JP2006270709A - Echo remover, electronic conference apparatus, and echo removing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To more surely remove an echo component of a sound pickup signal irrespective of the presence of the occurrence of double talk. <P>SOLUTION: An echo component removal part 512 estimates the echo component with an adaptive filter in a time domain from the sound pickup signal and a reference voltage corresponding to a voice to be reproduced and output to remove the echo component from the sound pickup signal. A parameter updating part 513 updates an adaptive filter parameter 501 in response to an update amount μdesignated by a μcalculation part 520. The update amount μof the adaptive filter parameter 501 is designated by the μcalculation part 520 in response to coherence between an error signal yielded by removing the echo component from the sound pickup signal and the reference signal. It is therefore possible to securely control the operation such that the update amount is optimized on the occurrence of the double talk. This improves the quality of the error signal. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an echo removal apparatus that removes an echo component generated by a wraparound of reproduced and outputted sound from a sound collection signal, an electronic conference apparatus equipped with the echo removal apparatus, an echo removal method, and an echo removal program In particular, the present invention relates to an echo removal apparatus, an electronic conference apparatus, an echo removal method, and an echo removal program that can favorably remove an echo component when double talk occurs.

  For example, in a system that communicates sound collected by a microphone in two ways, such as an electronic conference system, the sound signal collected by the other party is output from the speaker on its own side, and the reproduced sound wraps around the microphone on its own side. It is known that the quality of the voice transmitted / received is degraded due to the occurrence of an echo when the sound is picked up by the sound. For this reason, in such a conventional system, an echo canceller that removes an echo component from a collected sound signal by a microphone is generally used.

Conventional echo cancellers are known to estimate echo components with an adaptive filter that processes speech signals as they are in the time domain, and to use adaptive filters that process speech signals by converting them from the time domain to the frequency domain. ing. As an example of the latter, the power spectrum of the collected sound signal and the cross spectrum between this signal and the reproduced signal are obtained, and the coherence between the reproduced signal and the collected sound signal of one channel or more is obtained, and from these, the frequency spectrum is obtained. There has been a method of estimating the ratio of echo components in the collected sound signal, calculating an echo suppression gain from the ratio, and suppressing the echo of the collected sound signal (see, for example, Patent Document 1).
JP 2003-309493 A (paragraph numbers [0017] to [0020], FIG. 1)

  By the way, of the above-described adaptive filter processing methods, the method of processing an audio signal in the time domain is such that the learning speed of the adaptive filter is set in a double-talk state where both the other party and the other party emit voice simultaneously. It is known that the looser can reliably remove the echo component and improve the voice quality. On the other hand, as a method of converting an audio signal into a frequency domain and processing it, a method has been proposed that can improve audio quality after removing an echo component including a double talk state by using a scale called coherence. However, since the method of processing in the time domain cannot handle the coherence that is a measure of the frequency domain as it is, it is difficult to accurately detect the occurrence of double talk and improve the voice quality.

  The present invention has been made in view of these points, and an object of the present invention is to provide an echo removal apparatus that can more reliably remove an echo component of a collected sound signal regardless of the occurrence of double talk.

Another object of the present invention is to provide an electronic conference apparatus that can more reliably remove the echo component of the collected sound signal regardless of the occurrence of double talk.
Furthermore, another object of the present invention is to provide an echo removal method that can more reliably remove the echo component of the collected sound signal regardless of the occurrence of double talk.

  Another object of the present invention is to provide an echo removal program that can more reliably remove an echo component of a collected sound signal regardless of whether or not double talk occurs.

  In the present invention, in order to solve the above-described problem, an echo removal apparatus that removes an echo component generated due to the wraparound of the reproduced and output sound from the sound collection signal corresponds to the sound collection signal and the sound to be reproduced and output. An echo component removing means for estimating the echo component from a reference signal by a time domain adaptive filter and removing the echo component from the collected sound signal, a parameter updating means for updating a parameter of the adaptive filter, Echo removal comprising: an update amount specifying means for specifying an update amount of the parameter by the parameter update means in accordance with the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal An apparatus is provided.

  Here, the echo component removing means estimates the echo component from the collected sound signal and the reference signal corresponding to the sound to be reproduced and output by the time domain adaptive filter, and removes the echo component from the collected sound signal. The parameter update unit updates the parameter of the adaptive filter in accordance with the update amount designated by the update amount designation unit. The update amount of the parameter is specified by the update amount specifying means according to the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal. Since the coherence is usually low when double talk occurs, it is possible to reliably control the update amount to be optimized when double talk occurs.

  According to the echo removing apparatus of the present invention, the echo component to be removed is estimated by the time domain adaptive filter, so that the delay caused by the processing is reduced compared to the processing in the frequency domain, and the echo component estimating processing is followed. Can increase the sex. In addition to this, by changing the parameter of the adaptive filter according to the coherence between the error signal and the reference signal, it is possible to reliably control the update amount when the double talk occurs. Therefore, the echo component of the collected sound signal can be reliably removed regardless of whether or not double talk occurs.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings, taking as an example the case where the present invention is applied to a terminal device of an electronic conference system.
FIG. 1 is a diagram illustrating a configuration example of an electronic conference system according to an embodiment.

  The electronic conference system according to this embodiment has a configuration in which electronic conference terminals 10 and 20 are connected to a network 30 as shown in FIG. The electronic conference terminals 10 and 20 are terminals for conducting an electronic conference between remote conference rooms, and can transmit and receive image signals and audio signals through the network 30.

  Here, a schematic configuration of the electronic conference terminal 10 will be described as an example. The electronic conference terminal 10 includes a network interface (I / F) 11 that transmits and receives image / sound data through a network 30, an image CODEC (COder / DECoder) 12 that encodes / decodes an image signal, and a display image signal. And an image I / F 13 that performs input / output processing of a captured image signal, an audio CODEC 14 that encodes / decodes audio data, an echo canceller 15 that removes an echo component from the collected sound signal, and an output audio signal or collected sound And an audio I / F 16 for performing signal input / output processing. A camera 13a, a monitor 13b, a microphone 16a, and a speaker 16b are connected to the outside of the electronic conference terminal 10.

  In the electronic conference terminal 10, an image signal captured by the camera 13 a is converted into a digital signal by the image I / F 13, and is encoded by the image CODEC 12 using a predetermined encoding method. The audio signal collected by the microphone 16a is converted into a digital signal by the audio I / F 16, the echo component is removed by the echo canceller 15, and then encoded by the audio CODEC 14 using a predetermined encoding method. The encoded image and audio data is multiplexed into a packet by the network I / F 11 and sent out on the network 30.

  Also, the image and audio data received through the network 30 are separated by the network I / F 11 and input to the image CODEC 12 and the audio CODEC 14, respectively. The separated image data is decoded by the image CODEC 12, converted into a display image signal by the image I / F 13, and output to the monitor 13b, whereby the image is reproduced and displayed. The audio data separated by the network I / F 11 is decoded by the audio CODEC 14, converted into an analog signal by the audio I / F 16, and output to the speaker 16b, thereby reproducing and outputting the audio. The audio signal decoded by the audio CODEC 14 is also supplied to the echo canceller 15 and is used as a reference signal for echo cancellation processing.

  In such an electronic conference system, for example, the sound collected at the electronic conference terminal 10 may be reproduced on the electronic conference terminal 20 side, and the reproduced sound may wrap around and be collected. The same audio wrap may occur on the electronic conference terminal 10 side. Therefore, the electronic conference terminal 10 uses the echo canceller 15 to remove the echo component generated in such a case, thereby improving the quality of the audio transmitted and received.

  Note that the function of the echo canceller 15 is not limited to the electronic conference terminal 10, and may be incorporated in the microphone 16a, for example. In this case, for example, the microphone 16a receives a reference signal (audio signal from the other party's electronic conference terminal 20) from the electronic conference terminal 10 through a serial communication I / F or the like in addition to outputting a collected sound signal.

FIG. 2 is a diagram illustrating an internal configuration example of the echo canceller 15.
As shown in FIG. 2, the echo canceller 15 includes an echo cancellation processing unit 510 that performs processing for removing an echo component by an adaptive filter, and a μ calculation unit 520 that calculates an update amount μ of the parameter of the adaptive filter. The echo cancellation processing unit 510 includes a reference signal buffer 511, an echo component removal unit 512, a parameter update unit 513, a parameter storage unit 514, and a list storage unit 515. The μ calculator 520 includes a coherence calculator 521 and a coherence / μ converter 522.

  The reference signal buffer 511 temporarily stores the reference signal received from the electronic conference terminal 20 on the other side through the network 30 and decoded by the voice CODEC 14, and the echo component removal unit 512, the parameter update unit 513, and the list storage unit 515. Output to.

  The echo component removal unit 512 receives the sound collection signal from the microphone 16a through the sound I / F 16, removes the echo component from the sound collection signal, and outputs it as an error signal to the sound CODEC 14, the parameter update unit 513, and the list storage unit 515. To do. The echo component removal unit 512 uses an adaptive filter parameter (hereinafter simply referred to as a parameter) 501 stored in the parameter storage unit 514, and uses the input sound pickup signal and the reference signal from the reference signal buffer 511, An echo component is estimated by an adaptive filter processed in the time domain.

  The parameter update unit 513 is a block that performs learning processing of the parameter 501 of the adaptive filter. The parameter update unit 513 receives the designation of the update amount μ of the parameter 501 from the μ calculation unit 520, and is used for estimation of echo components according to the update amount μ. The filter parameter 501 is updated. The parameter storage unit 514 stores the parameter 501 updated by the parameter update unit 513 and outputs the parameter 501 to the echo component removal unit 512.

  The list storage unit 515 stores the error signal from the echo component removal unit 512 and the reference signal from the reference signal buffer 511 in the error signal list 502 and the reference signal list 503 that are sequentially accumulated at the same constant time, respectively. Here, the reference signal from the audio CODEC 14 and the collected sound signal from the audio I / F 16 have the same signal output timing to the echo canceller 15, and the error signal lists 502 and 503 in the list storage unit 515 include Audio data of the same number of samples is always accumulated.

  On the other hand, the coherence calculation unit 521 of the μ calculation unit 520 calculates a coherence value between the error signal and the reference signal based on the error signal list 502 and the reference signal list 503 stored in the list storage unit 515. The coherence / μ conversion unit 522 converts the coherence value calculated by the coherence calculation unit 521 into the update amount μ, and designates the update amount μ to the parameter update unit 513.

Next, processing by the echo cancellation processing unit 510 will be described.
The echo component removal processing in the echo cancellation processing unit 510 estimates the echo component using an adaptive filter that processes the audio signal in the time domain, and the processing procedure itself is the same as that used in the past. it can. The echo component in the n-th echo cancellation process can be estimated by the following equation (1).

Here, y (t) is the estimated value of the echo component at time t, w is the adaptive filter parameter 501, x is the time domain data of the reference signal, and k is the number of elements of the vector. Further, w (n) is a vector of parameters 501 arranged in the order of {w ₀ , w ₁ , w ₂ ,..., W _k−1 }, and x (n) is {x _t , x _{t− 1} , x _t−2 ,..., X _{t− (k−1)} } represents a vector of time domain data of reference signals, and w (n) x (n) An inner product of vectors w (n) and x (n) is shown.

  The echo component removing unit 512 estimates the echo component according to the above equation (1) based on the parameter 501 and the reference signal in the parameter storage unit 514, subtracts the component from the collected sound, and outputs an error signal. .

  On the other hand, the parameter update unit 513 updates the parameter 501 of the adaptive filter using a predetermined adaptive algorithm. When the projection method is used as an example of the adaptive algorithm, the n-th update of the parameter 501 is calculated by the following equation (2). Equation (3) is a determinant for obtaining a1 and a2 in Equation (2), and e indicates time domain data of the error signal.

  In the above equation (2), when the update amount μ is increased, the learning speed (update speed) of the parameter 501 is increased. Here, when double talk is not occurring, the correlation between the echo component in the collected sound signal and the reference signal is high. Therefore, increasing the amount of update μ to increase the learning speed ensures that the echo component is The quality of output sound can be improved by removing. However, in the state where double talk has occurred, the sound quality is better if the learning rate is decreased by decreasing the update amount μ.

  In view of this, the echo canceller 15 obtains the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal, and adjusts the parameter update amount μ according to the value to thereby determine whether or not double talk occurs. Improve the quality of the error signal regardless of. When the coherence between the error signal and the reference signal is high, the correlation between the echo component to be removed and the collected sound signal is high. Therefore, the echo component can be more reliably removed by increasing the update amount μ. Conversely, when the coherence is low, there is a high possibility that the parameter 501 has already converged or double talk has occurred, and the quality of the error signal can be improved by reducing the update amount μ.

FIG. 3 is a flowchart showing the flow of processing of the echo cancellation processing unit 510.
[Step S101] When the echo cancellation processing unit 510 receives an audio signal, the following processing is executed as interrupt processing.

[Step S102] The echo component removal unit 512 receives the collected sound signal and receives the reference signal via the reference signal buffer 511.
[Step S103] The echo component removal unit 512 reads the parameter 501 of the adaptive filter from the parameter storage unit 514, and estimates the echo component using the equation (1) based on the parameter 501.

[Step S104] The echo component removal unit 512 subtracts the estimated echo component from the collected sound signal and outputs an error signal.
[Step S105] The error signal output from the echo component removal unit 512 is output to the parameter update unit 513 and the audio CODEC 14, and is also supplied to the list storage unit 515. The list storage unit 515 converts the error signal into the error signal list. Store in 502. The list storage unit 515 stores the latest reference signal stored in the reference signal buffer 511 in the reference signal list 503.

  [Step S106] The parameter update unit 513 uses the formula (2) to calculate the parameter based on the update amount μ from the μ calculation unit 520, the error signal, the reference signal, and the parameter 501 stored in the parameter storage unit 514. The update value 501 is calculated, and the data stored in the parameter storage unit 514 is updated.

  On the other hand, in the processing by the μ calculator 520, the echo cancellation processor 510 receives the audio signal, removes the echo component from the collected sound signal, updates the adaptive filter parameter 501, and then receives the next audio signal. It is executed using the remaining time until. The reason for this is that the processing of the echo cancellation processing unit 510 must be executed every time an audio signal is received. Therefore, when the process of the μ calculation unit 520 is not completed before the next audio signal is received, the interrupt process by the echo cancellation processing unit 510 is executed when the audio signal is received, and the process of the μ calculation unit 520 is performed after the process is completed. Resumed.

Next, the process of the μ calculator 520 will be described in detail. First, examples of the error signal list 502 and the reference signal list 503 used for the μ calculation process will be described.
FIG. 4 is a diagram illustrating a configuration example of the list storage unit 515.

  As shown in FIG. 4, the list storage unit 515 includes error signal lists 502a and 502b and reference signal lists 503a and 503b each divided into two storage areas, and a list management unit 516 for managing these storage areas. It is equipped with. In the error signal lists 502a and 502b, error signals from the echo component removal unit 512 are sequentially accumulated so that when one of them is satisfied, the other is satisfied. Similarly, the reference signals are sequentially accumulated so as to alternately fill the reference signal lists 503a and 503b.

  These error signal lists 502a and 502b and reference signal lists 503a and 503b all store audio data having the same number of samples, and the number of samples is determined by DFT (Discrete Fourier) by a μ calculation unit 520 described later. (Transform) points.

  The list management unit 516 holds a list selection flag FL1 and a μ calculation permission flag FL2. The list selection flag FL1 is a flag indicating which of the error signal lists 502a and 502b the error signal from the echo component removal unit 512 is currently input. For example, “1” is set when data is being stored in the error signal list 502a, and is inverted to “0” when the list is filled and data storage in the error signal list 502b is started. Further, in this case, “1” indicates that data is being stored in the reference signal list 503a, and “0” indicates that data is being stored in the reference signal list 503b.

  The μ calculation permission flag FL2 is set to “1” when new audio data having a number of samples sufficient to satisfy one of the error signal lists 502a and 502b (or the reference signal lists 503a and 503b) is accumulated. Is done. When the audio data is read by the coherence calculation unit 521 of the μ calculation unit 520, the sound data is returned to “0”. Here, the calculation process of the update amount μ by the μ calculation unit 520 is a time during which any of the error signal list 502a and the reference signal list 503a newly started to be accumulated or the error signal list 502b and the reference signal list 503b is satisfied. Be completed within.

  The coherence calculation unit 521 of the μ calculation unit 520 reads either the error signal list 502a and the reference signal list 503a or the error signal list 502b and the reference signal list 503b in order to calculate the coherence. It is possible to determine which list should be read by referring to FL1. That is, the audio data may be read from the other list different from the list being selected indicated by the list selection flag FL1. Further, by referring to the μ calculation permission flag FL2, it can be determined whether or not new audio data can be read.

  FIG. 5 is a flowchart showing a processing flow of the μ calculator 520. When the next audio signal is received by the echo cancellation processing unit 510 during the execution of the series of processing of FIG. 5, the processing of the echo cancellation processing unit 510 is interrupted and executed after completion of the execution of FIG. Subsequent processing is executed.

  [Step S201] The coherence calculation unit 521 refers to the μ calculation permission flag FL2 of the list management unit 516, and executes the process of step S202 when the value becomes “1”.

  [Step S202] Based on the list selection flag FL1, the coherence calculation unit 521 reads audio data from the error signal list 502a or 502b in which audio data of a predetermined number of samples is newly accumulated and the reference signal list 503a or 503b, respectively.

[Step S203] The coherence calculation unit 521 converts the error signal and reference signal read from the list into values in the frequency domain by DFT, and calculates coherence. Specifically, based on the result of DFT, for each component of frequency f, each power spectrum W _xx (f) and W _yy (f) of the error signal and the reference signal and the cross spectrum W _xy (f) of each signal And calculate. Then, the coherence C (f) corresponding to the frequency f is calculated using the following equation (4).

In this equation (4), the average value of the spectrum of the error signal and reference signal of the past M frames (M is a natural number) including the latest W _xx (f), W _yy (f), and W _xy (f) is used. ing. The smaller the number M, the quicker the calculation can be made in response to the change in the sound collection state, but the calculated coherence value becomes less stable. On the contrary, if the number M is increased, the stability of coherence is improved, but the reaction becomes slow. Therefore, it is desirable to determine the number M in consideration of the balance between the reaction rate and the stability of coherence.

  In step S203, the spectrum of the error signal and the reference signal may be calculated by various Fourier transforms such as FFT (Fast Fourier Transform) instead of DFT.

  [Step S204] The coherence calculator 521 calculates an average coherence value C_avg based on the coherence value. The average coherence value C_avg is calculated by adding all the coherence C (f) for each frequency f calculated in step S203 and dividing the added value by the number of frequency components calculated by DFT.

  [Step S205] The coherence / μ conversion unit 522 converts the calculated average coherence value C_avg into a value of the update amount μ of the parameter 501. As described above, the error signal quality can be improved by setting the update amount μ to be higher as the correlation between the error signal and the reference signal is higher, that is, as the average coherence value C_avg is higher.

  In the coherence / μ conversion unit 522, for example, the update amount μ may be calculated by multiplying the average coherence value C_avg by a constant. However, as a whole, the convergence speed of the parameter 501 is slow, and an error signal is generated from the start of processing. It takes time until the parameter 501 converges so that the quality of the image becomes good. For this reason, for example, the conversion is performed using a conversion graph as shown in FIG.

  [Step S206] The coherence / μ conversion unit 522 sets the converted update amount μ in the parameter update unit 513. Thereafter, the μ calculation unit 520 repeats the process of FIG. 5 until the process is terminated in accordance with, for example, a user operation input.

FIG. 6 is a diagram illustrating an example of a conversion graph used in the coherence / μ conversion process.
In this conversion graph, conversion is performed so that the minimum value of the update amount μ is larger than 0, so that the convergence time of the parameter 501 of the adaptive filter is increased, and a sound quality improvement effect can be obtained in a shorter period. In FIG. 6, as an example, the update amount μ is set to the minimum value c1 when the average coherence value C_avg is 0 to a1, the update amount μ is increased at a constant rate when the average coherence value C_avg is a1 to b1, and the average When the coherence value C_avg is b1-1, the update amount μ is set to the maximum value d1.

  As described above, the echo canceller 15 according to the present embodiment obtains coherence based on each list in which the most recent error signal and reference signal are accumulated, and based on the coherence, the correlation between the error signal and the reference signal is high. By changing the update amount μ of the adaptive filter parameter 501 accordingly, the echo component can be more reliably removed from the collected sound signal regardless of the occurrence of double talk. In addition, for the estimation of the echo component, an adaptive filter that processes the audio signal in the time domain is used. Further, the processing of the μ calculation unit 520 that processes in the frequency domain is performed between the estimation processing intervals between the echo component estimation processes. By carrying out with the above period, it can process efficiently.

[Other configuration examples of list storage unit]
FIG. 7 is a diagram illustrating another configuration example of the list storage unit.
The list storage unit 515a shown in FIG. 7 is provided in place of the list storage unit 515 shown in FIG. 4, and includes ring buffers 517 and 518 for storing the error signal list 502 and the reference signal list 503, respectively. ing. Each of the ring buffers 517 and 518 has a capacity twice as large as the number of samples of audio data necessary for performing the DFT by the coherence calculation unit 521. The ring buffers 517 and 518 are all initialized with data “0” when the operation of the echo canceller 15 is started.

  The list storage unit 515a further includes a list management unit 518. The list management unit 518 includes a ring counter 519 that is a counter indicating a readable position in the ring buffers 517 and 518, and μ The calculation permission flag FL3 is held. The ring counter 519 updates the count value every time the storage area of each of the ring buffers 517 and 518 is filled with audio data by ¼. When the count value is updated, the μ calculation permission flag FL3 is set to “1”, and when the audio data is read from the corresponding reading position to the coherence calculation unit 521, the μ calculation permission flag FL3 is inverted to “0”.

  Here, it is assumed that the calculation process of the update amount μ by the μ calculation unit 520 can be completed within a time when the voice data is input by the capacity of ¼ of each of the ring buffers 517 and 518. At this time, when the μ calculation permission flag FL3 becomes “1”, the coherence calculation unit 521 starts from the corresponding position of the ring buffers 517 and 518 based on the count value of the ring counter 519, and ½ capacity of each buffer. Minute error signal and reference signal are read, and coherence calculation is performed.

  For example, if the storage area of the ring buffer 517 has four equal areas 517a to 517d in the data accumulation order, when the error signal is filled up to the area 517b, the error signals stored in the areas 517a and 517b are coherent. The data is read by the calculation unit 521. Next, when the error signal is satisfied up to the area 517c, the error signals in the areas 517b and 517c are read.

  By such an operation, for example, compared with the list storage unit 515 shown in FIG. 4, the output period of the update amount μ can be halved, and the adaptive filter parameter 501 can be updated more accurately. Further, without increasing the buffer capacity for storing the error signal list 502 and the reference signal list 503, the number of audio data samples used for coherence calculation remains unchanged, and the calculation is always performed using the latest audio data. And the quality of the error signal can be further improved.

  In the example of FIG. 7, 1/4 of the storage areas of the ring buffers 517 and 518 are read. However, the present invention is not limited to this, and the timing for calculating the coherence can be freely changed.

[Other processing examples of coherence calculation]
When the average coherence value C_avg is obtained by the coherence calculation unit 521, the stability of the average coherence value C_avg can be increased by using more (that is, a long period) coherence C (f). However, on the other hand, the processing load increases and the memory capacity required inside increases.

  Therefore, the average coherence value C_avg may be obtained by using only a part of the frequency component obtained by DFT. For example, the frequency components obtained by DFT are thinned out one by one, the coherence C (f) is obtained in the frequency components after the thinning, and the average coherence value C_avg is calculated. As a result, even when the memory capacity and processing capability are the same, calculation can be performed based on audio data for a longer period. In addition, for example, a predetermined number of frequency components used for calculation may be selected in ascending order of frequency, or may be selected at random.

  Furthermore, weighting may be performed for each frequency component when calculating the average coherence value C_avg. For example, the weighting ratio is set according to the characteristics of the microphone, speaker, CODEC, and the like. Alternatively, the weighting ratio may be set and changed according to the size of the room in which the electronic conference apparatus is installed, the number of microphones and speakers, and the like. As a result, it is possible to change the value taken by the update amount μ according to the situation, for example, so as not to take into account frequency components including stationary noise so much, and to suppress degradation of the quality of the error signal.

On the other hand, when the number of samples (sampling period) of the audio data used for coherence calculation is reduced because the internal memory amount of the μ calculation unit 520 and the storage capacity of the list storage unit 515 are reduced, the average coherence value C_avg is calculated. The time constant D (where 0 <D <1) may be used to improve the stability of the calculation result. In this case, the average coherence value C_avg (N) calculated at the Nth turn is calculated by the following equation (5), for example. In addition, in the Nth turn, after adding all the coherence C (f) calculated for each frequency component, a value obtained by dividing by the number of the frequency components is defined as C_small_ave (N).
C_avg (N) = {D × C_avg (N−1)} + {(1−D) × C_small_ave (N)} (5)
By performing such calculation, the average coherence value C_avg is stabilized even when the number of audio data samples is small, and the quality of the error signal is increased. If the time constant D is increased, the stability of the calculation result is increased, but the followability to changes in the components of the audio signal is lowered. Therefore, it is desirable to set the time constant D in consideration of such balance.

[Other processing examples of average coherence / μ conversion]
The conversion graph used when the average coherence value C_avg is converted into the update amount μ in the coherence / μ conversion unit 522 is not limited to that shown in FIG. 6, and various patterns can be used as follows.

8 to 10 are diagrams illustrating other examples of conversion graphs used in the coherence / μ conversion processing.
In FIG. 8A, as the average coherence value C_avg increases from 0 to 1, the update amount μ is increased from the minimum value a2 to the maximum value b2 at a constant rate. In FIG. 8B, the conversion processing is performed by setting the update amount μ as the minimum value b3 when the average coherence value C_avg is smaller than the threshold value a3, and the update amount μ as the maximum value c3 when the average coherence value is equal to or greater than the threshold value a3. Is simplified.

  In FIG. 9A, the rate of increase of the update amount μ (however, 0 or more) is changed with the case where the average coherence value C_avg becomes a4, b4, c4, d4 so that the update amount can be controlled more finely. It has become. In FIG. 9B, as the average coherence value C_avg increases, the update amount μ increases in a smooth curve from the minimum value a5 to the maximum value b5.

  Furthermore, as shown in FIG. 10, two curves 522a and 522b indicating the correspondence between the average coherence value C_avg and the update amount μ may be prepared and used depending on the situation. For example, when the history of the average coherence value C_avg is held and the average coherence value C_avg calculated in a recent fixed period tends to be high (for example, when the average value is equal to or greater than a threshold value), an error occurs. Since the degree of relevance between the signal and the reference signal is high and it is considered that many echo components remain in the error signal, conversion is performed using the curve 522a to increase the convergence speed of the parameter 501 of the adaptive filter. Conversely, when the recent average coherence value C_avg tends to be low, the conversion is performed using the curve 522b to stabilize the parameter 501 and prevent deterioration in sound quality. Also, three or more conversion graphs to be used may be prepared and used according to more detailed conditions.

  Further, when the average coherence value C_avg becomes lower than a certain threshold value, the update of the adaptive filter parameter 501 may be stopped. In order to stop the update, a method of setting the update amount μ to 0, the coherence / μ conversion unit 522 outputs an update stop signal, and when the parameter update unit 513 receives this update stop signal, the operation is stopped. Techniques can be applied. Further, at the time of stopping the update or at the time of returning from the update stop, for example, the process may be executed based on the conditions used in FIG. 11 or FIG.

  In each of the above examples, the change amount of the update amount μ accompanying the increase in the average coherence value C_avg is 0 or more. However, in general, when the average coherence value C_avg is too close to 1, there is a possibility that inaccurate detection is performed. For example, when the average coherence value C_avg exceeds a certain threshold value, the update amount μ is decreased. Such a conversion graph may be used.

  FIG. 11 is a flowchart showing another processing example (part 1) for converting the average coherence value C_avg into the update amount μ. The process of FIG. 11 corresponds to the process of step S205 of FIG. 5 executed by the coherence / μ conversion unit 522.

  [Step S301] It is determined whether or not the calculated average coherence value C_avg is less than the threshold value Th1, and if it is less than the threshold value Th1, the process proceeds to step S302. If it is greater than or equal to the threshold value Th1, step S305 is performed. Proceed to

[Step S302] "1" is added to the variable p.
[Step S303] It is determined whether or not the variable p is greater than or equal to a threshold value Th2. If it is greater than or equal to the threshold value Th2, the process proceeds to step S304, and if it is less than the threshold value Th2, the process proceeds to step S307.

[Step S304] An update stop signal for the parameter update unit 513 is set to H level.
[Step S305] The variable p is initialized to “0”.

[Step S306] The update stop signal is set to L level.
[Step S307] Based on a predetermined conversion graph, the average coherence value C_avg is converted into an update amount μ and set in the parameter update unit 513.

  In the above process, when the average coherence value C_avg is continuously smaller than the threshold value Th1 for the number of times indicated by the threshold value Th2, the parameter update unit 513 stops the update process of the parameter 501 (step S301). ~ S302). After that, when the average coherence value C_avg is equal to or greater than the threshold value Th1, the update amount μ converted based on the conversion graph is set in the parameter update unit 513, and the update process is resumed. Therefore, when the average coherence value C_avg tends to be low, the adaptive filter parameter 501 can be stabilized to prevent deterioration in sound quality.

  FIG. 12 is a flowchart showing another processing example (part 2) for converting the average coherence value C_avg into the update amount μ. The processing in FIG. 12 also corresponds to the processing in step S205 in FIG. 5 executed by the coherence / μ conversion unit 522.

  [Step S401] It is determined whether or not the calculated average coherence value C_avg is less than the threshold value Th3. If the calculated average coherence value C_avg is less than the threshold value Th3, the process proceeds to step S402. Proceed to

[Step S402] The variable q is initialized to “0”.
[Step S403] The update stop signal is set to H level.
[Step S404] "1" is added to the variable q.

  [Step S405] It is determined whether or not the variable q is greater than or equal to a threshold value Th4. If it is greater than or equal to the threshold value Th4, the process proceeds to step S406. If it is less than the threshold value Th4, the process proceeds to step S403.

[Step S406] The update stop signal is set to L level.
[Step S407] Based on a predetermined conversion graph, the average coherence value C_avg is converted into an update amount μ and set in the parameter update unit 513.

  In the above process, if the average coherence value C_avg is less than the threshold value Th3 even once, the update process of the parameter update unit 513 is stopped. Thereafter, the update process of the parameter 501 is not resumed until the average coherence value C_avg is continuously equal to or greater than the threshold value Th3 for the number of times indicated by the threshold value Th4. Therefore, when the average coherence value C_avg fluctuates substantially below the threshold value Th3, the adaptive filter parameter 501 can be stabilized so that the change in sound quality does not become unnatural.

  In the above embodiment, the case where the present invention is applied to an electronic conference terminal has been described. However, the present invention is not limited to this, and for example, a microphone that is used in the electronic conference system or a terminal that transmits and receives audio in two directions such as a telephone. On the other hand, an echo canceller to which the present invention is applied can be mounted. Further, the present invention can be applied to both a microphone and a terminal such as an electronic conference terminal for voice transmission / reception.

  Further, the above processing functions can be realized by a computer. In that case, a program describing the processing contents of the functions that the echo canceller described in the above embodiment should have is provided, and the processing functions are realized on the computer by executing the program on the computer. The program describing the processing contents can be recorded on a computer-readable recording medium. Examples of the computer-readable recording medium include a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory. Examples of the magnetic recording device include a hard disk device (HDD), a flexible disk (FD), and a magnetic tape. Examples of the optical disk include a DVD (Digital Versatile Disk), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc Read Only Memory), and a CD-R (Recordable) / RW (ReWritable). Magneto-optical recording media include MO (Magneto-Optical disk).

  When distributing the program, for example, a portable recording medium such as a DVD or a CD-ROM in which the program is recorded is sold. It is also possible to store the program in a storage device of a server computer and transfer the program from the server computer to another computer via a network.

  The computer that executes the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its own storage device and executes processing according to the program. The computer can also read the program directly from the portable recording medium and execute processing according to the program. Further, each time the program is transferred from the server computer, the computer can sequentially execute processing according to the received program.

It is a figure which shows the structural example of the electronic conference system which concerns on embodiment. It is a figure which shows the internal structural example of an echo canceller. It is a flowchart which shows the flow of a process of an echo cancellation process part. It is a figure which shows the structural example of a list memory | storage part. It is a flowchart which shows the flow of a process of (mu) calculation part. It is a figure which shows an example of the conversion graph used by a coherence / micro conversion process. It is a figure which shows the other structural example of a list memory | storage part. It is a figure which shows the other example (the 1) of the conversion graph used by a coherence / micro conversion process. It is a figure which shows the other example (the 2) of the conversion graph used by a coherence / micro conversion process. It is a figure which shows the other example (the 3) of the conversion graph used by a coherence / micro conversion process. It is a flowchart which shows the other process example (the 1) which converts an average coherence value into update amount. It is a flowchart which shows the other process example (the 2) which converts an average coherence value into update amount.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Teleconference terminal, 11 ... Network I / F, 12 ... Image CODEC, 13 ... Image I / F, 13a ... Camera, 13b ... Monitor, 14 ... Voice CODEC, 15 ... Echo canceller , 16... Voice I / F, 16 a... Microphone, 16 b. Speaker, 20... Teleconference terminal, 30... Network 501. List 510 ... Echo cancel processing unit 511 ... Reference signal buffer 512 ... Echo component removal unit 513 ... Parameter update unit 514 ... Parameter storage unit 515 ... List storage unit 520 ... μ Calculation unit 521 ... Coherence calculation unit 522 ... Coherence / μ conversion unit

Claims (11)

In an echo removing apparatus that removes an echo component generated by a wraparound of a reproduced and output sound from a sound collection signal,
Echo component removal means for estimating the echo component from a reference signal corresponding to the collected sound signal and the sound to be reproduced and output by a time domain adaptive filter, and removing the echo component from the collected sound signal;
Parameter updating means for updating parameters of the adaptive filter;
An update amount specifying means for specifying an update amount of the parameter by the parameter update means according to the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal;
An echo removing apparatus comprising: The update amount designation means includes:
An average value for calculating an average coherence value by adding the coherence calculated based on the collected sound signal and the reference signal input for a certain period for each frequency component and dividing the added value by the number of the frequency components. A calculation means;
Data conversion means for converting the average coherence value into an update amount of the parameter;
The echo removing apparatus according to claim 1, further comprising:   The data conversion unit converts the average coherence value into an update amount of the parameter so that a change amount of the update amount of the parameter becomes 0 or more with respect to an increase in the average coherence value. Item 3. The echo canceller according to Item 2.   The echo removal according to claim 2, wherein the data conversion means changes a conversion pattern when converting the average coherence value into the update amount according to a tendency of the average coherence value in a previous predetermined period. apparatus.   The average value calculating means calculates the average coherence value using coherence corresponding to a part of frequency components among coherence calculated based on the collected sound signal and the reference signal. 2. The echo canceller according to 2.   The average value calculation unit calculates the average coherence value using a value obtained by weighting the coherence calculated based on the collected sound signal and the reference signal for each frequency component. Echo removal device. The average value calculating means outputs a value obtained by adding the current average coherence value and the average coherence value calculated in the previous period at a predetermined ratio based on a time constant,
The echo removal apparatus according to claim 2, wherein the data conversion unit converts the addition value into an update amount of the parameter.   3. The parameter updating unit temporarily stops updating the parameter when the average coherence value calculated by the average value calculating unit is lower than a predetermined threshold value. Echo removal device. In an electronic conference device that realizes a conference with a remote place by transmitting and receiving audio signals and video signals to each other through a network,
Echo component removal means for estimating an echo component from a sound collection signal and a reference signal corresponding to the sound to be reproduced and output by a time domain adaptive filter, and removing the echo component from the sound collection signal;
Parameter updating means for updating parameters of the adaptive filter;
An update amount specifying means for specifying an update amount of the parameter by the parameter update means according to the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal;
An electronic conference apparatus comprising an echo removing unit having In an echo removal method for removing an echo component generated by a wraparound of a reproduced and output sound from a sound collection signal,
An echo component removing unit estimating the echo component by a time domain adaptive filter from the collected sound signal and a reference signal corresponding to the sound to be reproduced and output, and removing the echo component from the collected sound signal;
Parameter updating means updating the parameters of the adaptive filter;
An update amount designation means designates an update amount of the parameter by the parameter update means according to the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal;
An echo removal method comprising: In an echo removal program for causing a computer to execute a process of removing an echo component generated by wraparound of a reproduced and output sound into a sound collection unit,
An echo component removing means for estimating the echo component from the collected sound signal and a reference signal corresponding to the sound to be reproduced and output by a time domain adaptive filter, and removing the echo component from the collected sound signal;
Parameter updating means for updating parameters of the adaptive filter;
An update amount specifying means for specifying an update amount of the parameter by the parameter update means according to the coherence between the error signal obtained by removing the echo component from the collected sound signal and the reference signal;
An echo removal program for causing the computer to function as:

Images