the adaptive filter.
  • the detection threshold initialization unit 201 stores the near-end signal while the call link is established, and estimates the speech detection threshold according to Equation 1 and Equation 2, as the initial value of the near-end speech detection threshold of the near-end speech judging unit 202 and the far-end speech judgment.
  • the initial value of the far end speech detection threshold of unit 203 is the initial value of the far end speech detection threshold of unit 203.
  • ThresholdO K . E(v 2 ) [Formula 2] where ThmsholdQ is the initial value of the speech detection threshold, is the near-end signal stored by the detection threshold initialization unit 201, N is the signal length, generally 1000 3000, K is the speech threshold Estimate the gain, generally take 3 ⁇ 5.
  • the near-end speech judging unit 202 is for judging whether or not there is a near-end speech signal, and performs preliminary judgment according to Equation 3. Due to the unbalanced speech sound level, the near-end speech is misjudged at various points. According to the continuous appearance of the speech point, after the single point is judged whether there is near-end speech, the sliding window is added, and the near-end speech point in the window exceeds When the threshold value is determined, there is a near-end speech, that is, the final judgment is performed using Equation 4.
  • the far-end signal judging unit 203 is the same as the above-described working method, and only performs initial judgment according to Equation 5, and uses Equation 6 to perform final judgment.
  • Threshold is the first near-end voice detection threshold
  • Threshold2 is the second near-end voice detection threshold
  • Threshhold1 is the first far-end voice detection threshold
  • Threshold3 is the second far-end voice detection threshold.
  • is the length of the sliding window, generally 100 300
  • M M + S
  • M is the starting point of the sliding window
  • S is the sliding window sliding length, generally 50 ⁇ 150.
  • the result of the voice judgment is input to the control signal transmitting unit 204, and the control signal 108 is issued to control the adaptive filter section 103 and the path change detecting section 105.
  • the adaptive filter is turned off at this time;
  • the path change detecting section 105 performs the path change detection.
  • the near-end speech judging unit 202 still determines that there is near-end speech, so it is necessary to determine whether the path is changed after the path is changed. Adjust the threshold of voice detection.
  • Threshold G, ⁇ ⁇ 2 ) [Equation 7] where Threshold5 is the correlation comparison threshold, generally set to 0.2 ⁇ 0.4; is the signal variance,
  • G x is the speech threshold estimation gain, which is generally 3-5.
  • Fig. 3 is a structural diagram of a second adaptive filter
  • Fig. 4 is a structural diagram of a path change detecting section 105.
  • the input signal of the second adaptive filter 301 is the far-end speech signal x(n), the desired signal is dl(n), and the error signal is el(n).
  • the second adaptive filter 301 simulates the first L coefficients of the echo path 100. Since the order is smaller than the adaptive filter, the convergence effect is good. When the path is changed, the path change can be tracked more quickly, and the path change is judged by comparing the performance of the adaptive filter.
  • the performance of the two filters is calculated according to Equation 8, and the result is input to the performance comparison unit 40 to determine whether or not a path mutation has occurred.
  • d represent the desired signal and the error signal of the adaptive filter, respectively.
  • ERLE1 is the second adaptive filter performance value
  • ERLE2 is the adaptive filter performance value
  • Threshold4 is the path mutation threshold, which is generally set to 15 ⁇ 20.
  • control signal 107 is given to control the adaptive filter section 103 and the speech signal detecting section 104.
  • the adaptive filter 103 is restarted, and the correlation calculating unit 205 of the speech signal detecting section 104 is turned on.
  • FIG. 5 is the back of the embodiment 1 of the present embodiment.
  • the flow of the Wave 4 method including:
  • Step 501 The detection threshold initialization unit 201 of the voice signal detecting unit 104 stores the near-end signal, and the initialization threshold of the voice detection is obtained after the communication link is established.
  • Step 502 The voice signal detecting unit 104 detects a call state.
  • Step 503 The voice signal detecting unit 104 determines whether the call state is a near-end call or muting, if yes, step 505 is performed; otherwise, step 504 is performed;
  • Step 504 Turn on the adaptive filter 103, and perform step 506;
  • Step 505 Turn off the adaptive filter 103, return to step 502;
  • Step 506 Determine whether the call is a double-talk, if yes, go to step 507; otherwise, go to step 508;
  • Step 507 Stop updating the adaptive filter 103, returning to step 502;
  • Step 508 Start filter performance comparison unit 403;
  • Step 509 The performance comparison unit 403 determines whether a path mutation occurs. If a path mutation occurs, step 510 is performed; otherwise, step 511 is performed;
  • Step 510 Restart the adaptive filter 103, turn on the correlation calculation unit 205, update the voice detection threshold, and then return to step 502;
  • Step 511 The adaptive filter is normally updated according to the variable step size algorithm, and then returns to step 502 to perform voice signal detection again.
  • the detection threshold initializing unit 201 stores the near-end signal while the call link is established, and estimates the speech detection threshold according to Equation 1 and Equation 2, as the near-end speech judging unit 202 and the far-end speech judging unit 203.
  • the initial threshold In the present embodiment, while the call link is established, not only the detection threshold initialization unit 201 stores the near-end signal, but also the random sequence transmission unit 601 actively transmits a random sequence to initialize the adaptive filter 103.
  • Fig. 6 is a structural diagram of the echo canceler of this embodiment, and the random sequence is usually selected as an M sequence.
  • the random sequence transmitted by the random sequence transmitting unit 601 is transmitted to the microphone 102 via the echo path 100, the echo signal is formed to initialize the adaptive filter 103, and the adaptive filter 103 normalizes the minimum mean square error using the variable step size (VSS).
  • VSS variable step size
  • the algorithm trains and stores the coefficients obtained by training as the initial stage of the adaptive filter. Starting coefficient.
  • the initial coefficient of the adaptive filter is not set to a zero vector but is replaced by a preliminary echo path analog value, so that the filter in the initialization phase converges faster and the steady state error is smaller.
  • the present embodiment uses a random sequence that is not related to any speech signal, so that The adaptive filter itself is somewhat robust in double talk.
  • the random sequence sent after the voice communication link is established can remind the user that the voice channel has been connected, which has good practicability.
  • the flow of the echo canceler of this embodiment is the same as that of Embodiment 1 except that the random sequence is actively sent when the link is established, and details are not described herein again.
  • the order of the second adaptive filter is L (generally 32 to: 128), and dl(n) is the first L-order coefficient of the adaptive filter 103 and the convolution of the far-end signal x(n).
  • the near-end speech determining unit 202 of the speech signal detecting unit 104 determines whether there is a near-end speech signal, and performs preliminary judgment according to Equation 3, using Equation 4 Final judgment.
  • the adaptive filter is used to receive the signal d(n) as the input signal of the near-end speech judging unit 202 in the speech signal detecting section 104.
  • the frame structure is used to initially judge the speech signal.
  • > Threshold timing counter C(n) is incremented by 1 [Equation 10] where d(n) is the received signal of the adaptive filter unit 103, and ⁇ is the frame length, which is generally 100 300.
  • Other structures and implementation steps are exactly the same as those in Embodiment 1, and are not described herein again.
  • the present invention can achieve a good convergence effect in the initialization phase after using the above technical solution, and has certain stability in double-talking; the random sequence has a prompt sound effect, and has good practicability; The interaction between double-talk detection and path mutation detection is reduced; the echo cancellation performance of the echo suppressor in double-talk and path mutation is improved. Therefore, the present invention has strong industrial applicability.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
  • Telephone Function (AREA)

Abstract

一种回波抵消器和方法,该回波抵消器包括:自适应滤波器、语音信号探测部和路径改变探测部,远端语音信号通过扬声器在回波路径中传播,被话筒拾取,形成回波信号,其中:所述自适应滤波器设置成:接收远端语音信号作为训练信号模拟回波路径,对近端信号进行回波信号的消除;所述语音信号探测部设置成:检测通话状态,根据通话状态控制所述自适应滤波器,并根据通话状态控制所述路径改变探测部的启动;所述路径改变探测部设置成:检测回波路径是否发生改变,根据回波路径是否发生改变控制所述自适应滤波器。采用上述技术方案后,能够在初始化阶段达到很好的收敛效果和双端通话效果。

Description

一种回波抵消器及回波抵消方法
技术领域
本发明涉及移动通信领域中声学回波消除的方法, 一种回波 ·ί氏消器及回 波抵消方法。
背景技术
随着无线通信技术的飞速发展, 人们对语音通信的质量及舒适性提出了 更高的要求, 其中, 舒适自然的免提对话环境正成为人们日益增长的需求。 然而回波的存在影响了通信质量, 严重时可能使通信系统无法正常工作。 因 此, 必须釆取有效措施来抑制回波, 消除其影响, 才能提高语音通信质量。
回波抵消器一般釆用自适应回波抵消方法。 自适应滤波器通过对声回授 通道的冲击响应的辨识, 产生出同回声相同的信号, 再从近端话音和回声的 混合信号中减去回声信号, 来达到回声对消的目的。 这样既可以保证对话音 质量影响最小, 又使得回声得到最大的抑制。 自适应回波对消技术的最大特 点就是不限制声回授通道所在的空间, 不论内部空间尺寸如何, 不论其内部 陈设如何,也不论讲话者所处位置,它都会自动地跟踪房间声学特性的变化, 最大限度地抑制由声回授引起的回声乃至嘯叫。 因此,对驾驶抢或室内 LRM 系统通道的声回授特性的快速自动辨识及跟踪是声回波对消的关键。
然而, 在实际通信情况下存在双端通话和路径突变等现象, 这些现象都 会影响自适应滤波器的收敛性能, 导致回波不能有效地得到消除。 回波抵消 器中路径改变检测模块和双端检测模块性能的优劣直接影响了回波抑制的效 果。 在检测到双端通话模式时, 需要控制自适应滤波器停止系数更新, 否则 误差信号的突增将导致自适应滤波器发散; 在路径突变时需要重启滤波器, 以便能快速跟踪路径变化, 更好的消除回波。
传统的双端检测模块一般釆用 Geigel法和相关性检测法。 Geigel法复杂 度低易于实现, 但是门限的确定十分困难, 并且在噪声环境下的效果较差, 而相关检测法主要依靠对近端和远端语音的检测, 当噪声较大或者路径突变 时会造成滤波器的性能恶化。 传统的路径探测模块釆用主从滤波器结构, 加入的辅助滤波器一般为全 长为 128ms的窗, 用来覆盖整个回波路径, 但是, 这种方法的花销与复杂度 都很大。 同时在路径改变之后, 回波也发生改变, 使得原本性能良好的双端 检测方法因门限失效性能恶化。
发明内容
本发明要解决的技术问题是提供一种回波抵消器及回波抵消方法, 能够 在初始化阶段达到艮好的收敛效果和双端通话效果。
为解决上述技术问题, 本发明釆用如下技术方案: 一种回波抵消器, 包括: 自适应滤波器、 语音信号探测部和路径改变探 测部, 远端语音信号通过扬声器在回波路径中传播, 被话筒拾取, 形成回波 信号, 其中:
所述自适应滤波器设置成: 接收远端语音信号作为训练信号模拟回波路 径, 对近端信号进行回波信号的消除;
所述语音信号探测部设置成: 检测通话状态, 根据通话状态控制所述自 适应滤波器, 并根据通话状态控制所述路径改变探测部的启动;
所述路径改变探测部设置成: 检测回波路径是否发生改变, 根据回波路 径是否发生改变控制所述自适应滤波器。
其中, 所述自适应滤波器的初始系数为 0。
所述回波抵消器还包括随机序列发送部, 其中:
所述随机序列发送部设置成: 在建立起通话链路时, 向所述扬声器发送 随机序列, 所述随机序列经回波路径和麦克风形成回波信号, 传送到所述自 适应滤波器, 对所述自适应滤波器进行初始化;
所述自适应滤波器设置成:釆用所述随机序列形成的回波信号进行训练, 将训练得到的系数存储为初始系数。
其中, 所述随机序列与语音信号不相关。 其中, 所述语音信号探测部包括检测门限初始化单元、 近端语音判断单 元、 远端语音判断单元和控制信号发送单元, 其中:
所述检测门限初始化单元设置成: 在建立起通话链路时, 存储所述近端 信号, 釆用所存储的近端信号估计语音检测门限的初始值, 作为所述近端语 音判断单元的近端语音检测门限的初始值和所述远端语音判断单元的远端语 音检测门限的初始值;
所述近端语音判断单元设置成: 判断是否存在近端语音信号, 并将近端 语音判断的结果输入所述控制信号发送单元;
所述远端语音判断单元设置成: 判断是否存在远端语音信号, 并将远端 语音判断的结果输入所述控制信号发送单元;
所述控制信号发送单元设置成: 收到近端语音判断的结果和远端语音判 断的结果后,发出控制信号控制所述自适应滤波器部和所述路径改变探测部。
其中, 所述控制信号发送单元设置成按照以下方式发出控制信号控制所 述自适应滤波器部和所述路径改变探测部:
在近端通话或静音时, 关闭所述自适应滤波器;
双端通话时, 锁定所述自适应滤波器, 即不对所述自适应滤波器系数进 行更新:
非近端通话、 非静音、 且非双端通话时, 启动所述路径改变探测部。 其中, 所述检测门限初始化单元设置成按照以下方式估计语音检测门限 的初始值:
釆用 ThresholdO = K . E(v2 )计算语音检测门限的初始值, 其中, ThresholdO为 语音检测门限的初始值, K为语音门限估计增益, £(ν2) = ^; («)2 , 为近 端信号, N为信号长度。 其中:
所述近端语音判断单元设置成按照以下方式判断是否存在近端语音信号:
Μ+Νγ-\
在 〉7¾ ra 7oW时对计数器 C(n)力 1 , SUMC = C(i > Threshoid2 ^ , i=M
判断所述近端信号中存在近端语音信号, 其中, dl (n)为自适应滤波器的前 L 阶系数和远端语音信号的卷积, Threshold 为第一近端语音检测门限, Threshold2为第二近端语音检测门限, 为滑动窗长度, M = M + S , M为滑 动窗起点, S为滑动窗滑动长度; 或者,在2 (/)2〉7¾^^0 时对计数器 C(n)
i=l
Μ+Νλ-\
力口 1 , SUMC = C(i、> Threshold! ^ , 判断所述近端信号中存在近端语音 i=
信号, 其中, d(n)为近端信号, ^为帧长, Threshold为第一近端语音检测门 限, Threshold2为第二近端语音检测门限, 为滑动窗长度, M = M + S , M 为滑动窗起点, S为滑动窗滑动长度;
所述远端语音判断单元设置成按照以下方式判断是否存在远端语音信号:
M+N2-\
在 x(")2〉7¾m^o l时对计数器 D(n)力 1 , 在^ 7 D = Ό(Γ> > Threshold3 , i=M
判断所述近端信号中存在远端语音信号, 其中, Threshholdl为第一远端语音 检测门限, Threshold3为第二远端语音检测门限, ^为滑动窗长度。 其中:
所述第一近端语音检测门限与所述第一远端语音检测门限相等, 所述第 二近端语音检测门限与所述第二远端语音检测门限相等。 其中,所述语音信号探测部还包括相关性计算单元和门限改变控制单元, 其中:
所述相关性计算单元设置成: 接收所述路径改变探测部发送的路径改变 探测结果,在回波路径发生改变时,计算近端信号与远端语音信号的相关性, 当相关性大于相关性比较门限时, 确定所述近端信号中无近端语音信号, 并 检查所述近端语音判断单元的判断结果, 在所述判断结果为存在近端语音信 号时, 确定第一近端语音检测门限失效;
所述门限改变控制单元设置成: 在第一近端语音检测门限失效时, 重新 计算近端语音的检测门限, 并向所述近端语音判断单元发出控制信号改变近 端语音的检测门限。
其中, 所述门限改变控制单元设置成按照以下方式重新计算近端语音的 检测门限: 判断是否存在 r C( ) > Threshold! ,如果
Figure imgf000006_0001
存在, 则确定 Threshold = · (ί 2) , 其中, Threshold 为第一近端语音检测门 限, Thresholds为相关性比较门限, 为信号方差, 7,为语音门限估计增益。 其中, 所述路径改变探测部包括: 第二自适应滤波器、 自适应滤波器性 能计算单元、 第二自适应滤波器性能计算单元和性能比较单元, 其中:
所述第二自适应滤波器的输入信号为远端语音信号, 期望信号为由所述 自适应滤波器的前 L阶系数和远端语音信号卷积得到的值, 误差信号为所述 卷积的误差信号, 设置成: 模拟回波路径的前 L个系数;
所述自适应滤波器性能计算单元设置成:计算所述自适应滤波器的性能, 并将结果发送给所述性能比较单元;
所述第二自适应滤波器性能计算单元设置成: 计算所述第二自适应滤波 器的性能, 并将结果发送给所述性能比较单元;
所述性能比较单元设置成: 接收所述自适应滤波器性能计算单元和所述 第二自适应滤波器性能计算单元发送的结果后, 判断是否发生了路径突变, 在确定路径突变时重启所述自适应滤波器, 同时开启所述语音信号探测部的 相关性计算单元; 在确定未发生路径突变时, 正常更新所述自适应滤波器。
其中:
所述自适应滤波器性能计算单元设置成按照以下方式计算所述自适应滤 波器性能: 按照皿 £2 = 101ogl。 ^^计算自适应滤波器的性能, 其中,
ERLE2 为自适应滤波器性能值, d( )为近端信号, e(n)为近端信号的误差信 号;
所述第二自适应滤波器性能计算单元设置成按照以下方式计算所述第二 自适应滤波器性能: 按照 = 101ogl。∑ dl(n) 计算所述第二自适应滤波器 的性能, 其中, ERLE1为第二自适应滤波器性能值, dl(n)为所述第二自适应 滤波器的前 L阶系数和远端语音信号的卷积, el(n)为所述卷积的误差信号; 所述性能比较单元设置成按照以下方式判断是否发生了路径突变: 接收 所述自适应滤波器性能计算单元和所述第二自适应滤波器性能计算单元计算 得到的性能值, 判断是否存在 ERLEl-ERLE2> 7¾r^½ 4 , 其中, Threshold4 为 路径突变门限, 如果存在, 则确定回波路径突变。 一种回波 4氏消方法, 包括:
远端语音信号通过扬声器在回波路径中传播, 被话筒拾取, 形成回波信 号; 自适应滤波器接收远端语音信号作为训练信号模拟回波路径, 对近端信 号进行回波信号的消除;
语音信号探测部检测通话状态, 根据通话状态控制所述自适应滤波器, 并根据通话状态控制路径改变探测部的启动;
所述路径改变探测部检测回波路径是否发生改变, 根据回波路径是否发 生改变控制所述自适应滤波器。
该回波抵消方法还包括: 将所述自适应滤波器的初始系数置为 0。
该回波抵消方法还包括:
随机序列发送部在建立起通话链路时, 向所述扬声器发送随机序列, 所 述随机序列经回波路径和麦克风形成回波信号, 传送到所述自适应滤波器, 对所述自适应滤波器进行初始化;
所述自适应滤波器釆用所述随机序列形成的回波信号进行训练, 将训练 得到的系数存储为初始系数。 其中, 语音信号探测部检测通话状态, 根据通话状态控制所述自适应滤 波器, 并根据通话状态控制路径改变探测部的启动的步骤包括:
所述语音信号探测部在建立起通话链路时, 存储所述近端信号, 釆用所 存储的近端信号估计语音检测门限的初始值, 作为近端语音检测门限的初始 值和远端语音检测门限的初始值;
所述语音信号探测部判断是否存在近端语音信号, 且判断是否存在远端 语音信号, 发出控制信号控制所述自适应滤波器部和所述路径改变探测部。
其中, 所述语音信号探测部发出控制信号控制所述自适应滤波器部和所 述路径改变探测部的步骤包括:
在近端通话或静音时, 关闭所述自适应滤波器;
双端通话时, 锁定所述自适应滤波器, 即不对所述自适应滤波器系数进 行更新:
非近端通话、 非静音、 且非双端通话时, 启动所述路径改变探测部。 其中, 所述语音信号探测部釆用所存储的近端信号估计语音检测门限的 初始值的步骤包括:
所述语音信号探测部釆用 ThreshddO = Κ · Ε(ν2 )计算语音检测门限的初始值, 其中, Threshold^为语音检测门限的初始值, K 为语音门限估计增益, Ε(ν2) = ^^ν(η)2 , ")为近端信号, N为信号长度。
其中:
所述语音信号探测部判断是否存在近端语音信号的步骤包括: 在 〉 7¾ ra 7oW时对计数器 C(n)加 1 , SUMC = C(i > Threshoid2 ^ , 判
i=
断所述近端信号中存在近端语音信号, 其中, dl(n)为自适应滤波器的前 L阶 系数和远端语音信号的卷积, Threshold为第一近端语音检测门限, Threshold2 为第二近端语音检测门限, 为滑动窗长度, M = M + S , M为滑动窗起点,
S为滑动窗滑动长度; 或者, 在 J (/)2〉7¾m^oW时对计数器 C(n)加 1 , 在 SUMC = C(i、> Threshom , 判断所述近端信号中存在近端语音信号, 其
i=
中, d(n)为近端信号, ^为帧长, Threshold 为第一近端语音检测门限, Threshold2为第二近端语音检测门限, 为滑动窗长度, M = M + S , M为滑 动窗起点, S为滑动窗滑动长度;
所述语音信号探测部判断是否存在远端语音信号的步骤包括: 在
M+N2-\
: ί)2 > ¾m^2oM时对计数器 D(n)加 1 , SUMD = Ό(ί > Threshold^ , 判 i=M
断所述近端信号中存在远端语音信号, 其中, Threshholdl为第一远端语音检 测门限, Threshold3为第二远端语音检测门限, ^为滑动窗长度。 其中:
所述第一近端语音检测门限与所述第一远端语音检测门限相等, 所述第 二近端语音检测门限与所述第二远端语音检测门限相等。 该回波抵消方法还包括:
所述语音信号探测部接收所述路径改变探测部发送的路径改变探测结果, 在回波路径发生改变时, 计算近端信号与远端语音信号的相关性, 当相关性 大于相关性比较门限时, 确定所述近端信号中无近端语音信号, 并检查所述 近端语音判断单元的判断结果, 在所述判断结果为存在近端语音信号时, 确 定第一近端语音检测门限失效;
所述语音信号探测部在第一近端语音检测门限失效时, 重新计算近端语 音的检测门限, 并发出控制信号改变近端语音的检测门限。
其中, 重新计算 判断是否存在 r = C( ) > Threshold! ,如果
Figure imgf000010_0001
存在, 则确定 Threshold = · (ί 2) , 其中, Threshold 为第一近端语音检测门 限, Thresholds为相关性比较门限, 为信号方差, 7,为语音门限估计增益。 该回波抵消方法还包括:
将远端语音信号作为第二自适应滤波器的输入信号, 由所述自适应滤波 器的前 L阶系数和远端语音信号卷积得到的值作为所述第二自适应滤波器的 期望信号, 将所述卷积的误差信号作为所述第二自适应滤波器的误差信号, 所述第二自适应滤波器模拟回波路径的前 L个系数;
计算所述自适应滤波器的性能和所述第二自适应滤波器的性能, 判断是 否发生了路径突变, 在确定路径突变时重启所述自适应滤波器, 同时启动相 关性计算; 在确定未发生路径突变时, 正常更新所述自适应滤波器。
其中: 计算所述自适应滤波器的性能的步骤包括: 按照 ERLE2 = 101og10d(nf 2 计算自适应滤波器的性能, 其中, ERLE2为自适应滤波器性能值, d(n)为近 端信号, e(n)为近端信号的误差信号;
自 适应滤波器的性能的 步骤 包括: 按照 计算所述第二自适应滤波器的性能, 其中, ERLE1为
Figure imgf000010_0002
第二自适应滤波器性能值, dl(n)为所述第二自适应滤波器的前 L阶系数和远 端语音信号的卷积, el(n)为所述卷积的误差信号;
判断是否发生 了 路径突变的 步骤 包括: 判 断是否存在 ERLE1 - ERLE2 > Threshold^ , 其中, Threshold4为路径突变门限, 如果存在, 则 确定回波路径突变。
综上所述, 本发明釆用上述技术方案后能够在初始化阶段达到很好的收 敛效果, 在双端通话时具有一定的稳定性; 随机序列具有提示音作用, 具有 很好的实用性; 减小了双端通话检测和路径突变检测之间的相互影响; 提高 了回波抑制器在双端通话和路径突变时的回波抵消性能。 附图概述
图 1是本发明的实施方式 1的回波抵消器的结构图;
图 2是本发明的实施方式 1的回波抵消器中的语音信号探测部的结构图; 图 3是本发明的实施方式 1的回波抵消器中的路径改变探测部中第二自 适应滤波器的结构图;
图 4是本发明的实施方式 1的回波抵消器中的路径改变探测部的结构图; 图 5是本发明的实施方式 1的回波抵消器的流程图;
图 6是本发明的实施方式 2的回波抵消器的结构图。
本发明的较佳实施方式
传统的回波抵消器没有考虑自适应滤波器的初始化问题, 使得在回波抑 制的初始阶段自适应滤波器的性能不佳。 本实施方式利用通信链路建立后短 暂的静音资源, 发送随机序列用初始化自适应滤波器, 并且, 提取环境噪声 来估计语音检测的初始门限。 在回波抑制过程中, 自适应滤波器的初始系数 不是置为零向量而是由回波路径模拟值代替, 从而使得初始化阶段自适应滤 波器收敛更快速, 稳态误差更小。 发送与语音信号不相关的随机序列, 使得 自适应滤波器本身在双端通话时就具有一定的鲁棒性。 同时, 随机序列可以 提醒用户话路已经接通, 具有很好的实用性。
本实施方式的回波 ·ί氏消器, 包括: 自适应滤波器、 随机序列发送部、 语 音信号探测部和路径改变探测部, 其中:
所述自适应滤波器设置成: 从近端信号中消除回波信号;
所述随机序列发送部设置成:发送随机序列对自适应滤波器进行初始化; 所述语音信号探测部设置成:通过对近端语音和远端语音分别进行检测 , 判断通话状态, 4艮据通话状态控制自适应滤波器;
所述路径改变探测部设置成: 对自适应滤波器和第二自适应滤波器进行 性能比较, 判断是否发生路径改变, 以此控制自适应滤波器, 同时更新语音 检测部中的语音检测门限。
本实施方式的回波 ^^消方法, 包括:
第一步: 在通话链路建立起来的同时, 由随机序列发送部发送随机序列 对自适应滤波器进行训练, 自适应滤波器存储训练得到的系数作为自适应滤 波器的初始系数;
第二步: 语音信号探测部存储近端信号, 根据存储的近端信号估计语音 检测门限作为初始门限, 进行语音检测;
第三步: 语音信号探测部根据语音检测的不同结果控制自适应滤波器, 包括: 近端通话或静音时, 关闭自适应滤波器; 双端通话时锁定自适应滤波 器; 无以上状态时, 由路径改变探测部探测是否发生路径突变, 发生路径突 变时, 重启自适应滤波器, 同时由语音信号探测部调整近端语音检测门限; 未发生路径突变时, 正常更新自适应滤波器。
在回波 ·ί氏消器进行完三个步骤之后, 跳到第二步循环运行。
本实施方式釆用基于语音能量估计的语音检测方法利用第二自适应滤波 器的期望信号进行双端通话检测。 第二自适应滤波器由于阶数较小, 更易于 区分回声和近端语音, 所以能得到比自适应滤波器更精确的检测结果。 当远 端通话同时近端通话时判为双端通话。 然而在检测近端语音时由于响度的不 均衡导致近端语音在各个点上的误判, 本实施方式根据语音点连续出现的特 点, 在判断完单点的是否有近端语音后加入滑动窗, 在窗口内的近端语音点 超过门限值时判为存在近端语音, 提高了语音检测的准确性。
由于人在噪声环境中通话时会根据噪声的情况调节说话响度, 所以语音 信号探测部能根据噪声的初步估计设定语音检测的初始门限。
路径改变探测部利用第二自适应滤波器与自适应滤波器的性能差异判断 是否发生路径改变, 其中第二自适应滤波器的期望信号由自适应滤波器系数 和远端语音信号得到。 由于第二自适应滤波器模拟的是回波路径的前几阶系 数, 所以能够达到很好的收敛效果, 当路径改变后能较清晰地判断出路径的 改变。
当路径改变以后, 回波的能量也发生改变, 本实施方式在路径改变后自 动调整近端语音检测门限: 当接收信号和远端语音信号的相关性大于门限, 即相关性检测时不存在近端语音, 同时按照原门限可检测到出现近端语音时 进行门限更新。 借助对近端信号能量的估计和设定的增益得出新的门限, 直 至两种检测结果相一致时停止。 本实施方式提高了语音检测的准确性, 减少 了路径检测和双端检测的相互影响, 增强了回波抵消器的鲁棒性。
下面结合附图对本实施方式进行详细说明。
实施例 1 :
图 1是本实施例的回波抵消器的结构图,回波抵消器包括:扬声器 101、 话筒 102、 自适应滤波器 103、 语音信号探测部 104和路径改变探测部 105。 在图 1中, x(n)是远端语音信号, d(n)是接收信号 (近端信号) , e(n)是误差 信号, n表示时刻。 接收信号中包含近端语音信号、 回波信号和环境噪音的 一种或多种。
远端语音信号 x(n)通过扬声器 101在回波路径 100中传播, 被话筒 102 拾取, 形成回波信号。 自适应滤波器部 103接收远端语音信号 x(n)作为训练 信号模拟回波路径 100 , 消除接收信号中的回波信号。 在自适应滤波器 103 中, 通常情况下使用变步长归一化最小均方误差( VSS-NLMS , Variable Step Size Normalized Least Mean Square ) 算法即可, 阶数一般选为 512~1024。
回波路径中可能出现近端语音信号 (双端通话) , 由于近端语音信号会 对自适应滤波器 103产生不良影响, 所以由语音信号探测部 104检测近端信 号中是否含有近端语音信号, 以此控制自适应滤波器 103下一步工作。
同时, 回波路径 100可能会随着用户的移动而发生改变, 路径改变探测 部 105探测回波路径 100是否发生改变, 路径一旦改变, 需要自适应滤波器 103能快速地随之改变, 路径改变探测部 105也用来控制自适应滤波器 103 工作。 由于双端通话和路径改变之间存在相互影响, 路径改变之后由路径改 变探测部 105发出控制信号 107控制语音信号探测部 104的门限更新。
图 2是本实施例的回波抵消器中的语音信号探测部 104的结构图, 假设 釆用第二自适应滤波器模拟回波路径 100的前 L个系数。 第二自适应滤波器 的阶数为 L (一般取 32-128 ) , dl (n)是自适应滤波器 103的前 L阶系数和远 端语音信号 x(n)卷积得到 , 即第二自适应滤波器的期望信号。
检测门限初始化单元 201在通话链路建立起来的同时存储近端信号, 按 照公式 1和公式 2估计语音检测门限, 作为近端语音判断单元 202的近端语 音检测门限的初始值和远端语音判断单元 203的远端语音检测门限的初始值。
1 N
E(v2 ) = -∑v(n)2 [公式 1]
丄 V i=l 语音检测门限初始化估计:
ThresholdO = K . E(v2 ) [公式 2] 其中, ThmsholdQ为语音检测门限的初始值, 是检测门限初始化单元 201存储的近端信号, N是信号长度, 一般取 1000 3000 , K是语音门限估计 增益, 一般取 3~5。
近端语音判断单元 202用来判断是否存在近端语音信号, 按照公式 3进 行初步判断。 由于语音响度的不均衡导致近端语音在各个点上的误判, 根据 语音点连续出现的特点, 在判断完单点是否有近端语音后加入滑动窗, 在窗 口内的近端语音点超过门限值时判为存在近端语音, 即利用公式 4进行最终 判断。 远端信号判断单元 203与上述工作方法相同, 只是按照公式 5进行初 步判断, 利用公式 6进行最终判断。
近端语音判断:
当 c l(«)2 > Threshold时令计数器 C(n)加 1 [公式 3]
Μ+Νλ-\
SUMC = ^ C(i) > Threshold! [公式 4]
i=
远端语音判断: 当 x(w)2 > Threshold^时令计数器 D(n)加 1 [公式 5]
M+N2-\
SUM = D(/) > Threshold^ [公式 6]
i=
其中, Threshold为第一近端语音检测门限, Threshold2为第二近端语音 检测门限, Threshholdl为第一远端语音检测门限, Threshold3为第二远端语 音检测门限, 在初始检测阶段 Threshold=Thresholdl=ThresholdO; ,^为滑 动窗长度, 一般取 100 300 , M = M + S , M为滑动窗起点, S为滑动窗滑动 长度, 一般取 50~150。 近端语音检测门限和远端语音检测门限一般情况下取 相等的值, 即 Threshold=Thresholdl , Threshhold2=Threshold3。
将语音判断的结果输入控制信号发送单元 204 , 发出控制信号 108控制 自适应滤波器部 103和路径改变探测部 105。
近端通话或静音时(扬声器不发声) 由于不需要进行回波的消除, 此时 关闭自适应滤波器;
双端通话时会使自适应滤波器发散, 此时锁定自适应滤波器; 即滤波器 系数不进行更新: w{n + l) = w(n) ;
无以上状态时(排除近端通话、 静音和双端通话的远端通话情况, 此时 只有远端语音, 可能附加有噪声) , 由路径改变探测部 105进行路径改变探 测。
由于回波路径改变后可能会导致回波信号的突增, 致使无近端语音信号 只有回波信号时, 近端语音判断单元 202仍判断为有近端语音, 因此在路径 改变之后需要判断是否对语音检测的门限进行调整。
路径改变探测的结果由控制信号 107反馈回来,控制相关性计算单元 205。 如果探测到有路径改变, 就启动相关性计算单元 205。 相关性计算单元 205 计算接收信号 d(n)和远端语音信号 x(n)的相关性, 当相关性大于设定的相关 性比较门限时说明此时无近端语音信号, 如果近端语音判断单元 202的检测 结果 207是有近端语音信号时, 说明该门限已经失效, 此时由门限改变控制 单元 206发出控制信号 208改变近端语音的检测门限。 门限的更新由公式 7 给出。 当 r = > Threshold!时
Figure imgf000016_0001
Threshold = G, · Ε{ά2) [公式 7] 其中, Threshold5是相关性比较门限, 一般设为 0.2~0.4; 是信号方差,
Gx是语音门限估计增益, 一般取 3-5。
路径改变探测部 105的具体实现由图 3和图 4给出。 图 3为第二自适应 滤波器的结构图, 图 4为路径改变探测部 105的结构图。
如图 3所示, 第二自适应滤波器 301的输入信号为远端语音信号 x(n), 期望信号为 dl(n), 误差信号为 el(n)。 第二自适应滤波器 301模拟的是回波 路径 100的前 L个系数。 由于阶数比自适应滤波器小, 收敛效果好, 当路径 改变后能更快速的跟踪路径的改变, 通过对比自适应滤波器的性能判断是否 发生了路径改变。
如图 4所示, 按照公式 8分别计算两个滤波器的性能, 将结果输入性能 比较单元 40 得出是否发生了路径突变。 [公式 8]
Figure imgf000016_0002
其中, d,e分别代表所求自适应滤波器的期望信号与误差信号。
ERLE1 - ERLE2 > Threshold^ [公式 9]
ERLE1 是第二自适应滤波器性能值, ERLE2 是自适应滤波器性能值, Threshold4是路径突变门限, 一般设为 15~20。
最终给出控制信号 107控制自适应滤波器部 103和语音信号探测部 104。 路径突变时重启自适应滤波器 103 , 同时开启语音信号探测部 104的相 关性计算单元 205。
未发生路径突变时, 使用变步长归一化最小均方误差 (VSS-NLMS , Variable Step Size Normalized Least Mean Square )算法正常更新自适应滤波器 c 图 5是本实施方式的实施例 1的回波 4氏消方法的流程, 包括:
步骤 501: 在通信链路建立的同时, 由语音信号探测部 104的检测门限 初始化单元 201储存近端信号, 经过计算得到语音检测的初始化门限; 步骤 502: 语音信号探测部 104检测通话状态;
步骤 503:语音信号探测部 104判断通话状态是否为近端通话或者静音, 如果是, 则执行步骤 505; 否则, 执行步骤 504;
步骤 504: 开启自适应滤波器 103 , 执行步骤 506;
步骤 505: 关闭自适应滤波器 103 , 返回步骤 502;
步骤 506: 判决是否为双端通话, 如果是, 则执行步骤 507; 否则, 执行 步骤 508;
步骤 507: 停止更新自适应滤波器 103 , 返回步骤 502;
步骤 508: 启动滤波器性能比较单元 403 ;
步骤 509:性能比较单元 403判决是否发生路径突变,若发生路径突变, 则执行步骤 510; 否则, 执行步骤 511 ;
步骤 510: 重启自适应滤波器 103 , 开启相关性计算单元 205 , 对语音检 测门限进行更新, 之后回到步骤 502;
步骤 511 : 按照变步长算法正常更新自适应滤波器, 之后回到步骤 502 重新进行语音信号检测。
实施例 2:
在上述实施例 1中, 检测门限初始化单元 201在通话链路建立起来的同 时存储近端信号, 按照公式 1和公式 2估计语音检测门限, 作为近端语音判 断单元 202和远端语音判断单元 203的初始门限。 在本实施例中, 在通话链 路建立起来的同时, 不仅检测门限初始化单元 201存储近端信号, 而且随机 序列发送部 601主动发送随机序列, 对自适应滤波器 103进行初始化。 图 6 是本实施例的回波抵消器的结构图, 通常随机序列选为 M序列。
随机序列发送部 601发送的随机序列经过回波路径 100传到麦克风 102 , 形成回波信号对自适应滤波器 103进行初始化, 自适应滤波器 103使用变步 长归一化最小均方误差 (VSS-NLMS , Variable Step Size Normalized Least Mean Square )算法进行训练, 将训练得到的系数存储作为自适应滤波器的初 始系数。
这样在回波抑制过程中, 自适应滤波器的初始系数不是置为零向量而是 由初步的回波路径模拟值代替, 从而使得初始化阶段滤波器收敛更快速, 稳 态误差更小。
由于在远端语音信号和近端语音信号不相关时, 双端通话对自适应滤波 器的系数更新几乎无影响, 因此, 本实施例釆用与任何语音信号均不相关的 随机序列, 所以使得自适应滤波器本身在双端通话时就具有一定的鲁棒性。 同时, 在语音通信链路建立后发送的随机序列可以提醒用户话路已经接通, 具有很好的实用性。
本实施例的回波抵消器的流程,除在链路建立时主动发送随机序列以外, 其他步骤与实施例 1完全相同, 在此不再赘述。
实施例 3 :
实施例 1中假设第二自适应滤波器的阶数为 L (一般取 32〜: 128 ) , dl(n) 是自适应滤波器 103的前 L阶系数和远端信号 x(n)卷积得到, 即第二自适应 滤波器的期望信号, 以此信号作为输入由语音信号探测部 104的近端语音判 断单元 202判断是否存在近端语音信号, 按照公式 3进行初步判断, 利用公 式 4进行最终判断。 在本实施例中, 釆用自适应滤波器接收信号 d(n)作为语 音信号探测部 104中近端语音判断单元 202的输入信号。 同时, 由于语音信 号具有短时平 按照公式 10 , 釆用帧结构对语音信号进行初始判断。 当
Figure imgf000018_0001
> Threshold时令计数器 C(n)加 1 [公式 10] 其中, d(n)为自适应滤波器部 103的接收信号, ^为帧长,一般取 100 300。 其他结构及实施步骤与实施方式 1完全相同, 这里不再赘述。
本领域普通技术人员可以理解上述方法中的全部或部分步骤可通过程序 来指令相关硬件完成, 所述程序可以存储于计算机可读存储介质中, 如只读 存储器、 磁盘或光盘等。 可选地, 上述实施例的全部或部分步骤也可以使用 一个或多个集成电路来实现, 相应地, 上述实施例中的各装置 /模块 /单元可 以釆用硬件的形式实现, 也可以釆用软件功能模块的形式实现。 本发明不限 制于任何特定形式的硬件和软件的结合。
以上所述仅为本发明的优选实施例而已, 并不用于限制本发明, 对于本 领域的技术人员来说, 本发明可以有各种更改和变化。 凡在本发明的精神和 原则之内, 所作的任何修改、 等同替换、 改进等, 均应包含在本发明的保护 范围之内。
工业实用性
综上所述, 本发明釆用上述技术方案后能够在初始化阶段达到很好的收 敛效果, 在双端通话时具有一定的稳定性; 随机序列具有提示音作用, 具有 很好的实用性; 减小了双端通话检测和路径突变检测之间的相互影响; 提高 了回波抑制器在双端通话和路径突变时的回波抵消性能。 因此本发明具有很 强的工业实用性。

Claims

权 利 要 求 书
1、 一种回波抵消器, 包括: 自适应滤波器、 语音信号探测部和路径改变 探测部, 远端语音信号通过扬声器在回波路径中传播, 被话筒拾取, 形成回 波信号, 其中:
所述自适应滤波器设置成: 接收远端语音信号作为训练信号模拟回波路 径, 对近端信号进行回波信号的消除;
所述语音信号探测部设置成: 检测通话状态, 根据通话状态控制所述自 适应滤波器, 并根据通话状态控制所述路径改变探测部的启动;
所述路径改变探测部设置成: 检测回波路径是否发生改变, 根据回波路 径是否发生改变控制所述自适应滤波器。
2、如权利要求 1所述的回波抵消器, 其中, 所述自适应滤波器的初始系 数为 0。
3、如权利要求 1所述的回波抵消器,所述回波抵消器还包括随机序列发 送部, 其中: 所述随机序列发送部设置成: 在建立起通话链路时, 向所述扬声器发送 随机序列, 所述随机序列经回波路径和麦克风形成回波信号, 传送到所述自 适应滤波器, 对所述自适应滤波器进行初始化;
所述自适应滤波器设置成:釆用所述随机序列形成的回波信号进行训练, 将训练得到的系数存储为初始系数。
4、 如权利要求 3所述的方法, 其中, 所述随机序列与语音信号不相关。
5、 如权利要求 2、 3或 4所述的回波抵消器, 其中, 所述语音信号探测 部包括检测门限初始化单元、 近端语音判断单元、 远端语音判断单元和控制 信号发送单元, 其中:
所述检测门限初始化单元设置成: 在建立起通话链路时, 存储所述近端 信号, 釆用所存储的近端信号估计语音检测门限的初始值, 作为所述近端语 音判断单元的近端语音检测门限的初始值和所述远端语音判断单元的远端语 音检测门限的初始值; 所述近端语音判断单元设置成: 判断是否存在近端语音信号, 并将近端 语音判断的结果输入所述控制信号发送单元;
所述远端语音判断单元设置成: 判断是否存在远端语音信号, 并将远端 语音判断的结果输入所述控制信号发送单元;
所述控制信号发送单元设置成: 收到近端语音判断的结果和远端语音判 断的结果后,发出控制信号控制所述自适应滤波器部和所述路径改变探测部。
6、如权利要求 5所述的回波抵消器, 其中, 所述控制信号发送单元设置 成按照以下方式发出控制信号控制所述自适应滤波器部和所述路径改变探测 部:
在近端通话或静音时, 关闭所述自适应滤波器;
双端通话时, 锁定所述自适应滤波器, 即不对所述自适应滤波器系数进 行更新:
非近端通话、 非静音、 且非双端通话时, 启动所述路径改变探测部。
7、如权利要求 5所述的回波抵消器, 其中, 所述检测门限初始化单元设 置成按照以下方式估计语音检测门限的初始值:
釆用 ThresholdO = K . E(v2 )计算语音检测门限的初始值, 其中, ThresholdO为 语音检测门限的初始值, K为语音门限估计增益, £(ν2 ) = ^; («)2 , 为近 端信号, N为信号长度。
8、 如权利要求 5所述的回波抵消器, 其中:
所述近端语音判断单元设置成按照以下方式判断是否存在近端语音信号:
Μ+Νγ-\
在 〉7¾ ra 7oW时对计数器 C(n)力 1 , SUMC = C(i > Threshoid2 ^ , i=M
判断所述近端信号中存在近端语音信号, 其中, dl(n)为自适应滤波器的前 L 阶系数和远端语音信号的卷积, Threshold 为第一近端语音检测门限, Threshold2为第二近端语音检测门限, 为滑动窗长度, M = M + S , M为滑 动窗起点, S为滑动窗滑动长度; 或者,在 J (/)2〉 J¾m^oW时对计数器 C(n)
i=l
Μ+Νλ-\
力口 1 , SUMC = C(i、> Threshold! ^ , 判断所述近端信号中存在近端语音
i=
信号, 其中, d(n)为近端信号, ^为帧长, Threshold为第一近端语音检测门 限, Threshold2为第二近端语音检测门限, 为滑动窗长度, M = M + S , M 为滑动窗起点, S为滑动窗滑动长度;
所述远端语音判断单元设置成按照以下方式判断是否存在远端语音信号:
M+N2-\
在 x(")2〉7¾m^o l时对计数器 D(n)力 1 , 在^ 7 D = Ό(Γ> > Threshold3 , i=M
判断所述近端信号中存在远端语音信号, 其中, Threshholdl为第一远端语音 检测门限, Threshold3为第二远端语音检测门限, ^为滑动窗长度。
9、 如权利要求 8所述的回波抵消器, 其中:
所述第一近端语音检测门限与所述第一远端语音检测门限相等, 所述第 二近端语音检测门限与所述第二远端语音检测门限相等。
10、 如权利要求 5所述的回波抵消器, 其中, 所述语音信号探测部还包 括相关性计算单元和门限改变控制单元, 其中:
所述相关性计算单元设置成: 接收所述路径改变探测部发送的路径改变 探测结果,在回波路径发生改变时,计算近端信号与远端语音信号的相关性, 当相关性大于相关性比较门限时, 确定所述近端信号中无近端语音信号, 并 检查所述近端语音判断单元的判断结果, 在所述判断结果为存在近端语音信 号时, 确定第一近端语音检测门限失效;
所述门限改变控制单元设置成: 在第一近端语音检测门限失效时, 重新 计算近端语音的检测门限, 并向所述近端语音判断单元发出控制信号改变近 端语音的检测门限。
1 1、如权利要求 10所述的回波抵消器, 其中, 所述门限改变控制单元设 置成按照以下方式 判断是否存在 r C( ) > Threshold! ,如果
Figure imgf000022_0001
存在, 则确定 Threshold = · (ί 2) , 其中, Threshold 为第一近端语音检测门 限, Thresholds为相关性比较门限, 为信号方差, 7,为语音门限估计增益。
12、 如权利要求 1-1 1中任一项所述的回波抵消器, 其中, 所述路径改变 探测部包括: 第二自适应滤波器、 自适应滤波器性能计算单元、 第二自适应 滤波器性能计算单元和性能比较单元, 其中:
所述第二自适应滤波器的输入信号为远端语音信号, 期望信号为由所述 自适应滤波器的前 L阶系数和远端语音信号卷积得到的值, 误差信号为所述 卷积的误差信号, 设置成: 模拟回波路径的前 L个系数;
所述自适应滤波器性能计算单元设置成:计算所述自适应滤波器的性能, 并将结果发送给所述性能比较单元;
所述第二自适应滤波器性能计算单元设置成: 计算所述第二自适应滤波 器的性能, 并将结果发送给所述性能比较单元;
所述性能比较单元设置成: 接收所述自适应滤波器性能计算单元和所述 第二自适应滤波器性能计算单元发送的结果后, 判断是否发生了路径突变, 在确定路径突变时重启所述自适应滤波器, 同时开启所述语音信号探测部的 相关性计算单元; 在确定未发生路径突变时, 正常更新所述自适应滤波器。
13、 如权利要求 12所述的回波抵消器, 其中:
所述自适应滤波器性能计算单元设置成按照以下方式计算所述自适应滤 波器性能: 按照皿 £2 = 101ogl。 ^^计算自适应滤波器的性能, 其中,
ERLE2 为自适应滤波器性能值, d( )为近端信号, e(n)为近端信号的误差信 号;
所述第二自适应滤波器性能计算单元设置成按照以下方式计算所述第二 自适应滤波器性能: 按照 = l。logl。∑ dl(n) 计算所述第二自适应滤波器 的性能, 其中, ERLE1为第二自适应滤波器性能值, dl(n)为所述第二自适应 滤波器的前 L阶系数和远端语音信号的卷积, el(n)为所述卷积的误差信号; 所述性能比较单元设置成按照以下方式判断是否发生了路径突变: 接收 所述自适应滤波器性能计算单元和所述第二自适应滤波器性能计算单元计算 得到的性能值, 判断是否存在 ERLEl - ERLE2 > 7¾r^½ 4 , 其中, Threshold4 为 路径突变门限, 如果存在, 则确定回波路径突变。
14、 一种回波 4氏消方法, 包括:
远端语音信号通过扬声器在回波路径中传播, 被话筒拾取, 形成回波信 号;
自适应滤波器接收远端语音信号作为训练信号模拟回波路径, 对近端信 号进行回波信号的消除; 语音信号探测部检测通话状态, 根据通话状态控制所述自适应滤波器, 并根据通话状态控制路径改变探测部的启动;
所述路径改变探测部检测回波路径是否发生改变, 根据回波路径是否发 生改变控制所述自适应滤波器。
15、如权利要求 14所述的回波抵消方法, 该回波抵消方法还包括: 将所 述自适应滤波器的初始系数置为 0。
16、 如权利要求 14所述的回波抵消方法, 该回波抵消方法还包括: 随机序列发送部在建立起通话链路时, 向所述扬声器发送随机序列, 所 述随机序列经回波路径和麦克风形成回波信号, 传送到所述自适应滤波器, 对所述自适应滤波器进行初始化;
所述自适应滤波器釆用所述随机序列形成的回波信号进行训练, 将训练 得到的系数存储为初始系数。
17、 如权利要求 15或 16所述的回波抵消方法, 其中, 语音信号探测部 检测通话状态, 根据通话状态控制所述自适应滤波器, 并根据通话状态控制 路径改变探测部的启动的步骤包括:
所述语音信号探测部在建立起通话链路时, 存储所述近端信号, 釆用所 存储的近端信号估计语音检测门限的初始值, 作为近端语音检测门限的初始 值和远端语音检测门限的初始值;
所述语音信号探测部判断是否存在近端语音信号, 且判断是否存在远端 语音信号, 发出控制信号控制所述自适应滤波器部和所述路径改变探测部。
18、如权利要求 17所述的回波抵消方法, 其中, 所述语音信号探测部发 出控制信号控制所述自适应滤波器部和所述路径改变探测部的步骤包括: 在近端通话或静音时, 关闭所述自适应滤波器;
双端通话时, 锁定所述自适应滤波器, 即不对所述自适应滤波器系数进 行更新:
非近端通话、 非静音、 且非双端通话时, 启动所述路径改变探测部。
19、如权利要求 17所述的回波抵消方法, 其中, 所述语音信号探测部釆 用所存储的近端信号估计语音检测门限的初始值的步骤包括:
所述语音信号探测部釆用 ThreshddO = Κ · Ε(ν2 )计算语音检测门限的初始值, 其中, Threshold^为语音检测门限的初始值, K 为语音门限估计增益, Ε(ν2) = ^^ν(η)2 , ")为近端信号, N为信号长度。
20、 如权利要求 17所述的回波抵消方法, 其中:
所述语音信号探测部判断是否存在近端语音信号的步骤包括: 在 〉 7¾ ra 7oW时对计数器 C(n)加 1 , SUMC = C(i > Threshoid2 ^ , 判
i=
断所述近端信号中存在近端语音信号, 其中, dl(n)为自适应滤波器的前 L阶 系数和远端语音信号的卷积, Threshold为第一近端语音检测门限, Threshold2 为第二近端语音检测门限, 为滑动窗长度, M = M + S , M为滑动窗起点,
S为滑动窗滑动长度; 或者, 在 J (/)2〉7¾m^oW时对计数器 C(n)加 1 , 在 SUMC = C(i、> Threshom , 判断所述近端信号中存在近端语音信号, 其
i=
中, d(n)为近端信号, ^为帧长, Threshold 为第一近端语音检测门限, Threshold2为第二近端语音检测门限, 为滑动窗长度, M = M + S , M为滑 动窗起点, S为滑动窗滑动长度;
所述语音信号探测部判断是否存在远端语音信号的步骤包括: 在
M+N2-\
: ί)2 > ¾m^2oM时对计数器 D(n)加 1 , SUMD = Ό(ί > Threshold^ , 判 i=M
断所述近端信号中存在远端语音信号, 其中, Threshholdl为第一远端语音检 测门限, Threshold3为第二远端语音检测门限, ^为滑动窗长度。
21、 如权利要求 20所述的回波抵消方法, 其中:
所述第一近端语音检测门限与所述第一远端语音检测门限相等, 所述第 二近端语音检测门限与所述第二远端语音检测门限相等。
22、 如权利要求 17所述的回波 4氏消方法, 该回波抵消方法还包括: 所述语音信号探测部接收所述路径改变探测部发送的路径改变探测结果, 在回波路径发生改变时, 计算近端信号与远端语音信号的相关性, 当相关性 大于相关性比较门限时, 确定所述近端信号中无近端语音信号, 并检查所述 近端语音判断单元的判断结果, 在所述判断结果为存在近端语音信号时, 确 定第一近端语音检测门限失效;
所述语音信号探测部在第一近端语音检测门限失效时, 重新计算近端语 音的检测门限, 并发出控制信号改变近端语音的检测门限。
23、如权利要求 22所述的回波抵消方法, 其中, 重新计算近端语音的检 测门限的步骤包括: 判断是否存在 r &&SUMC = ¾C( ) > Threshold! ,如果
Figure imgf000026_0001
存在, 则确定 Threshold = · (ί 2) , 其中, Threshold 为第一近端语音检测门 限, Thresholds为相关性比较门限, 为信号方差, 7,为语音门限估计增益。
24、 如权利要求 14-23中任一项所述的回波抵消方法, 该回波抵消方法 还包括:
将远端语音信号作为第二自适应滤波器的输入信号, 由所述自适应滤波 器的前 L阶系数和远端语音信号卷积得到的值作为所述第二自适应滤波器的 期望信号, 将所述卷积的误差信号作为所述第二自适应滤波器的误差信号, 所述第二自适应滤波器模拟回波路径的前 L个系数;
计算所述自适应滤波器的性能和所述第二自适应滤波器的性能, 判断是 否发生了路径突变, 在确定路径突变时重启所述自适应滤波器, 同时启动相 关性计算; 在确定未发生路径突变时, 正常更新所述自适应滤波器。
25、 如权利要求 24所述的回波抵消方法, 其中: 计算所述自适应滤波器的性能的步骤包括: 按照 ERLE2 = 101og10d(nf 2 计算自适应滤波器的性能, 其中, ERLE2为自适应滤波器性能值, d(n)为近 端信号, e(n)为近端信号的误差信号;
波器的性能的 步骤 包括: 按照 二自适应滤波器的性能, 其中, ERLE1为
Figure imgf000026_0002
第二自适应滤波器性能值, dl(n)为所述第二自适应滤波器的前 L阶系数和远 端语音信号的卷积, el(n)为所述卷积的误差信号;
判断是否发生 了 路径突变的 步骤 包括: 判 断是否存在 ERLE1 - ERLE2 > Threshold , 其中, Threshold4为路径突变门限, 如果存在, 则 确定回波路径突变。
PCT/CN2012/072395 2011-12-26 2012-03-15 一种回波抵消器及回波抵消方法 WO2013097357A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US14/368,953 US9282195B2 (en) 2011-12-26 2012-03-15 Echo canceller and echo cancellation method
ES12863916.8T ES2644579T3 (es) 2011-12-26 2012-03-15 Cancelador de eco y procedimiento de cancelación de eco
EP12863916.8A EP2785032B1 (en) 2011-12-26 2012-03-15 Echo canceller and echo cancellation method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201110442191.3 2011-12-26
CN201110442191.3A CN103179296B (zh) 2011-12-26 2011-12-26 一种回波抵消器及回波抵消方法

Publications (1)

Publication Number Publication Date
WO2013097357A1 true WO2013097357A1 (zh) 2013-07-04

Family

ID=48638906

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2012/072395 WO2013097357A1 (zh) 2011-12-26 2012-03-15 一种回波抵消器及回波抵消方法

Country Status (6)

Country Link
US (1) US9282195B2 (zh)
EP (1) EP2785032B1 (zh)
CN (1) CN103179296B (zh)
ES (1) ES2644579T3 (zh)
PT (1) PT2785032T (zh)
WO (1) WO2013097357A1 (zh)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2515592B (en) * 2013-12-23 2016-11-30 Imagination Tech Ltd Echo path change detector
GB2536742B (en) * 2015-08-27 2017-08-09 Imagination Tech Ltd Nearend speech detector
EP3354004B1 (en) * 2015-09-25 2021-10-27 Microsemi Semiconductor (U.S.) Inc. Acoustic echo path change detection apparatus and method
CN105677240B (zh) * 2015-12-30 2019-04-23 上海联影医疗科技有限公司 数据删除方法及系统
CN107786755B (zh) * 2016-08-30 2020-12-15 合肥君正科技有限公司 一种双端通话检测方法和装置
CN108076239B (zh) * 2016-11-14 2021-04-16 深圳联友科技有限公司 一种改善ip电话回声的方法
WO2019051841A1 (zh) * 2017-09-18 2019-03-21 深圳市汇顶科技股份有限公司 确定滤波器系数的方法及其装置、终端
US10880440B2 (en) * 2017-10-04 2020-12-29 Proactiv Audio Gmbh Echo canceller and method therefor
CN107888792B (zh) * 2017-10-19 2019-09-17 浙江大华技术股份有限公司 一种回声消除方法、装置及系统
CN108712585A (zh) * 2018-05-09 2018-10-26 质音通讯科技(深圳)有限公司 通话终端、滤波器的初始化方法、装置、设备及存储介质
GB2577905A (en) * 2018-10-10 2020-04-15 Nokia Technologies Oy Processing audio signals
CN110600048B (zh) * 2019-08-23 2022-03-25 Oppo广东移动通信有限公司 音频校验方法、装置、存储介质及电子设备
CN110995951B (zh) * 2019-12-13 2021-09-03 展讯通信(上海)有限公司 基于双端发声检测的回声消除方法、装置及系统
CN111277718B (zh) * 2020-01-21 2021-10-08 上海推乐信息技术服务有限公司 一种回声消除系统及其方法
CN111556210B (zh) * 2020-04-23 2021-10-22 深圳市未艾智能有限公司 通话语音处理方法与装置、终端设备和存储介质
CN111968658B (zh) * 2020-06-30 2024-02-06 北京百度网讯科技有限公司 语音信号的增强方法、装置、电子设备和存储介质
CN111970610B (zh) * 2020-08-26 2022-05-20 展讯通信(上海)有限公司 回声路径检测方法、音频信号处理方法及系统、存储介质、终端
CN112053700B (zh) * 2020-08-31 2023-02-07 浙江大华技术股份有限公司 场景识别方法、装置、电子设备以及计算机可读存储介质
DE102020213051A1 (de) * 2020-10-15 2022-04-21 Sivantos Pte. Ltd. Verfahren zum Betrieb eines Hörhilfegeräts sowie Hörhilfegerät
TWI761075B (zh) 2021-02-09 2022-04-11 瑞昱半導體股份有限公司 回音抵消系統以及回音抵消方法
CN114978238B (zh) * 2021-02-18 2024-07-02 瑞昱半导体股份有限公司 回音抵消系统以及回音抵消方法
CN113362842B (zh) * 2021-06-30 2022-11-11 北京小米移动软件有限公司 音频信号处理方法及装置
TWI790718B (zh) * 2021-08-19 2023-01-21 宏碁股份有限公司 會議終端及用於會議的回音消除方法
CN115881080B (zh) * 2023-03-02 2023-05-26 全时云商务服务股份有限公司 一种语音通信系统中的声反馈处理方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1494229A (zh) * 2002-10-30 2004-05-05 冲电气工业株式会社 带有回声路径变化探测器的回声消除器
US20070206777A1 (en) * 2006-02-22 2007-09-06 Alcatel Lucent Method of controlling an adaptation of a filter
CN101087320A (zh) * 2006-06-07 2007-12-12 米特尔网络公司 声学回波消除器中检测回波路径变化的改进方法和装置
CN101826328A (zh) * 2010-04-29 2010-09-08 浙江工业大学 嵌入式无线可视门铃中的回音抵消方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6035034A (en) 1996-12-20 2000-03-07 Telefonaktiebolaget L M Ericsson (Publ) Double talk and echo path change detection in a telephony system
FR2952263B1 (fr) 2009-10-29 2012-01-06 Univ Paris Descartes Procede et dispositif d'annulation d'echo acoustique par tatouage audio

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1494229A (zh) * 2002-10-30 2004-05-05 冲电气工业株式会社 带有回声路径变化探测器的回声消除器
US20070206777A1 (en) * 2006-02-22 2007-09-06 Alcatel Lucent Method of controlling an adaptation of a filter
CN101087320A (zh) * 2006-06-07 2007-12-12 米特尔网络公司 声学回波消除器中检测回波路径变化的改进方法和装置
CN101826328A (zh) * 2010-04-29 2010-09-08 浙江工业大学 嵌入式无线可视门铃中的回音抵消方法

Also Published As

Publication number Publication date
US9282195B2 (en) 2016-03-08
ES2644579T3 (es) 2017-11-29
EP2785032A1 (en) 2014-10-01
US20140329568A1 (en) 2014-11-06
EP2785032B1 (en) 2017-08-16
CN103179296B (zh) 2017-02-15
CN103179296A (zh) 2013-06-26
EP2785032A4 (en) 2015-06-10
PT2785032T (pt) 2017-10-30

Similar Documents

Publication Publication Date Title
WO2013097357A1 (zh) 一种回波抵消器及回波抵消方法
US6434110B1 (en) Full-duplex speakerphone circuit including a double-talk detector
KR101503991B1 (ko) 휴대형 오디오 장치 및 휴대형 오디오 통신 장치를 이용하여 통화를 행하기 위한 방법
EP2783504B1 (en) Acoustic echo cancellation based on ultrasound motion detection
US6282176B1 (en) Full-duplex speakerphone circuit including a supplementary echo suppressor
US8498407B2 (en) Systems and methods for double-talk detection in acoustically harsh environments
JP3922997B2 (ja) エコーキャンセラ
CN106713570B (zh) 一种回声消除方法和装置
US8116448B2 (en) Acoustic echo canceler
KR20060097759A (ko) 에코 소거 회로용 다운링크 활동 및 더블 토크 확률 검출기및 방법
US7613291B1 (en) Echo path change detection using dual sparse filtering
JP3979169B2 (ja) エコーキャンセラ
US7539300B1 (en) Echo canceller with enhanced infinite and finite ERL detection
US9083783B2 (en) Detecting double talk in acoustic echo cancellation using zero-crossing rate
JP4600423B2 (ja) エコーキャンセラ
CN111355855B (zh) 回声处理方法、装置、设备及存储介质
JP2000059270A (ja) 音響エコーキャンセラ
TW201039606A (en) Speakerphone apparatus
JP4900185B2 (ja) 拡声通話装置
JP2009218849A (ja) 音響エコー・キャンセル制御方法と装置。
JP2005311903A (ja) 拡声通話装置
JPH04252524A (ja) 反響消去装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12863916

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

REEP Request for entry into the european phase

Ref document number: 2012863916

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 14368953

Country of ref document: US

Ref document number: 2012863916

Country of ref document: EP