JP2011166239A - Echo canceling method, echo canceler, program thereof and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an echo canceling method that stably and efficiently cancels echo with less computation complexity even if an acoustic path between a reproduction means and a sound collection means is nonlinear, and sound volume significantly varies , and prevents deterioration in speech quality. <P>SOLUTION: A received signal is divided into a plurality of components by a predetermined method. A false echo is generated for each component by using a plurality of adaptive filters with multiplication coefficients generated according to the size of the components set thereto, and then the false echo is subtracted from a collected signal to obtain a transmission signal from which the echo is canceled. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an echo canceling method, an echo canceling apparatus, a program, and a recording medium using adaptive filter technology for performing effective echo canceling during communication of an audio signal.

  An echo canceller (echo canceling device), when picking up sound by a sound collecting means such as a microphone in a hands-free call device or the like, is emitted from a reproducing means such as a speaker and is converted into a sound collecting signal through the acoustic characteristics of the room. It is an apparatus used to secure a high-quality audio signal in which the echo component is suppressed by estimating the mixed echo component and subtracting it from the collected sound signal.

  Examples of conventional echo canceller techniques include a frequency domain echo canceller using “linear adaptive filter” in Non-Patent Document 1 and “nonlinear echo canceller using Volterra filter” in Non-Patent Document 2.

  FIG. 15 shows a functional configuration example of the echo removal apparatus 10 according to the technique disclosed in Non-Patent Document 1. The echo removal apparatus 10 generates a pseudo echo from each frequency component of the received signal, a first frequency converting unit 110 that divides the received signal in the time domain into frequency domain components by Fourier transform, and also receives the difference signal from the received signal. The adaptive filter 13 that sequentially updates its own filter coefficient from the second, the second frequency converter 140 that divides the time-domain sound-collected signal into frequency-domain components by Fourier transform, and the pseudo echo from the sound-collected signal for each frequency component The difference signal generation unit 150 generates a difference signal obtained by subtracting the difference signal, and the frequency inverse conversion unit 160 generates a time domain transmission signal by synthesizing each difference signal for each frequency component by inverse Fourier transform. The principle of operation is as follows.

(L is an integer greater than or equal to 2 in frame length, and T is transposed). Each of these Fourier transforms is

(In this case, the frequency division number K = L / 2 + 1 of Fourier transform). The filter coefficient of the frequency domain adaptive filter is

The update of the filter coefficient of the adaptive filter is

Further, μ is a step size (0 <μ <2) for controlling the update amount.
Next, the operation principle of the echo canceller of Non-Patent Document 2 will be described. In the echo canceller of Non-Patent Document 2, a linear adaptive filter is used.

In addition, an adaptive filter corresponding to a higher-order term of the Volterra series is prepared, and the received signal is processed as an input. For example, the second-order Volterra filter at time n and its input are considered for the past L ₂ samples of x.

Is calculated.

M.Dentino, J.McCool and B.Widrow, "Adaptive Filtering in Frequency Domain", Proceedings of the IEEE, 1978, Vol.66, No.12, p.1658 A. Stenger, L. Trautmann and R. Rabenstein, "Nonlinear acoustic echo cancellation with 2nd order adaptive volterra filters", International Conference on Acoustics, Speech, and Signal Processing, 1999, Vol.2, p.877-880

  In a hands-free communication device, the acoustic path between the speaker and the microphone may be nonlinear because the speaker has a small diameter. However, even in such a case, since the filter of Non-Patent Document 1 estimates only a linear echo assuming the linearity of the acoustic path, a non-linear echo remains in the transmitted signal, and the speech quality deteriorates. Further, in the filter of Non-Patent Document 2, the case where the acoustic path is nonlinear is also considered. However, when a high-order term is taken into consideration, the filter length is increased by the power of the order and the amount of calculation is increased. In addition, although it shows high performance for steady sounds in actual measurement, estimation may become unstable and may not be effective for voices with large fluctuations in volume, leaving problems for practical use as an echo canceller .

  It is an object of the present invention to cancel echoes stably and sufficiently with a small amount of computation even when the acoustic path between the reproducing means and the sound collecting means is non-linear and the volume fluctuation is large, thereby degrading the call quality. It is an object of the present invention to provide an echo canceling method that can prevent the above-described problem.

  The echo cancellation method of the present invention uses a plurality of adaptive filters in which a received signal is divided into a plurality of components by a predetermined method, and a multiplication coefficient generated according to the size of each component is set for each component. Then, a pseudo echo is generated and subtracted from the collected sound signal to obtain a transmission signal in which the echo is eliminated.

  The echo canceling method of the present invention is “noting the characteristics for each band and separately preparing the number of filters to be prepared for each band” and “preparing a plurality of filters for each band and inputting coefficients of their linear combination” This is a reflection of the technical feature of “variation according to volume” in the specific configuration and processing. Thereby, even if each filter estimates a fixed value, it is possible to accurately estimate an impulse response that fluctuates as a whole. Therefore, even if the acoustic path between the playback means and the sound collection means is non-linear and the fluctuation in volume is large, pseudo echoes close to the echoes that have passed through the acoustic path can be generated, so that the echoes can be erased stably and sufficiently. Therefore, it is possible to prevent deterioration of call quality. In addition, since a suitable number of filters can be prepared for each band, the efficiency of the memory can be improved. Furthermore, since the configuration is simple, it can be realized with a small amount of calculation.

FIG. 2 is a block diagram illustrating a configuration example of an echo cancellation apparatus 100. FIG. 3 is a block diagram showing a configuration example of a composite adaptive filter unit 130. The figure which shows the example of a processing flow of the echo cancellation method which concerns on the echo cancellation apparatus. The block diagram which shows the structural example of the echo cancellation apparatus. FIG. 3 is a block diagram illustrating a configuration example of a composite adaptive filter unit 230. The figure which shows the example of a processing flow of the echo cancellation method which concerns on the echo cancellation apparatus. The block diagram which shows the structural example of the echo cancellation apparatus 300. FIG. The block diagram which shows the structural example of the echo cancellation apparatus 400. FIG. The block diagram which shows the structural example of the composite adaptive filter part 430. FIG. The figure which shows the example of a processing flow of the echo cancellation method which concerns on the echo cancellation apparatus. The figure which shows the example of the impulse response measured with five types of sound volumes. The enlarged view of the part shown by A of FIG. The enlarged view of the part shown by B of FIG. The figure which shows the comparison of the average echo cancellation amount of the echo cancellation method of this invention, and the conventional echo cancellation method. The block diagram which shows the structural example of the echo cancellation apparatus 10 by a prior art.

  Hereinafter, embodiments of the present invention will be described in detail.

  FIG. 1 shows a configuration example of the echo canceling apparatus 100 according to the first embodiment, and FIG. 3 shows a processing flow example by an echo canceling method using the apparatus.

The echo canceling apparatus 100 includes a first frequency conversion unit 110, K multiplication coefficient calculation units 120 (K is the number of frequency components), K composite adaptive filter units 130, second frequency conversion units 140, and K differences. A signal generation unit 150 and a frequency inverse conversion unit 160 are provided. In addition, the k-th (k = 1, 2,..., K) composite adaptive filter unit 130 has M _k adaptations (M _k is a positive integer corresponding to the value of _k ) as shown in FIG. The filter 131 and the addition part 132 which adds those outputs are provided. The first frequency conversion unit 110, the second frequency conversion unit 140, the difference signal generation unit 150, and the frequency inverse conversion unit 160 play a role similar to that of the conventional echo removal apparatus 10.

  The first frequency converter 110 receives the time domain received signal at time n.

(L is the frame length) is input, frequency analysis is performed by discrete Fourier transform in units of frames, and divided into K frequency components X (1), X (2), ..., X (K) for output (S1). Here, K is L / 2 + 1. There are various frame shift and frame length setting methods and frequency analysis methods. For example, the frame shift is L / 2 with respect to the frame length L, and the Hamming window w ( i) = 0.54-0.46 cos (2πn / L)
Using,

And can be calculated.

The multiplication coefficient calculation unit 120 and the composite adaptive filter unit 130 are provided for each frequency component X (k). The k-th multiplication coefficient calculation unit 120 corresponding to the frequency component X (k) has an absolute value | X (k) | that is the magnitude of the frequency component X (k) obtained by the first frequency conversion unit 110, and Predetermined M _k −1 threshold values v _{k, j} (v _{k, j} is a positive real number v _{k, j−1} <v _{k, j} (j = 2,..., M _k −1)) _Are used to generate M _k multiplication coefficients β _{k, m} (m = 1, 2,..., M _k ) (S2). The multiplication coefficient β _{k, m} is a coefficient for setting with which weight each of the M _k adaptive filters 131 (filter coefficients thereof) is applied to the frequency component X (k). The threshold value v _{k, j} defines which of the M _k adaptive filters 131 is applied according to the value of | X (k) |. Therefore, there is only one threshold when M _k = 2 and there is no threshold when M _k = 1 (that is, the composite adaptive filter unit 130 is a single adaptive filter).

When M _k is 2 or more, the multiplication coefficient β _{k, m} is specifically determined as follows. Note that i = 1, 2,..., M _k except for the excluded portion.

Further, the threshold value v _{k, j} is, for example, the maximum value that can be taken by the absolute value | X (k) | of the frequency component is equally divided into M _k on a linear or logarithmic scale,

And so on.

Note that the threshold value v _{k, j} and the number M _k of the adaptive filters 131 may be the same value without depending on k, or may be changed depending on k. A value of about 2 to 10 is normally used for M _k according to a change in the magnitude of | X (k) |.

The filter coefficient of the mth adaptive filter 131 of the kth composite adaptive filter unit 130 corresponding to the frequency component X (k) is H _{k, m} , and the adaptive filter 131 has a corresponding multiplication coefficient β _{k, m} and Using the frequency component X (k), for example, pseudo echo Y ′ (k, m) by the m-th adaptive filter 131 is obtained by the following equation (S3).

The adaptive filter 131 updates the filter coefficient H _{k, m} using the multiplication coefficient β _{k, m} , the frequency component X (k), and the difference signal E (k), which will be described later.

The adder 132 adds the pseudo echoes Y ′ (k, 1), Y ′ (k, 2),..., Y ′ (k, M _k ) for each adaptive filter 131, and the frequency component X (k). A total pseudo echo Y ′ (k) based on the above is generated.

  The second frequency converter 140 is a time domain sound pickup signal at time n.

Are converted into K frequency components Y (1), Y (2),..., Y (K) by discrete Fourier transform in units of frames and output (S4). Here, K is L / 2 + 1. Specific conversion methods include those exemplified in the description of the first frequency conversion unit 110.

  A total of K difference signal generation units 150 are provided for each frequency component, and the difference signal E (k) is generated by subtracting the pseudo echo Y ′ (k) from the frequency component Y (k) of the collected sound signal (S5). .

The frequency inverse transform unit 160 includes difference signals E (1), E (2),..., E (L / 2 + 1), E ^* (L / 2), E ^* (L / 2-1) of each frequency component. ), ..., E ^* (2) (* indicates a complex conjugate) is frequency-synthesized by inverse Fourier transform to generate a time domain transmitted signal.

Is generated (S6). When the frequency analysis is performed by the method exemplified in the description of the frequency conversion unit 110, at the time of synthesis, the re-synthesis results from two adjacent frames are added,

Calculate according to X _n (k) is a frequency analysis value calculated from signals x (n−L + 1) to x (n), and X _{nL / 2} (k) is x (n−3L / 2 + 1) to x ( n−L / 2) is a frequency analysis value calculated with a signal, and X _{n + L / 2} (k) is a frequency calculated with a signal from x (n−L / 2 + 1) to x (n + L / 2). Analytical value.

While the sound pickup signal and the reception signal are continuously input to the echo canceller 100, the mth adaptive filter 131 of the kth composite adaptive filter unit 130 corresponding to the frequency component X (k) is a difference signal generation unit. The filter coefficient H _{k, m} is sequentially updated using the difference signal E (k) generated at 150, the frequency component X (k) of the received signal, and the multiplication coefficient β _{k, m} (S7). When the frame number is t and the filter coefficient before update is H _{k, m} (t), the updated filter coefficient H _{k, m} (t + 1) can be obtained by the following equation.

Where * represents the complex conjugate and P _t (k) is the smoothed signal power.

It is. Further, μ is a step size for controlling the update amount (0 <μ <2), and α is a smoothing constant (0 <α <1).

In the above description, the absolute value | X (k) | is applied to the mathematical expression as a value indicating the magnitude of each frequency component X (k), but the value indicating the applied magnitude is not necessarily | X. (k) | is not necessary, and other values such as | X (k) | ² may be used. The same applies to the following embodiments.

  As described above, the echo canceling apparatus 100 and the echo canceling method using the same according to the present invention include “a focus on the characteristics of each band and separate the number of filters prepared for each band” and “a plurality of filters for each band. Is prepared, and the technical feature of changing the coefficient of the linear combination according to the volume of the input is reflected in the specific configuration and processing. Thereby, even if each filter estimates a fixed value, it is possible to accurately estimate an impulse response that fluctuates as a whole. Therefore, even if the acoustic path between the playback means and the sound collection means is non-linear and the fluctuation in volume is large, pseudo echoes close to the echoes that have passed through the acoustic path can be generated, so that the echoes can be erased stably and sufficiently. Therefore, it is possible to prevent deterioration of call quality. In addition, since a suitable number of filters can be prepared for each band, the efficiency of the memory can be improved. Furthermore, since the configuration is simple, it can be realized with a small amount of calculation.

The configuration of the present invention can be applied to the configuration of Reference 1 which is a variation of the frequency domain echo canceller. The configuration of Reference 1 is a configuration that can obtain higher calculation accuracy in the frequency domain. The echo canceling apparatus 200 of the second embodiment is a configuration example in which the configuration of the present invention is applied to the configuration of Reference Document 1. FIG. 4 is a block diagram thereof, and FIG. Indicates.
[Reference 1] ERFerrara, "Fast Implimentation of LMS Adaptive Filters", IEEE Transactions on Acoustics, Speech, and Signal Processing, 1980, Vol.28, No.4, p.474-475

The echo canceling apparatus 200 includes a first frequency conversion unit 210, K (K is the number of frequency components) multiplication coefficient calculation unit 120, K composite adaptive filter units 230, a frequency inverse conversion unit 260, and a block extraction unit 270. A zero addition unit 280, a second frequency conversion unit 240, and an update coefficient calculation unit 290 are provided. Further, the k-th (k = 1, 2,..., K) composite adaptive filter unit 230 has M _k adaptations (M _k is a positive integer corresponding to the value of _k ) as shown in FIG. The filter 231 and the addition part 132 which adds those outputs are provided. Note that the multiplication coefficient calculation unit 120 and the addition unit 132 that are assigned the same reference numerals as those of the echo canceling apparatus 100 of the first embodiment are the same as those of the echo canceling apparatus 100, and thus detailed description thereof is omitted.

  The first frequency converter 210 receives the time domain received signal at time n.

(L is the frame length) is input, frequency analysis is performed by discrete Fourier transform in units of frames, and converted into K frequency components X (1), X (2),..., X (K) and output. (S11). Here, K is 2L. The difference from the first frequency conversion unit 110 of the first embodiment is that the captured frame length is 2L.

Subsequently, as in the first embodiment, the multiplication coefficient calculation unit 120 generates the multiplication coefficient β _{k, m} (S12), the adaptive filter 231 determines the pseudo echo Y ′ (k, m), and the addition unit 132 The pseudo echoes Y ′ (k, m) for each of the M _k adaptive filters are added to generate a total pseudo echo Y ′ (k) based on the frequency component X (k) (S13). The generation method of the pseudo echo Y ′ (k, m) in the adaptive filter 231 is the same as that of the first embodiment, but is given a different code because the filter coefficient update method is different. A method for updating the filter coefficient will be described later.

  The frequency inverse transform unit 260 synthesizes the frequency of the pseudo echoes Y ′ (1), Y ′ (2),..., Y ′ (K) of each frequency component by inverse Fourier transform, and produces a time domain pseudo echo.

Is generated (S14).
The block extraction unit 270 uses the L elements in the latter half of the time domain pseudo echo elements obtained by the frequency inverse transform unit 260.

Is extracted (S15).
A total of K difference signal generation units 250 are provided for each frequency component, and the L elements extracted by the block extraction unit 270 are used as sound collection signals.

Send signal by subtracting from each corresponding element of

Is generated (S16).

While the collected sound signal and the received signal are continuously input to the echo canceling apparatus 200, the m-th adaptive filter 231 of the k-th composite adaptive filter unit 230 corresponding to the frequency component X (k) is subjected to successive filter coefficients H _{k. , m} is updated. In updating, the echo canceling apparatus 200 executes the following processing.

  The zero addition unit 280 adds L zeros to the head of the transmission signal generated by the difference signal generation unit 250, and adds the zero addition transmission signal.

Is generated (S17).

  The second frequency transform unit 240 receives the zero-added transmission signal generated by the zero adder 280 and receives K frequency components E ′ (1), E ′ (2),. .., Converted to E ′ (K) and output (S18). Here, K is 2L. The difference from the second frequency conversion unit 140 of the first embodiment is that the captured frame length is 2L.

  The update coefficient calculation unit 290 generates an update coefficient S (k) from the frequency component E ′ (k) of the zero-added transmission signal and the frequency component X (k) of the reception signal (S19). In particular,

Generate by.

The m-th adaptive filter 231 of the k-th composite adaptive filter unit 230 corresponding to the frequency component X (k) includes the update coefficient S (k), the frequency component X (k) of the received signal, and the multiplication coefficient β _{k, The} filter coefficient H _{k, m} is sequentially updated using _m (S20). When the frame number is t and the filter coefficient before update is H _{k, m} (t), the updated filter coefficient H _{k, m} (t + 1) can be obtained by the following equation.

Where P _t (k) is the smoothed signal power.

It is. Further, μ is a step size for controlling the update amount (0 <μ <2), and α is a smoothing constant (0 <α <1).

  In addition to the effects obtained by the configuration of the first embodiment, the configuration of the second embodiment has higher calculation accuracy in the frequency domain than the configuration of the first embodiment, so that the echo cancellation amount can be further increased.

In the third embodiment, the multiplication coefficient calculation unit 120 in the first and second embodiments is replaced with a multiplication coefficient calculation unit 320. FIG. 7 shows an echo removal apparatus 300 as an example of the configuration. In the first and second embodiments, the multiplication coefficient β _{k, m} is calculated for each frequency component X (k) in the multiplication coefficient calculation unit 120. By evaluating several frequency components collectively, it corresponds to the sound volume. The change of the value becomes gradual, and it is possible to calculate the stable multiplication coefficient without being sensitive to the change of each frequency.

  Specifically, only for the calculation in the multiplication coefficient calculation unit 320, instead of K = L / 2 + 1,

, L _k −L _{k−1, the} number of sets of frequency components bundled is defined as K (L ₁ ,..., L _k is an integer of 2 or more and 0 = L ₀ <L ₁ < ... <L _K = L / 2 + 1). In the multiplier coefficient calculation unit 120 of the first and second embodiments, β _{k, m} is calculated using the absolute value of X (k). In the multiplier coefficient calculation unit 320 of the third embodiment,

Since the multiplication coefficient is obtained in this way, when the multiplication coefficient is applied to each adaptive filter, the multiplication coefficient β _{k, m} related to the bundled frequency components is expressed as L _i <k ≦ L _{i + 1} (0 ≦ i ≦ It becomes the same value in the range of (K-1). For example, as shown in FIG. 7, when X (1) and X (2) are bundled and evaluated in the multiplication coefficient calculation unit 320,

It becomes.

  Note that the memory can be used efficiently by, for example, not bundling in a frequency region where the volume change is large or bundling less, and bundling in a frequency region where the volume change is small.

  FIG. 8 shows a configuration example of the echo canceling apparatus 400 according to the fourth embodiment, and FIG. 10 shows a processing flow example by an echo canceling method using the apparatus.

The echo canceling apparatus 400 includes a first band division unit 410, K multiplication coefficient calculation units 420 (K is the number of band divisions), K composite adaptive filter units 430, second band division units 440, and K difference signals. A generation unit 450 and a band synthesis unit 460 are provided. Further, the k-th (k = 1, 2,..., K) composite adaptive filter unit 430 is adapted to M _k (M _k is a positive integer corresponding to the value of _k ) as shown in FIG. The filter 431 and the addition part 432 which adds those outputs are provided. The main difference from the echo canceling apparatus 100 of the first embodiment is that the first and second frequency conversion units 110 and 140 are replaced with the first and second band dividing units 410 and 440, and the frequency inverse conversion unit 160 is the band synthesizing unit. 460 is replaced by 460. That is, in the echo canceling apparatus 100 of the first embodiment, the input signal is converted into the frequency domain, and the processing is performed in the frequency domain and then reversely converted into the time domain. The difference is that processing is performed in the time domain without conversion to. The other differences are all due to the difference in the processing areas described above, and the functions performed by the multiplication coefficient calculation unit, the composite adaptive filter unit, and the difference signal generation unit are the same as those in the first embodiment.

The first band dividing unit 410 converts the time domain received signal x (n) (where n is an integer representing time) into K band components x _k (k = 1, 2,..., K). , K is an integer of 2 or more) (S21). The method of dividing into band components is, for example, the method disclosed in Reference 2, that is, the k-th band component,

Calculate according to M is an integer of 2 or more as an interval for thinning out the band-limited signal, I is an integer of 2 or more as the number of signals after thinning, r is n / M, and f (i) is 1 ≦ i ≦ M · A low-pass filter defined by I and having a pass band of [−π / K, π / K].
[Reference Document 2] Shoji Makino, “Subband Signal Processing: Real Time Realization”, Journal of the Acoustical Society of Japan, 2000, Vol. 56, No. 12, p.845-851
Multiplication coefficient calculation section 420 calculates M _k multiplication coefficients β _{k, m} (m = 1, 2,..., M _k ) for each band component x _k (r) according to the size of each band component. (S22). Specifically, it is generated as follows.

Band component vector based on band component x _k (r) output from first band dividing section 410

First, the volume P _k (r) of each signal is calculated. Note that L is a filter tap length after thinning M intervals, and is an integer of 2 or more. If the tap lengths before and after thinning are the same, I = L, but in principle, I = I = L need not be. As the volume P _k (r), for example, a vector L2 norm is used.

Calculate according to And this P _k (r) and a preset M _k −1 threshold value v _{k, j} (v _{k, j} is a positive real number and v _{k, j−1} <v _{k, j} (j = 2,... _-, with M k -1)) and, _{M k-number} of multiplication factors _{β k, m (m = 1,2} , ···, generates a M _k).

When M _k is 2 or more, the multiplication coefficient β _{k, m} is specifically determined as follows. Note that i = 1, 2,..., M _k except for the excluded portion.

Further, the threshold value v _{k, j} is, for example, the maximum value that the norm P _k (r) can take is linearly or logarithmically divided into M _k ,

And so on.

Note that v _{k, j} and M _k may be the same value without depending on k, or may be changed depending on k.

The filter coefficient of the mth adaptive filter 431 of the kth composite adaptive filter unit 430 corresponding to the band component x _k (r) is a vector.

The adaptive filter 431 includes a corresponding multiplication coefficient β _{k, m} and a band component vector.

For example, the pseudo echo y ′ _{k, m} (r) by the adaptive filter 431 is obtained by the following equation (S23).

The adaptive filter 431 is

This is updated using the difference signal e _k (r), which will be described later.

  The adding unit 432 performs pseudo echo for each adaptive filter 431.

To generate a total pseudo echo y ′ _k (r) based on the band component x _k (r).

The second band dividing unit 440 uses the same method as the first band dividing unit 410 to convert the time domain sound pickup signal y (n) (n is an integer representing time) into K band components y _k ( r) (k = 1, 2,..., K, K is an integer of 2 or more) (S24).
A total of K difference signal generation units 450 are provided for each band component, and the difference signal e _k (r) is generated by subtracting the pseudo echo y ′ _k (r) from the band component y _k (r) of the collected sound signal. (S25).

The band synthesizing unit 460 performs band synthesis on the difference signals e ₁ (r), e ₂ (r),..., E _K (r) of the respective band components to generate a transmission signal e (n) (S26). ). Band synthesis is, for example,

To do.

While the sound pickup signal and the reception signal are continuously input to the echo canceller 400, the m-th adaptive filter 431 of the k-th composite adaptive filter unit 430 corresponding to the band component x _k (r) generates a difference signal. Difference signal e _k (r) generated by unit 450 and

It can ask for.
Here, μ is a step size (0 <μ <2) for controlling the update amount, and δ is a small positive constant.

"Experimental result"
In order to explain why the present invention is effective, experimental results are shown below.
First, in order to investigate the influence of nonlinear distortion on the impulse response, the impulse response was measured using a plurality of M series having different amplitudes. FIG. 11 shows five impulse responses measured using the M series. The impulse responses are expressed as impulse responses 1 to 5 in order from the one measured with the M series having a small amplitude. It can be seen that each impulse response has the same general shape but the exact value is different.

  Enlarged portions A and B surrounded by dotted lines are shown in FIGS. 12 and 13, respectively. In FIG. 12, a plot near 950 Hz is displayed. In this figure, the response value decreases as the amplitude of the input M-sequence increases. In FIG. 13, a plot near 2300 Hz is displayed. In this figure, the response value increases as the amplitude of the input M-sequence increases.

  Thus, it can be seen that the frequency characteristics differ depending on the magnitude of the input signal, and the behavior differs for each band.

  Next, a performance comparison was made between the conventional method and the method of the present invention based on the average echo canceling amount. First, an audio signal was reproduced and recorded in an actual environment, and Example 2 was implemented as a program on a PC. Next, the amount of echo cancellation obtained by averaging the recorded audio signals in the time direction was calculated. Note that the number K of divisions in the frequency direction is 8, and L = 1024. The results are shown in FIG. In the case of a single filter (conventional method), the amount of echo cancellation was 13.495 dB, whereas when two or four filters were prepared in the entire band, they were 14.227 dB and 14.257 dB, respectively.

  On the other hand, an echo cancellation amount of 14.260 dB could be obtained by changing the number of filters between 2 and 4 for each band. The threshold value for filter switching at this time is the same value as when there are four filters at all frequencies (for example, in the frequency where two filters are used, only one of the three threshold values at the frequency where four filters are used). Was used). The amount of memory required for the filter is 100 when the number of filters is 4 at all frequencies, 50 when the number of filters is 2, and the number of filters is changed between 2 and 4 Was 75. Therefore, it is possible to reduce the memory usage by optimizing the number for each band, rather than preparing a large number of filters (in this case, 4) for all bands, and canceling equal or more echoes. It can be seen that the quantity can be obtained.

  When the echo canceling method is realized by a computer, the processing functions performed by the assignment control unit are described by a program. By executing this program on the personal computer or portable terminal through the exchange of data between the input means and various storage means and the CPU, the hardware and software cooperate to realize the above processing functions on the computer. The effect of the echo canceling method of the invention is achieved. In this case, at least a part of the processing function may be realized by hardware. Further, the various processes described above are not only executed in time series according to the description, but may be executed in parallel or individually according to the processing capability of the apparatus that executes the processes or as necessary. In addition, it can change suitably in the range which does not deviate from the meaning of this invention.

  The program describing the processing contents can be recorded on a computer-readable recording medium. The computer-readable recording medium may be any medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, or a semiconductor memory. Specifically, for example, the magnetic recording device may be a hard disk device or a flexible Discs, magnetic tapes, etc. as optical disks, DVD (Digital Versatile Disc), DVD-RAM (Random Access Memory), CD-ROM (Compact Disc Read Only Memory), CD-R (Recordable) / RW (ReWritable), etc. As the magneto-optical recording medium, MO (Magneto-Optical disc) or the like can be used, and as the semiconductor memory, EEP-ROM (Electronically Erasable and Programmable-Read Only Memory) or the like can be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  As an execution form different from the above-described embodiment, the computer may read the program directly from the portable recording medium and execute processing according to the program. Each time is transferred, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

Claims (8)

  The received signal is divided into a plurality of components by a predetermined method, and a pseudo echo is generated for each component using a plurality of adaptive filters each having a multiplication coefficient generated according to the size of each component. An echo canceling method for obtaining a transmission signal from which echoes have been canceled by subtracting from the collected sound signal. The echo cancellation method according to claim 1,
Receive signal in time domain

First frequency conversion for converting (n is an integer representing time, L is a frame length) into K frequency components X (k) (k = 1, 2,..., K, K are integers of 2 or more) Steps,
For each frequency component X (k), M _k (M _k is a positive integer depending on the value of k) multiplication coefficients β _{k, m} (m = 1, 2,..., M _k ) A multiplication coefficient generation step for generating the frequency component according to the magnitude of the frequency component;
In M _k adaptive filters prepared for each frequency component X (k), the filter coefficient H _{k, m} of the adaptive filter, the multiplication coefficient β _{k, m} and the frequency component X (k) are used. Then, pseudo echoes Y ′ (k, 1), Y ′ (k, 2),..., Y ′ (k, M _k ) are obtained for each adaptive filter, and these are added together to obtain a frequency component X (k). A pseudo echo generation step for generating a total pseudo echo Y ′ (k) for
Time domain pickup signal

A second frequency conversion step of converting the frequency into K frequency components Y (k);
A difference signal generating step of subtracting the pseudo echo Y ′ (k) from the frequency component Y (k) to generate a difference signal E (k);
Said difference signal E (1), E (2 ), ···, E (L / 2 + 1), E * (L / 2), E * (L / 2-1), ···, E * (2 ) (* Indicates complex conjugate)

A frequency inverse transform step to convert to
A filter coefficient updating step for updating the filter coefficient H _{k, m} of the adaptive filter using the difference signal E (k), the frequency component X (k), and the multiplication coefficient β _{k, m} ;
Perform echo reverberation method. The echo cancellation method according to claim 1,
Receive signal in time domain

First frequency conversion for converting (n is an integer representing time, L is a frame length) into K frequency components X (k) (k = 1, 2,..., K, K are integers of 2 or more) Steps,
For each frequency component X (k), M _k (M _k is a positive integer depending on the value of k) multiplication coefficients β _{k, m} (m = 1, 2,..., M _k ) A multiplication coefficient generation step for generating the frequency component according to the magnitude of the frequency component;
In M _k adaptive filters prepared for each frequency component X (k), the filter coefficient H _{k, m} of the adaptive filter, the multiplication coefficient β _{k, m} and the frequency component X (k) are used. Then, pseudo echoes Y ′ (k, 1), Y ′ (k, 2),..., Y ′ (k, M _k ) are obtained for each adaptive filter, and these are added together to obtain a frequency component X (k). A pseudo echo generation step for generating a total pseudo echo Y ′ (k) for
Synthesizing the K pseudo echoes Y ′ (1), Y ′ (2),..., Y ′ (K) into a pseudo echo in the time domain

A frequency inverse transform step to convert to
L elements in the latter half of the time domain pseudo echo elements obtained in the frequency inverse transform step.

A block extraction step for extracting
The L elements extracted in the block extraction step are used as a sound collection signal.

Send signal by subtracting from each corresponding element of

A transmission signal generation step for generating
Add L zeros to the head of the above transmission signal, and add zero transmission signal

A zero addition step to generate
A second frequency conversion step of converting the zero-added transmission signal into K frequency components E ′ (k);
An update coefficient calculation step of calculating an update coefficient S (k) from the frequency component X (k) and the frequency component E ′ (k) of the zero-added transmission signal;
A filter coefficient updating step for updating the filter coefficient H _{k, m} of the adaptive filter using the update coefficient S (k), the frequency component X (k), and the multiplication coefficient β _{k, m} ;
Perform echo reverberation method. The echo cancellation method according to any one of claims 1 to 3,
An echo cancellation method, wherein the multiplication coefficient is generated by collectively evaluating the magnitudes of a plurality of frequency components X (k). The echo cancellation method according to claim 1,
A first band dividing step of dividing the received signal x (n) (n is an integer representing time) into K band components x _k (k = 1, 2,..., K, K is an integer of 2 or more). When,
For each band component x _k , M _k (M _k is a positive integer corresponding to the value of k) multiplication coefficients β _{k, m} (m = 1, 2,..., M _k ) A multiplication coefficient generation step that is generated according to the magnitude of
In M _k pieces of adaptive filters respectively prepared for each of the band components x _k, by using the filter coefficient and the multiplication factor beta _{k, m} and the band components x _k of the adaptive filter, respectively echo replica

And a pseudo echo generation step of adding these to generate a pseudo echo y ′ _k ;
A second band dividing step of dividing the collected sound signal y (n) into K band components y _k ;
And a difference signal generating step of generating a difference signal e _k by subtracting the echo replica y _'k from the band components y _k,
A band synthesis step of synthesizing the _K difference signals e ₁ , e ₂ ,..., E _K and converting them into a time domain transmission signal e (n)
A filter coefficient update step of updating the filter coefficient of the adaptive filter using the difference signal e _k , the band component x _k and the multiplication coefficient β _{k, m} ;
Perform echo reverberation method.   The received signal is divided into a plurality of components by a predetermined method, and a pseudo echo is generated for each component using a plurality of adaptive filters each having a multiplication coefficient generated according to the size of each component. An echo canceling device that obtains a transmission signal from which echoes have been canceled by subtracting from the collected sound signal.   The program for making a computer perform the echo cancellation method in any one of Claims 1 thru | or 5.   A computer-readable recording medium having recorded thereon a program for causing a computer to execute the echo canceling method according to claim 1.