WO2010110071A1 - 信号処理方法、装置及びプログラム - Google Patents

信号処理方法、装置及びプログラム Download PDF

Info

Publication number
WO2010110071A1
WO2010110071A1 PCT/JP2010/054032 JP2010054032W WO2010110071A1 WO 2010110071 A1 WO2010110071 A1 WO 2010110071A1 JP 2010054032 W JP2010054032 W JP 2010054032W WO 2010110071 A1 WO2010110071 A1 WO 2010110071A1
Authority
WO
WIPO (PCT)
Prior art keywords
signal sequence
value
function
signal
correction
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.)
Ceased
Application number
PCT/JP2010/054032
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
薫 鈴木
皇 天田
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Toshiba Corp
Original Assignee
Toshiba Corp
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 Toshiba Corp filed Critical Toshiba Corp
Publication of WO2010110071A1 publication Critical patent/WO2010110071A1/ja
Priority to US13/240,353 priority Critical patent/US8630850B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B3/00—Line transmission systems
    • H04B3/02—Details
    • H04B3/20—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other
    • H04B3/23—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other using a replica of transmitted signal in the time domain, e.g. echo cancellers
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04M—TELEPHONIC COMMUNICATION
    • H04M9/00—Arrangements for interconnection not involving centralised switching
    • H04M9/08—Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic
    • H04M9/082—Two-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

Definitions

  • the present invention relates to a signal processing method, apparatus and program for echo cancellation.
  • the acoustic echo canceller transmits / receives voice through, for example, a transmission path, amplifies the voice (far-end voice) from the received counterpart terminal from the speaker, and transmits the voice of the own terminal (close-end voice) received by the microphone It is used for the loud speaker communication device.
  • a phenomenon occurs in which the voice produced by itself is returned as an echo (referred to as an acoustic echo) at the opposite terminal.
  • acoustic echo is known to interfere in conversation because a certain amount of delay occurs in the transmission path.
  • An echo canceller is used to remove or reduce such acoustic echoes. Since the receiving input that is the source of the acoustic echo is known on the own terminal side, if the same conversion as the transfer characteristic of the echo path, which is the wraparound path of the acoustic echo, is applied to the receiving input, the acoustic echo mixed in the microphone output Echo replicas can be generated internally to simulate. By subtracting this echo replica from the microphone output, it is possible to generate an output signal (error signal) in which the acoustic echo is suppressed. The trick to perform this process is the echo canceller.
  • a non-recursive linear filter with N taps is often used as a linear filter for simulating the transfer characteristics of the echo path, and an echo replica is generated by convolving the filter's tap coefficient with the reception input Do.
  • an adaptive algorithm is used which finds asymptotically based on the observation signal.
  • probability gradient algorithms are known that correct tap coefficients in the gradient (probability gradient) direction of the instantaneous squared error with respect to the tap coefficients.
  • LMS algorithm minimum mean square error algorithm
  • NLMS algorithm normalized LMS algorithm
  • the tap coefficient correction amount in the LMS algorithm or the NLMS algorithm is an amount of blue ceiling proportional to the reference signal and also proportional to the error signal. Therefore, in the case of double talk where near end speech and far end speech are simultaneously present, a large erroneous correction occurs in response to the near end speech included in the error signal. In order to avoid this erroneous correction, it is necessary to quantitatively suppress or completely stop the correction of the tap coefficient during double talk. When these algorithms are employed, it is necessary to provide a double talk detector in order to detect double talk and control the tap coefficient correction unit.
  • Patent Document 1 Japanese Patent No. 3870861 discloses an echo in which independent component analysis (Infomax method) based on information entropy maximization is applied to correct tap coefficients so that a reference signal and an error signal become independent of each other.
  • a canceller is disclosed.
  • the echo canceller of Patent Document 1 uses a function G (e (t)) of the error signal e (t) as a sign function: sign (e (t)), a hyperbolic tangent function: tanh (e (t)), or a sigmoid
  • the tap coefficient correction formula is used with the function: 1 / (1 + exp (-e (t))).
  • An algorithm using these functions is generically referred to as the Infomax system hereinafter.
  • the correction at the time of double talk can be suppressed because the correction scale of the coefficient becomes a limit with the Infomax method at the limit of a fixed level. From this, the Infomax system has an advantage that the apparatus configuration can be further reduced without the need to provide a double talk detector.
  • the residual echo level of the Infomax system is high, that is, the echo remains unerased. This is considered to be due to the occurrence of an overshoot since the coefficient correction scale at the time of a minute error is excessive.
  • Patent Document 2 Japanese Patent No. 2885269
  • a tap coefficient is provided in a function G (e (t)) in which a linear band in which the correction amount is proportional to the error signal e (t) in the small error range of the sign algorithm
  • G e (t)
  • the algorithm of Patent Document 2 Due to the overshoot suppression effect of the linear band, the algorithm of Patent Document 2 has a residual echo level smaller than that of the coding algorithm.
  • the algorithm of Patent Document 2 is an algorithm in which a limiter of a correction scale is provided to the LMS algorithm. An algorithm using this function is hereinafter referred to as the Ideal Limiter scheme.
  • Patent Document 3 Japanese Patent Laid-Open No. 2004-64681 discloses an adaptive algorithm that has the same effect of accelerating the convergence as the LMF (Least Mean Fourth) algorithm.
  • the algorithm of Patent Document 3 does not provide robustness to double talk as the LMS algorithm does.
  • the correction scale of the tap coefficient monotonously increases in accordance with the instantaneous absolute value of the error signal, erroneous correction at the time of double talk becomes large as in the LMF (Least Mean Fourth) algorithm.
  • An object of the present invention is to provide a signal processing method, apparatus and program with improved echo cancellation capability and convergence speed.
  • a filter that performs filter processing to convolute tap coefficients into a first signal sequence to generate a second signal sequence, and a third signal sequence that includes an echo of the first signal sequence
  • a subtractor configured to subtract the second signal sequence from the second signal sequence to generate a fourth signal sequence
  • a correction unit configured to correct the tap coefficient according to a correction amount using a function of the fourth signal sequence as a parameter
  • the fourth signal series is smaller than the straight line in a negative range with respect to a straight line having the slope and the value of the function as the slope and the intercept, respectively, when the fourth signal series is 0.
  • at least one of the first region having a value and the second region having a value larger than the straight line in the positive range, and the value of the function is limited. Characteristics and Providing that the signal processing device.
  • Block diagram showing a signal processing apparatus (acoustic echo canceller) according to an embodiment of the present invention A flowchart showing the flow of processing of the signal processing device of FIG. 1 Diagram showing graphs of three functions of Infomax method Diagram showing a graph of the function of Ideal Limiter method Diagram schematically showing the shape of a function G (e (t)) based on the prior art Diagram schematically showing the shape of a function G (e (t)) based on an embodiment of the present invention Diagram showing a graph of the function of the overcorrected band Ideal Limiter scheme according to an embodiment of the present invention Transition of ERLE maximum value when over-correction bandwidth B is changed Transition of convergence speed when the overcorrection bandwidth B is changed Transition of ERLE maximum value when amplitude is changed Figure showing a graph of the sign function with deadband Diagram showing transition of ERLE maximum value when dead zone width C is changed Diagram showing transition of ERLE maximum value when amplitude is changed Diagram showing transition of ERLE maximum value when changing function Diagram showing transition (normalization) of ERLE maximum
  • a signal processing apparatus 1 includes a linear filter 2, a subtractor 3 and a tap coefficient correction unit 4.
  • the far-end voice which is the reception input from the transmission path to the reception input terminal 5, is input to the linear filter 2 and the tap coefficient correction unit 4 of the signal processing device 1 as a reference signal x (t) (first signal sequence).
  • the speaker 6 outputs a loud sound.
  • an output m (t) (third signal sequence) from the microphone 7 from which the near-end voice and the acoustic echo in which the loudspeaker output of the loudspeaker 6 wraps around through the echo path is received is input to the signal processing device 1 Be done.
  • the linear filter 2 is a non-recursive linear filter (FIR filter) with N taps for simulating the transfer characteristic of the echo path, and the echo replica signal y (t of acoustic echo) from the received reference signal x (t) ) (Second signal sequence) is generated according to equation (1). That is, in the linear filter 2, the echo replica signal y (t) is generated by performing filter processing in which the tap coefficient w (k, t) of the filter 2 is convoluted with the reference signal x (t).
  • FIR filter non-recursive linear filter
  • the echo replica signal y (t) is a time-series signal and represents the value of the echo replica at time t.
  • w (k, t) represents the value of the tap coefficient of the tap number k of the filter 2 at time t.
  • x (tk) is the value of the reference signal which went back to time k from time t.
  • N is the number of taps.
  • W (t) and X (t) are values of tap coefficient w (k, t) and reference signal x (tk) when k is changed from 0 to N-1 as shown in the following equation (2) It is a column vector which arranged each.
  • the echo replica signal y (t) thus generated is subtracted from the microphone output m (t) in the subtractor 3 as shown in equation (3).
  • an error signal (fourth signal sequence) e (t) instantaneous value which is a transmission output is generated and output from the transmission output terminal 8.
  • the tap coefficient correction unit 4 corrects the tap coefficient of the linear filter 2 using the tap coefficient correction equation represented by the recurrence equation of Equation (4).
  • Equation (4) is a generalized tap coefficient correction equation based on the probability gradient algorithm.
  • W (t) is a tap coefficient before correction
  • W (t + 1) is a tap coefficient after correction.
  • the second term on the right side of Equation (4) represents the correction amount of the tap coefficient.
  • a positive number ⁇ is a normalization coefficient
  • a positive number ⁇ is a step size for controlling the magnitude of correction
  • G (e (t)) is a function of the error signal (instant value) e (t), both of which are scalar quantities It is.
  • the product of the normalization coefficient ⁇ and the step size ⁇ is referred to as a step size as one quantity, but for the sake of explanation, these two quantities ⁇ and ⁇ are treated separately.
  • the shape of the function G (e (t)) of the error signal e (t), which is one of the parameters of the coefficient correction amount used in the tap coefficient correction unit 4, is It is different from technology.
  • the state of the signal processing device 1 is initialized. Specifically, the tap coefficient of the linear filter 2 is set to an initial value (in many cases, this initial value is 0).
  • the microphone output m (t) and the far-end voice x (t) are input.
  • the signal input processing step S2 includes processing to output the far-end voice x (t) by the speaker 6 as a loud output.
  • an echo replica y (t) of acoustic echo is generated according to the equation (1) using the input far-end speech x (t) as a reference signal. This process is performed by the linear filter 2.
  • an error signal e (t) is generated by subtracting the echo replica y (t) from the microphone output m (t) according to the equation (2).
  • the subtraction processing step S4 includes processing of transmitting and outputting the obtained error signal e (t). This process is performed by the subtractor 3.
  • the tap coefficients of the linear filter 2 are corrected from the error signal e (t) using the tap coefficient correction equation of equation (4). This process is executed by the tap coefficient correction unit 4.
  • an algorithm in which the normalization coefficient ⁇ of the LMS algorithm is not 1 but 1 / X T X is the NLMS algorithm (normalized LMS algorithm).
  • the definition of the NLMS algorithm will be described later.
  • X T X is the sum of powers of N reference signal values from the present to N-1 past times.
  • the LMS algorithm or the NLMS algorithm is an algorithm that finds tap coefficients that minimize the mean square value of the error signal asymptotically using the error signal e (t) (instant value) of each time.
  • the tap coefficient correction amount is generally proportional to the reference signal x (t) as shown in the equation (4)
  • FIGS. 3 (a), (b) and (c) The graphs of these functions are shown in FIGS. 3 (a), (b) and (c).
  • the horizontal axis of the graph is the value (instantaneous value) of the error signal e (t), and the vertical axis is the value of the function G (e (t)).
  • a scheme adopting sign (e (t)) is known as a sign algorithm.
  • tanh (e (t)) and 1 / (1 + exp (-e (t)) are sign (sign) in the full wave range (both positive and negative range) and half wave range (only positive range) of e (t) respectively It is a function approximating e (t)), and these functions are considered to function in the same way as the sign algorithm.
  • the correction scale of the coefficient becomes a ceiling with a certain level at the limit, so erroneous correction at double talk is suppressed and a double talk detector is unnecessary. It becomes.
  • the residual echo level is high in the Infomax system. This is considered to be due to the occurrence of an overshoot since the coefficient correction scale at the time of a minute error (the origin of FIGS. 3A, 3B, 3C) is excessive.
  • the correction amount of the tap coefficient is within the small error range (
  • a linear band proportional to the error is provided in the function G (e (t)).
  • FIG. 4 A graph of the function of equation (7) is shown in FIG.
  • the horizontal axis of the graph of FIG. 4 is the error signal e (t) (instantaneous value), and the vertical axis is the value of the function G (e (t)).
  • a linear band (referred to as a width A) in which the correction amount of the tap coefficient is proportional to the error signal e (t) in the small error range of the sign algorithm is provided in the function G (e (t)).
  • the residual echo level decreases due to the overshoot suppression effect of the linear band.
  • the algorithm of Patent Document 2 provides the LMS algorithm with a limiter on the scale of correction of tap coefficients. Therefore, the system of Patent Document 2 is hereinafter referred to as the Ideal Limiter system. Due to the effect of this limiter, similar to the Infomax system, the Ideal Limiter system exhibits robustness to double talk.
  • FIG. 5 schematically shows the geometrical feature of the function G (e (t)) of the above-described prior art (LMS algorithm, NLMS algorithm, Infomax method and Ideal Limiter method).
  • FIG. 6 schematically shows the shape feature of the function G (e (t)) in the present embodiment.
  • the function G (e (t)) in the prior art will be described below.
  • the function G (e (t)) is identical to the straight line F (e (t)). That is, it has a linear correction characteristic over the entire range of the error. It is said that the correction scale of the tap coefficient is constant.
  • the function G (e (t)) passes through the left side of the straight line F (e (t)) at e (t) ⁇ 0, and the straight line F (e (t)> 0 at e (t)> 0. Pass the right side of t)). That is, it has a characteristic of correcting the tap coefficient more than others in the minute error range. In other words, the correction magnitudes of the other error bands are smaller than the correction magnitudes of the small error band.
  • the linear band is the same as the LMS algorithm, and the other is the same as the Infomax system.
  • the function G (e (t)) is a straight line F (e (t) at (t) ⁇ 0. It does not pass on the right side of) and does not pass on the left side of the straight line F (e (t)) when e (t)> 0.
  • ⁇ D: C> 0, D> C) away from the origin is set to a small error range (
  • C and D are a first threshold and a second threshold.
  • the function G (e (t)) of the shape shown in FIG. 6 based on the present embodiment has two regions corresponding to the error signal e (t) as an overcorrection band, that is, a range where the error signal e (t) is negative.
  • (E (t) ⁇ 0) passes through the right side of the straight line F (e (t)) (that is, it has a smaller value than F (e (t))), and the error signal is in a positive range (e (t)> 0) has a second region passing through the left side of the straight line F (e (t)) (that is, having a value larger than F (e (t))).
  • the first region and the second region can be expressed as follows using the magnitude relationship between the function G (e (t)) and the straight line F (e (t)).
  • the first area is an area where G (e (t)) ⁇ F (e (t)) when e (t) ⁇ 0
  • the second area is G (e (e (t)) when e (t)> 0. t))> F (e (t)).
  • the function G (e (t)) in FIG. 6 has both the first region and the second region as the overcorrection zone, but may have one or both as the overcorrection zone.
  • the function G (e (t)) in FIG. 6 is provided with a limiter so that the robustness to double talk can be obtained as in the Infomax system and the Ideal Limiter system. Specifically, it is assumed that the value of the function G (e (t)) does not deviate from ⁇ ⁇ , that is,
  • ⁇ , a function of
  • the function G (e (t)) it needs to be included.
  • the intercept b does not necessarily have to be 0, by using a function G (e (t)) in which b is 0, tap coefficient correction characteristics symmetrical with respect to the sign of the error signal e (t) can be obtained. It becomes possible to give.
  • a tap coefficient correction equation having an overcorrection band which is defined as described above, will be collectively referred to as an overcorrection banded algorithm.
  • the function G (e (t)) is such that (a) the absolute value
  • the tap coefficient correction unit 4 in FIG. 1 can realize one or more implementation methods of the overcorrection banded algorithm described below, and can be appropriately selected according to the setting.
  • the function G (e (t)) of the equation (8) is an edge portion of the linear band of the Ideal Limiter scheme (A ⁇ B ⁇
  • AB corresponds to the first threshold D described above.
  • Equation (8) linearly corrects in a small error region where
  • This algorithm is expected to accelerate convergence of the Ideal Limiter method by applying correction.
  • this will be referred to as an over-correction banded Ideal Limiter method.
  • the overcorrection bandwidth B is a threshold for the amplitude value of the signal, it depends on the scale of the error signal e (t). Therefore, the echo cancellation ability (ERLE maximum value) was investigated when the amplitudes of the microphone output m (t) and the reference signal x (t) were doubled, tripled and quadrupled. This corresponds to doubling, three times, or four times the amplitude of the error signal e (t).
  • 10 (a) (b) (c) (d) show the relationship between the overcorrection band width B and the ERLE maximum value when the amplitude magnification is changed.
  • C A ⁇ B (0 ⁇ C ⁇ A)
  • the amplitude is doubled, tripled, quadrupled from this experiment.
  • C A-B which becomes ERLE maximum also becomes large.
  • C0 is the lower end of the over-correction band at which ERLE maximum is obtained by a previous experiment
  • M0 is the average value (average amplitude absolute value) of the amplitude absolute values of the error signal e (t) used at that time. is there.
  • Me is an average (moving average amplitude absolute value) of amplitude absolute values from the present of the actual error signal e (t) to the past of a predetermined period.
  • the ERLE maximum value drops sharply on the right side of the over-correction bandwidth B at which the ERLE is maximum. Therefore, if the overcorrection band lower end C calculated according to the equation (10) is applied as it is, there is a risk that the ERLE maximum value may rush into a region where the ERLE maximum value falls sharply due to measurement errors such as C0, M0, Me.
  • the tap coefficient correction unit 4 sets the overcorrection band lower limit C to a value obtained by adding a slight margin to the value while calculating the overcorrection band lower end C according to the equation (10). This margin shall be experimentally selected as a good value. At least the tap coefficient correction unit 4 performs control to reduce the overcorrection bandwidth B according to the increase of the moving average amplitude absolute value Me.
  • the function G (e (t)) of the equation (11) is a dead zone in which the correction magnitude is 0 within the minute error range (
  • the feature is that the The function G (e (t)) in FIG. 11 can be realized by adding only one threshold process. At this time, a value C larger than 0 corresponds to the lower end of the overcorrection band of the equation (9), but here, it will be particularly referred to as a dead band width.
  • the algorithm of equation (11) is expected to have the effect of suppressing over-correction of the Infomax system by the dead zone.
  • the dead band width C is related to the size of the signal. Of course, if the signal level in the operation environment is expected to some extent, a fixed dead band C may be used. On the other hand, if not, it is better to control C dynamically. Since the dead zone width C is a threshold for the absolute value
  • the function G (e (t)) has an absolute value
  • the first function which outputs 0, and is given by the second function which outputs the value of the same sign as that of the error signal e (t) at other times.
  • the second function is specifically sign (e (t)), tanh (e (t)), 1 / (1 + exp (-e (t))), or 1 / (1 + exp (-e (t))).
  • FIG. 11 is an example of the case where the second function is sign (e (t)).
  • FIG. 12 shows the change of the ERLE maximum value in 29 seconds when the dead zone width C is changed.
  • the dead zone width C is a threshold for the amplitude value of the signal, it depends on the scale of the error signal e (t). Therefore, the echo cancellation capability (ERLE maximum value) was investigated when the amplitudes of the microphone output m (t) and the reference signal x (t) were doubled, tripled, and quadrupled. This corresponds to doubling, three times, or four times the amplitude of the error signal e (t).
  • the dead zone width C 0 is the Infomax system.
  • the dead band width C at which the ERLE is maximum becomes approximately four times, nine times, or sixteen times. That is, the dead band width C at which ERLE is maximum is proportional to the square of the amplitude (power). Therefore, the dead band width C at which ERLE is maximum in this example is determined by equation (12).
  • C0 is a dead band width in the vicinity of the ERLE maximum obtained by a preliminary experiment
  • P0 is the average power of the error signal e (t) used at that time.
  • C0 and P0 are referred to as reference dead zone width C0 and reference power P0.
  • Pe is the moving average power from the present of the actual error signal e (t) to the past of a predetermined period.
  • the dead zone width C is experimentally selected to be a good value.
  • the ERLE maximum value falls steeper on the left side of the dead zone width C where ERLE is maximum than on the right side. Therefore, if the dead zone width C calculated according to the equation (12) is applied as it is, there is a risk that the ERLE maximum value may rush into a region where the ERLE maximum value falls sharply due to measurement errors such as C0, P0 and Pe.
  • the tap coefficient correction unit 4 sets the dead zone width to a value obtained by adding a margin to that value. This margin shall be experimentally selected as a good value. At least the tap coefficient correction unit 4 performs control to increase the dead zone width C according to the increase of the moving average power Pe.
  • Equation (7) (Normalized Ideal Limiter method with overcorrection zone)
  • the Ideal Limiter scheme shown in Equation (7) is a scheme in which a limiter is provided to the correction scale of the LMS algorithm. Then, what happens if a similar limiter is provided to the NLMS algorithm defined by the following equation (13)?
  • Equation (13) is called a normalized Ideal Limiter method, and is defined by the following equation (14).
  • 1 / X T X is the reciprocal of the sum of the power of N reference signal values from the present to N-1 time past, but instead of calculating it each time, the current value to be determined of X T X Pw (t), the value of one time past already calculated is Pw (t-1), and the current value x (t) of the reference signal and the positive coefficient ⁇ (forgetting factor) less than 1 are used to obtain the equation (19) It is also possible to calculate as in). Only by calculating X T X once at first, thereafter, the approximate value P w (t) of X T X is obtained with a small amount of calculation by equation (19). This is also a method of approximating 1 / X T X.
  • the tap coefficient correction unit 4 sets the overcorrection band width B to a value obtained by subtracting the margin from the overcorrection band width B at which the ERLE is maximum. This margin shall be experimentally selected as a good value.
  • FIG. 16 shows the result of comparison of ERLE at 4.46 seconds in the input signal 1-fold data as a measure of convergence speed.
  • the shape of the function G (e (t)) is not limited to the above (Equation 8) or (Equation 15), and e (t) falls within the over-correction band range.
  • G (e (t) is negative G (e (t)) ⁇ e (t) / A
  • e (t) is positive G (e (t))> e (t) / A
  • the function G (e (t)) conforms to the Ideal Limiter method shown in Expression (7) and the normalized Ideal Limiter method shown in Expression (14) over the entire range of the error signal e (t).
  • the over-correction band is the same as the over-correction band Ideal Limiter method shown in equation (10) or the over-correction band normalizing Ideal Limiter method shown in equation (15) It can produce an effect.
  • the function G (e (t)) conforms to the Infomax system of Equation (6) in the entire region of the error signal e (t).
  • H (e (t)) satisfies the feature of the function G (e (t)) based on this embodiment, where H (e (t)) is a function corresponding to the product of ⁇ , ⁇ and G.
  • H (e (t)) is a function corresponding to the product of ⁇ , ⁇ and G.
  • the adaptive filter converges quickly when the step size ⁇ is large but with low accuracy, and converges accurately when the step size ⁇ is small. Therefore, for example, by monitoring the tendency of the magnitude of the error signal becoming smaller and knowing that the filter convergence is progressing, control is performed such as making the step size ⁇ smaller according to the degree of progress, for example. It is possible to perform long-term control of ⁇ in which the convergence speed is prioritized and the convergence is advanced, the accuracy is prioritized. However, since the instantaneous value e (t) of the error signal is the signal itself, the value dynamically changes momentarily.
  • the convergence speed is first prioritized exclusively by the former effect, while maintaining the robustness to the double talk by the latter effect, If convergence is advanced, it is expected that it is possible to improve speed while giving priority to accuracy solely by the latter effect. That is, both are techniques that can be combined.
  • the LMF algorithm uses e 3 (t) rather than e (t) as the function G (e (t)). Therefore, in the whole range of error excluding 0, when e (t) ⁇ 0, G (e (t)) ⁇ F (e (t)), when e (t)> 0, G (e (t))> It becomes F (e (t)). This is a condition to be satisfied of the overcorrection band according to the present embodiment. Further, in the LMF algorithm, the correction scale monotonously increases in accordance with the instantaneous absolute value
  • the LMF algorithm is not limited to another condition to be satisfied by the function G (e (t)) based on the present embodiment,
  • the robustness to double talk can not be obtained.
  • the correction error at the double talk becomes larger than the LMS algorithm because the correction scale increases monotonously according to
  • Patent Document 3 is not limited to another condition to be satisfied by G (e (t)) based on the present embodiment,
  • the correction scale of the tap coefficient monotonously increases in accordance with the instantaneous absolute value
  • the linear filter 2 of FIG. 1 has been used as a filter for simulating the transfer characteristic of the echo path in the above description, it is also possible to replace the linear filter 2 with a non-linear filter.
  • a non-linear filter instead of the linear filter 2, it is possible to use a second or higher order Volterra filter known as a non-linear filter.
  • an arithmetic expression for calculating the echo replica y (t) in the second-order Volterra filter is shown in Expression (24).
  • the first term on the right side is a first-order linear term, which is similar to the linear filter equation described above.
  • the second term on the right side is a second-order nonlinear term.
  • the tap coefficients are corrected by the tap coefficients shown in equations (25) and (26) for N tap coefficients w1 forming a linear term of the Volterra filter and N ⁇ N tap coefficients w2 forming a nonlinear term. It is implemented by applying a correction expression.
  • the normalization coefficient ⁇ , the step size ⁇ , and the function G can be determined independently by Equation (25) and Equation (26).
  • the embodiment of the present invention not only makes the signal processing apparatus shown by the block diagram of FIG. 1 or the signal processing method shown by the flow chart of FIG. 2 but makes a computer function as the signal processing apparatus of FIG.
  • the present invention can also be implemented as a program that executes a procedure, or a computer-readable recording medium that stores the program.
  • the present invention can be implemented using a computer as shown in FIG.
  • the microphone 11 and the speaker 23 correspond to the microphone 7 and the speaker 6 shown in FIG.
  • the voice on the near end side is converted into an electrical sound signal by the microphone 11 and converted into digital sound data by the A / D converter 12.
  • the digital audio data from A / D converter 112 is processed by CPU 13 which executes program instructions.
  • Connected to the CPU 13 are a RAM 14, a ROM 15, an HDD 16, a LAN 17, a mouse / keyboard 18 and a display 19 which are standard devices constituting a computer.
  • the line interface 20 is a device that transmits and receives at least digital audio data to and from the far end.
  • Drives (other storage) 21 for supplying programs and data to the computer from outside via storage media are specifically CD-ROM drive, floppy (registered trademark) disk drive, CF / SD card slot and USB It is an interface etc.
  • the digital audio data on the far end side received and input via the line interface 20 is converted into an audio signal by the D / A converter 22 and output as an audio signal by the speaker 23.
  • a signal processing program for echo cancellation that executes the processing steps shown in FIG.
  • the microphone 11 and the A / D converter 12 are used for the input of the microphone output m (t) on the near end side, and the reception input x (t) from the far end side received and input via the line interface 20
  • the echo cancellation output (error signal) e (t) is generated by processing the microphone output m (t) and the reception input x (t) by the CPU 13 using the D / A converter 22 and the speaker 23 for the loud sound output. And transmit to the far end via the line interface 20.
  • the computer system of FIG. 17 functions as an acoustic echo canceller.
  • the computer apparatus can also receive the echo cancellation processing program from a recording medium inserted into the other storage 21 or another apparatus connected via the LAN 17.
  • the computer device can also receive an operation input from the user or present information to the user by using the mouse / keyboard 18 and the display 19.
  • the present invention can be implemented as a recording medium storing a program as shown in FIG.
  • a recording medium 31 realized by a CD-ROM or CF, an SD card, a floppy disk, a USB storage, or the like in which an echo cancellation signal processing program according to an embodiment of the present invention is recorded is used as the electronic device 32 or 33 or the robot 34.
  • the program can be executed, or the program can be supplied from the electronic device 33 supplied to the other electronic device 35 or the robot 34 to the electronic device 35 or the robot 34 by communication. Make the program executable.
  • the signal processing apparatus has been described above as an example of the acoustic echo canceller for removing the echo (acoustic echo) of the speaker output from the microphone output.
  • the acoustic echo canceller for removing the echo (acoustic echo) of the speaker output from the microphone output.
  • line echo generation of an echo in which the transmission output loops around to the reception input due to the hybrid transformer. This echo is called line echo.
  • the signal processing apparatus according to the embodiment of the present invention can also be used for a line echo canceller for removing such line echo.
  • FIG. 19 shows a speech communication apparatus provided with both the acoustic echo canceller 1 and the line echo canceller 9.
  • the signal processing apparatus described in the first embodiment is used for each echo canceller 1 and 9.
  • the acoustic echo is removed by the acoustic echo canceller 1 from the microphone output m (t) from the microphone 7 that receives and receives near-end speech, and the error signal e (t) is output.
  • the line echo canceller 9 the line echo is removed from the reception input x (t) received via the transmission line, and the error signal f (t) is output.
  • the error signal f (t) is amplified and output by the speaker 6.
  • the acoustic echo canceller 1 uses the error signal f (t) output from the line echo canceller 9 as the reference signal x (t).
  • the line echo canceller 9 uses the error signal e (t) output from the acoustic echo canceller 1 as a reference signal.
  • the present invention is not limited to the above embodiment as it is, and at the implementation stage, the constituent elements can be modified and embodied without departing from the scope of the invention.
  • various inventions can be formed by appropriate combinations of a plurality of constituent elements disclosed in the above embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, components in different embodiments may be combined as appropriate.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
PCT/JP2010/054032 2009-03-25 2010-03-10 信号処理方法、装置及びプログラム Ceased WO2010110071A1 (ja)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US13/240,353 US8630850B2 (en) 2009-03-25 2011-09-22 Signal processing method, apparatus and program

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2009-073902 2009-03-25
JP2009073902A JP5430990B2 (ja) 2009-03-25 2009-03-25 信号処理方法、装置及びプログラム

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US13/240,353 Continuation US8630850B2 (en) 2009-03-25 2011-09-22 Signal processing method, apparatus and program

Publications (1)

Publication Number Publication Date
WO2010110071A1 true WO2010110071A1 (ja) 2010-09-30

Family

ID=42780764

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2010/054032 Ceased WO2010110071A1 (ja) 2009-03-25 2010-03-10 信号処理方法、装置及びプログラム

Country Status (3)

Country Link
US (1) US8630850B2 (enExample)
JP (1) JP5430990B2 (enExample)
WO (1) WO2010110071A1 (enExample)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8363821B2 (en) 2010-03-31 2013-01-29 Kabushiki Kaisha Toshiba Apparatus and method for canceling echo
CN110767245A (zh) * 2019-10-30 2020-02-07 西南交通大学 基于s型函数的语音通信自适应回声消除方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8583428B2 (en) * 2010-06-15 2013-11-12 Microsoft Corporation Sound source separation using spatial filtering and regularization phases
JP5887535B2 (ja) * 2012-02-17 2016-03-16 パナソニックＩｐマネジメント株式会社 エコー消去装置、エコー消去方法、及び、通話装置
US8832170B2 (en) 2012-03-26 2014-09-09 King Fahd University Of Petroleum And Minerals System and method for least mean fourth adaptive filtering
US10381031B2 (en) 2015-03-31 2019-08-13 Seagate Technology Llc Adaptive disturbance rejection using dead zone filter
JP6678545B2 (ja) * 2016-09-12 2020-04-08 株式会社東芝 修正システム、修正方法及びプログラム
JP6672209B2 (ja) 2017-03-21 2020-03-25 株式会社東芝 情報処理装置、情報処理方法、および情報処理プログラム
JP7187183B2 (ja) * 2018-06-14 2022-12-12 株式会社トランストロン エコー抑圧装置、エコー抑圧方法およびエコー抑圧プログラム

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08213880A (ja) * 1995-02-01 1996-08-20 Nec Corp 適応制御ろ波器
JPH1028080A (ja) * 1996-07-12 1998-01-27 Nec Corp 低消費電力型通信用適応等化フィルタ
JP2000286763A (ja) * 1999-03-31 2000-10-13 Mitsubishi Electric Corp 波形等化装置
JP2005124026A (ja) * 2003-10-20 2005-05-12 Nippon Telegr & Teleph Corp <Ntt> 反響消去装置、方法、及び反響消去プログラム、そのプログラムを記録した記録媒体
JP2005124027A (ja) * 2003-10-20 2005-05-12 Nippon Telegr & Teleph Corp <Ntt> 反響消去装置、方法、及び反響消去プログラム、そのプログラムを記録した記録媒体

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4654863A (en) * 1985-05-23 1987-03-31 At&T Bell Laboratories Wideband adaptive prediction
JPS6365722A (ja) * 1986-09-05 1988-03-24 Fujitsu Ltd ハウリングキヤンセラ
EP0310055B1 (en) * 1987-10-02 1994-01-12 Advanced Micro Devices, Inc. Adaptive echo-canceller with double-talker detection
US5784532A (en) * 1994-02-16 1998-07-21 Qualcomm Incorporated Application specific integrated circuit (ASIC) for performing rapid speech compression in a mobile telephone system
JPH102239A (ja) * 1996-06-14 1998-01-06 Toyota Motor Corp ハイブリッド型車両のエンジン制御装置
KR100233463B1 (ko) * 1997-03-07 1999-12-01 윤종용 반향제거장치 및 방법
US6563803B1 (en) * 1997-11-26 2003-05-13 Qualcomm Incorporated Acoustic echo canceller
JP3296330B2 (ja) * 1999-05-20 2002-06-24 日本電気株式会社 適応フィルタ、適応フィルタの制御方法及びプログラムを記憶した記憶媒体
US6804203B1 (en) * 2000-09-15 2004-10-12 Mindspeed Technologies, Inc. Double talk detector for echo cancellation in a speech communication system
JP3870861B2 (ja) 2002-07-10 2007-01-24 株式会社デンソー エコーキャンセラ装置、及び音声通信装置
JP2004064681A (ja) 2002-07-31 2004-02-26 Japan Science & Technology Corp 適応フィルタ

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH08213880A (ja) * 1995-02-01 1996-08-20 Nec Corp 適応制御ろ波器
JPH1028080A (ja) * 1996-07-12 1998-01-27 Nec Corp 低消費電力型通信用適応等化フィルタ
JP2000286763A (ja) * 1999-03-31 2000-10-13 Mitsubishi Electric Corp 波形等化装置
JP2005124026A (ja) * 2003-10-20 2005-05-12 Nippon Telegr & Teleph Corp <Ntt> 反響消去装置、方法、及び反響消去プログラム、そのプログラムを記録した記録媒体
JP2005124027A (ja) * 2003-10-20 2005-05-12 Nippon Telegr & Teleph Corp <Ntt> 反響消去装置、方法、及び反響消去プログラム、そのプログラムを記録した記録媒体

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8363821B2 (en) 2010-03-31 2013-01-29 Kabushiki Kaisha Toshiba Apparatus and method for canceling echo
CN110767245A (zh) * 2019-10-30 2020-02-07 西南交通大学 基于s型函数的语音通信自适应回声消除方法

Also Published As

Publication number Publication date
JP5430990B2 (ja) 2014-03-05
JP2010226629A (ja) 2010-10-07
US20120072210A1 (en) 2012-03-22
US8630850B2 (en) 2014-01-14

Similar Documents

Publication Publication Date Title
WO2010110071A1 (ja) 信号処理方法、装置及びプログラム
JP4697465B2 (ja) 信号処理の方法、信号処理の装置および信号処理用プログラム
US9543926B2 (en) Signal processing method and device
JP5036874B2 (ja) エコー消去装置
JP4377952B1 (ja) 適応フィルタ及びこれを有するエコーキャンセラ
US20130191119A1 (en) Signal processing device, signal processing method and signal processing program
JP2003158476A (ja) エコーキャンセラ
JP4457639B2 (ja) エコーキャンセラ
CN106657507A (zh) 一种声学回声消除方法及装置
WO2012105880A1 (en) Estimation and suppression of harmonic loudspeaker nonlinearities
US9312913B2 (en) Echo cancellation apparatus, echo cancellation method and communications apparatus
EP2380339B1 (en) Determining an acoustic coupling between a far-end talker signal and a combined signal
CN111654585B (zh) 回声声场状态确定方法及装置、存储介质、终端
US8406430B2 (en) Simulated background noise enabled echo canceller
JP5057109B2 (ja) エコーキャンセラ装置
JP3982823B2 (ja) 音声処理装置、音声処理方法および音声処理プログラム
CN101765982A (zh) 回声消除器
JP2008263441A (ja) 非線形エコーキャンセラ装置
JP5025753B2 (ja) エコーキャンセル装置、方法
JP4903843B2 (ja) 適応フィルタ及びこれを有するエコーキャンセラ
JP4964267B2 (ja) 適応フィルタ及びこれを有するエコーキャンセラ
JP4452162B2 (ja) 通話状態値算出装置および算出方法
JP4903842B2 (ja) 適応フィルタ及びこれを有するエコーキャンセラ

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: 10755871

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 10755871

Country of ref document: EP

Kind code of ref document: A1