JP2009302885A - Sound pickup apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To prevent the deterioration of adaptive filter precision to reduce an acoustic echo without a complicated operation in a sound pickup apparatus of a fixed point DSP system. <P>SOLUTION: A scale attenuation part 15 attenuates an error signal, a scale amplification part 14 performs amplification by an inverse number of attenuation in the scale attenuation part to output of an adaptive filter part 11 to control magnitude of an adaptive filter factor. An attenuation amount is determined by a scale adjustment part 13, and controlled so that the adaptive filter factor does not exceed the maximum value and becomes within a range in which the precision is sufficiently maintained. As information when the attenuation amount is determined, one or more of the maximum absolute value of the adaptive filter factor, the absolute value summation of the adaptive filter factor, and an adaptive filter output value are used. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a sound pickup apparatus, and more particularly to a sound pickup apparatus that employs a fixed-point digital signal processor (DSP) as an echo canceller.

  Plays audio signals from remote locations (hereinafter referred to as “far-end signals”) and collects audio signals (hereinafter referred to as “near-end signals”) at the point where the sound pickup device is installed. Thus, acoustic echo becomes a problem in a sound pickup apparatus that must simultaneously reproduce the far-end signal and pick up the near-end signal. The acoustic echo is a signal that is generated because a signal such as a sound reproduced from a reproducing unit represented by a speaker is collected by a sound collecting unit such as a microphone, and is a signal that circulates from the reproducing unit to the sound collecting unit. .

  Conventionally, a number of methods called echo cancellers have been proposed and used as means for canceling acoustic echoes. The acoustic echo is generated by the combination of the reproducing means and the sound collecting means, and the path of the acoustic echo varies depending on the gain in the reproducing means, the gain in the sound collecting means, and the spatial coupling between the reproducing means and the sound collecting means. It varies greatly depending on the shape of the apparatus itself and the positional relationship between the reproducing means and the sound collecting means. Therefore, the echo canceller estimates the path of the acoustic echo, generates a pseudo acoustic echo signal from the far end signal, and cancels the acoustic echo signal and the pseudo acoustic echo signal. To prevent the output of.

  The theory of adaptive signal processing is used for path estimation of the acoustic echo, and an adaptive filter that operates adaptively so as to output a pseudo acoustic echo signal that cancels the acoustic echo signal is used. As a result of the adaptive operation, the coefficients of the adaptive filter are equivalent to the acoustic echo path, and the pseudo acoustic echo can be the same as the acoustic echo.

  As DSPs used for adaptive signal processing in the sound pickup apparatus, floating point type DSPs are avoided and fixed point type DSPs are often used from the viewpoint of price and processing speed. In the fixed-point type DSP, the calculation accuracy is limited, and for example, only 16 bits (= 65536 levels) can be expressed. Actually, the adaptive filter coefficient of the echo canceller is often expressed by 16 bits from the viewpoint of the calculation amount, calculation accuracy, and bit accuracy of the input signal.

  When an echo canceller is realized with such a fixed-point DSP, the position of the decimal point is adjusted in advance (hereinafter referred to as “scaling”) in consideration that the acoustic echo path may exceed 1 (overflow). It is common. That is, the decimal point position is shifted (so as not to overflow). For example, if the adaptive filter coefficient is not expected to exceed 4, the 16 bits are divided into 2 bits and 14 bits, and the former is an integer part and the latter is a decimal part to represent coefficients up to 4. . Such a countermeasure is relatively simple, and can be realized only by shifting the error signal to the right by 2 bits and shifting the output signal to the left by 2 bits in the adaptive operation of the adaptive filter.

  However, whether or not the coefficient of the adaptive filter actually exceeds 1 depends on the environment where the sound pickup device is placed, the gain of the reproduction means, and the gain of the sound collection means. Is not limited. In such a case, the fixed-point DSP has only a 16-bit representation and is not sufficiently accurate, but the above-mentioned 16 bits are divided into 2 bits and 14 bits. If the coefficient does not exceed 1, the effective number of bits is 14 bits (16384 steps), and the coefficient accuracy of the adaptive filter, and hence the accuracy of the pseudo acoustic echo signal, is greatly degraded.

  Furthermore, in order to reduce the amount of computation of adaptive signal processing, an echo canceller that uses a band division filter bank to estimate an echo path in each band has been proposed, but such a band division method was adopted. In the echo canceller, the coefficient maximum absolute value of the adaptive filter differs for each band. In this case, it is necessary to adjust the decimal point position of the adaptive filter in each band, but since the acoustic echo level in each band is an independent relationship, it cannot be determined uniquely, so the maximum that can be handled Generally, the absolute value is set larger. As a result, there is a problem in that the accuracy of the adaptive filter is remarkably deteriorated particularly in a high frequency range and the acoustic echo cannot be sufficiently canceled.

  In response to such a problem, an error signal used for learning in an acoustic echo canceller apparatus (see, for example, Patent Document 1) or an adaptive filter which has been solved by providing an automatic gain adjustment function for a signal from a sound collection unit. Therefore, an echo canceller (see, for example, Patent Document 2) that determines a coefficient for scaling an adaptive filter coefficient and performs control so that the adaptive filter coefficient does not exceed a maximum value is known. An echo canceller (see, for example, Patent Document 3) that adjusts an error signal that is a difference between a cancel signal and an input signal as an echo from an echo path is also known.

JP2002-043985 (3rd page, FIG. 1) JP 04-047720 (2nd page, Fig. 1) JP 06-204917 (2nd page, Fig. 1)

  However, in the technique described in Patent Document 3, the accuracy of the signal itself from the sound collecting means is deteriorated, or the adaptive filter learning is performed by using data having no meaning in accuracy included in the signal from the sound collecting means. There is a problem that it will be done.

  In the technique described in Patent Document 2, in order to adjust the scale of the adaptive filter coefficient itself, it is necessary to multiply all the adaptive filter coefficients every time by the scale adjustment coefficient, which is not realistic.

  Further, in the technique described in Patent Document 3, since the amount of adjustment of the error signal level is constant, in the echo canceller using the band division filter bank, the optimum amount of scaling adjustment differs for each band. There is a problem that the accuracy of the adaptive filter coefficient is not constant.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a fixed-point DSP type sound pickup apparatus that prevents deterioration of adaptive filter accuracy and reduces acoustic echo without complicated operation.

  In the present invention, a scale value is provided for an adaptive filter (for each band in the case of using band dividing means), and the value is adaptively adjusted. The scale value is a coefficient to be multiplied by an error signal used for adaptive filter learning or a shift amount. Further, the adaptive filter output (pseudo echo signal) is multiplied by the inverse of the scale value.

  The scale value is automatically adjusted according to the following three criteria. The first is the maximum absolute value of the adaptive filter coefficient, which is adjusted so that the filter coefficient does not overflow. The second is the value of the sum of the absolute values of the adaptive filter coefficients, and is adjusted so that the output of the filter does not overflow. The third is the value of the convolution output of the adaptive filter, which is adjusted so that the filter output does not overflow. The scale value may be adjusted by combining the above three or by using one of them.

  The sound pickup apparatus of the present invention is a sound pickup apparatus that employs a fixed-point DSP as an echo canceller, and an acoustic echo that circulates from a playback unit that reproduces a far-end signal to a sound collection unit that collects a near-end signal. An adaptive filter unit that generates a pseudo-acoustic echo signal for canceling out the sound, determines a scale value based on an output from the adaptive filter unit, and the acoustic echo passes through the sound collection unit between the acoustic echo signal and the pseudo-acoustic echo signal. The level of the pseudo acoustic echo signal is automatically adjusted by attenuating the error signal, which is a difference signal, by the scale value, and providing a scale adjusting means for amplifying the pseudo acoustic echo signal by the reciprocal of the attenuation amount and using it for cancellation. It is characterized by.

  Specifically, the scale adjustment means determines a scale value based on an output from the adaptive filter unit, and determines an attenuation amount and an amplification degree that is a reciprocal of the attenuation amount based on the scale value, and attenuates an error signal. A scale attenuating unit for attenuating by the amount and a scale amplifying unit for amplifying the pseudo acoustic echo signal output from the adaptive filter unit by the amplification degree.

  The automatic adjustment is performed based on the maximum absolute value of the filter coefficient in the adaptive filter unit so that the filter coefficient does not overflow the value determined by the number of bits of the DSP, and the accuracy of the filter coefficient can be utilized to the maximum in the DSP. You may do it.

  In the automatic adjustment, based on the absolute value sum of the filter coefficients in the adaptive filter unit, the output of the adaptive filter unit does not overflow the value determined by the number of bits of the DSP, and the accuracy of the filter coefficient is maximized in the DSP. You may make it carry out so that it can be used for a limited time.

  In the automatic adjustment, based on the convolution output value in the adaptive filter unit, the output of the adaptive filter unit does not overflow the value determined by the number of bits of the DSP, and the accuracy of the filter coefficient can be utilized to the maximum in the DSP. You may make it perform.

  Further, the sound collecting device of the present invention is provided with band dividing means for dividing the far-end signal and the output signal of the sound collecting section into a signal of a limited frequency band, and the adaptive filter section and the scale adjusting means are divided frequencies. You may make it operate | move corresponding to the signal of a zone | band.

  According to the present invention, an error signal required for learning in the adaptive filter unit is attenuated by a scale value and the output of the adaptive filter unit is amplified by the inverse of the attenuation amount, thereby automatically adjusting the level of the pseudo acoustic echo. Since the adjusting means is provided, it is possible to make the best use of the precision of the fixed-point DSP, and the echo canceling characteristics of the echo canceller can be improved.

  Further, it is not necessary to input the filter parameter scaling parameter or to adjust the gain of the reproducing means, which is necessary for the initial setting of the echo canceller, and the complicated initial setting is reduced.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a sound pickup apparatus of the present invention.

  The far-end signal input terminal 1 inputs a far-end signal, and the reproduction unit 2 represented by a speaker reproduces the far-end signal. A sound collection unit 3 represented by a microphone collects a near-end signal such as voice, and a sound collection signal output terminal 4 outputs a near-end signal. Ideally, this output signal is not affected by the far-end signal but only the near-end signal. However, in an actual apparatus, a part of the signal reproduced by the reproduction unit 2 generates an acoustic echo that wraps around the sound collection unit 3 due to the combination of the reproduction unit 2 and the sound collection unit 3, and is mixed into the near-end signal and collected. It may be output from the sound signal output terminal 4. A signal (near end signal + acoustic echo signal) output from the sound collection unit 3 is referred to as a sound collection signal.

  In order to eliminate the influence of the acoustic echo, an echo canceller comprising an adaptive filter unit 5, a subtracting unit 6 and a scale adjusting means 7 is provided. The echo canceller generates a pseudo acoustic echo and cancels the acoustic echo. The echo canceller involves adaptive signal processing, but a band division method is employed to reduce the amount of computation. That is, a band dividing unit 8 that divides the far-end signal into a signal limited to each frequency band, a band dividing unit 9 that divides the collected sound signal into a signal limited to each frequency band, and an output signal of each frequency band And a band synthesizing unit 10 that outputs the collected sound signals in the entire frequency band. The adaptive filter unit 5, the subtracting unit 6 and the scale adjusting unit 7 operate in correspondence with the frequency bands hidden.

  The band dividing unit 8 divides the far-end signal from the far-end signal input terminal 1 into signals limited to the respective frequency bands, and the band-dividing unit 9 separates the collected sound signal from the near-end signal input terminal 3. Divide into signals limited to frequency band. For the band dividing units 8 and 9, for example, a band dividing filter bank using FFT or a filter bank using SSB modulation is used.

  The signal limited to each frequency band by the band dividing unit 8 is input to the adaptive filter unit 5, and the adaptive filter unit 5 generates a pseudo acoustic echo signal in each frequency band. The subtracting unit 6 obtains a difference signal between the acoustic echo signal included in the near-end signal from the sound collecting unit 3 and the pseudo acoustic echo signal.

  That is, the adaptive filter unit 5 operates so that the difference signal between the sound collection signal and the pseudo acoustic echo signal in each frequency band becomes 0 (silence). However, when there is a near-end signal, such an adaptive operation is stopped and control is performed so as not to erase the near-end signal.

  The algorithm in the adaptive filter unit 5 performs learning by an adaptive algorithm such as a learning identification method or an affine projection algorithm. The adaptive algorithm means in the present invention is not limited.

  By the way, the path of the acoustic echo varies depending on the gain in the reproducing unit 2, the gain in the sound collecting unit 3, and the spatial coupling between the reproducing unit 2 and the sound collecting unit 3. And the position of the sound collection unit 3 vary greatly depending on the environment. Therefore, the adaptive filter coefficient related to the path of the acoustic echo also changes greatly. In such a situation, even when the fixed-point DSP method is adopted, the scale adjustment unit 7 is provided to maintain the coefficient accuracy of the adaptive filter, and hence the accuracy of the pseudo acoustic echo signal.

  The scale adjusting means 7 is a core part of the present invention, and controls the scale of the adaptive filter unit 5 in each band. The scale adjustment unit 7 determines a scale value based on the output from the adaptive filter unit 5, and attenuates an error signal (difference signal between the acoustic echo signal and the pseudo acoustic echo signal) necessary for learning in the adaptive filter unit 5 by the scale value. Let Further, the output of the adaptive filter unit 5 is amplified by the reciprocal of the attenuation amount. That is, the scale value is a coefficient or shift amount to be multiplied by the error signal used for learning in the adaptive filter unit 5.

  As a result, the maximum value of the adaptive filter coefficient can be arbitrarily determined according to the environment. That is, when the adaptive filter coefficient is largely determined, the attenuation amount of the error signal is increased and the amplification degree of the output of the adaptive filter unit 5 is decreased. On the other hand, when the adaptive filter coefficient is determined to be small, the attenuation amount of the error signal is reduced and the amplification degree of the output of the adaptive filter unit 5 is increased. In this way, the level of the pseudo acoustic echo signal output from the adaptive filter unit 5 is automatically adjusted.

  FIG. 2 shows a detailed configuration of the echo canceller for one frequency band, and such a configuration exists in a one-to-one correspondence with the frequency band into which the configuration is divided. FIG. 2 shows an adaptive filter unit 11 for one frequency band, a subtracting unit 12 for one frequency band, and a scale adjusting means 16 for one frequency band. The scale adjustment unit 16 amplifies the output of the scale adjustment unit 13, the adaptive filter unit 11 and outputs it to the subtraction unit 12, and attenuates the output of the subtraction unit 12 to output to the adaptive filter unit 11. And a scale attenuating unit 15.

  The scale adjustment unit 13 takes in the output of the adaptive filter unit 11 and determines the scale value. As information for determining the scale value, any one of (1) the maximum absolute value of the adaptive filter coefficient, (2) the sum of absolute values of all the adaptive filter coefficients, and (3) the convolution output value of the adaptive filter unit 5 is used. .

  It is ideal to acquire each piece of information in audio processing, but from a viewpoint of calculation amount, even if information is acquired at a certain interval (for example, every 10 ms), there is a big problem. Hard to become. However, since the adaptive filter may fail depending on circumstances, it is necessary to provide a margin for adjusting the scale. The margin will be specifically described in an example described later.

  Hereinafter, each method will be described separately for each case. As a condition for all examples, consider the case of a 16-bit fixed point DSP. The adaptive filter length is L_hreg. The scale value is automatically adjusted according to the following three criteria. The first method is adjusted so that the filter coefficient does not overflow, and the second and third methods are adjusted so that the output of the adaptive filter does not overflow. These three methods may be used alone or in any combination.

(1) When using the maximum absolute value of the adaptive filter coefficient The maximum absolute value of the adaptive filter coefficient does not overflow the value determined by the number of bits of the adaptive filter coefficient, and the accuracy of the adaptive filter coefficient is fixed point DSP. This is information for adjusting to a scale that can be used to the maximum. In this method, the scale value is controlled so that the maximum absolute value of the adaptive filter coefficient falls within a certain range. For example, in the case of a 16-bit fixed point DSP, the range that can be expressed by a signed integer is -32768 to 32767. Under this condition, the maximum absolute value of the adaptive filter coefficient needs to be close to 32767 in order for the adaptive filter coefficient not to overflow the value determined by the number of bits of the DSP and to maximize the accuracy.

The maximum absolute value of the adaptive filter coefficient is determined by the scale adjustment unit 13 searching all values of the adaptive filter coefficient (Hreg) and obtaining a value having the largest absolute value. Here, if this value is written as Hreg_Max,
Hreg_max = Max (abs (Hreg [n])) (n = 0 ~ L_hreg) (1)
The scale adjustment unit 13 searches the absolute value of the coefficient of the adaptive filter in each frequency band and obtains the maximum value thereof, thereby determining the attenuation amount of the error signal to be given to the adaptive filter unit 11. Specifically, when the maximum absolute value Hreg_max of the adaptive filter coefficient is x and the target adaptive filter maximum value is target,
delta_s = target / x (2)
s_new = s_old * delta_s (3)
x_new = x * s_new / s_old (4)
Here, s_old is a scale value before scale adjustment, s_new is a scale value after scale adjustment, and x_new is a maximum absolute value of the adaptive filter coefficient after scale adjustment.

For example, if the maximum absolute value x of the adaptive filter coefficient when the scale value s_old is 2.5 times is 16384, and the target maximum value target is 32000,
delta_s = 32000 / (16384) = 1.953125
s_new = 2.5 * 1.953125 = 4.8828
The maximum absolute value x_new of the adaptive filter coefficient after scale adjustment is
x_new = x_old / s_new * s_old = 16384 / 2.5 * 4.8828 = 32000
It is corrected to.

  The scale adjusting unit 13 outputs s_old / s_new = 2.5 / 4.8828 to the scale attenuating unit 15 and outputs s_new / s_old = 4.8828 / 2.5 to the scale amplifying unit 14. That is, in this numerical example, the scale attenuating unit 15 amplifies the error signal, and the scale amplifying unit 14 amplifies the adaptive filter output.

  In addition, since correction of the scale value including division as shown in Expression (2) requires a calculation amount, correction using a bit shift may be performed. In this case, it is impossible to correct the maximum absolute value of the adaptive filter to 32767, but it is possible to correct it to about 30000 or more. For example, if the maximum absolute value of the adaptive filter is less than 15000, increase the right shift amount (or decrease the left shift amount) of the scale value (in this method, not the multiplication value) and exceed 30000 In this case, the maximum absolute value of the adaptive filter can always be within the range of 15000 to 30000 by reducing the right shift amount (or increasing the left shift amount). In terms of bit accuracy, it is not fully utilized, but a sufficient effect can be obtained compared to the case where there is no automatic correction of the scale value.

(2) When using the absolute value sum of adaptive filter coefficients As in the method of (1), if only the maximum absolute value of the adaptive filter coefficient is viewed, the value that the adaptive filter output value is determined by the number of bits of the DSP It may overflow. Since the output of the adaptive filter is a convolution operation of the adaptive filter coefficient and the reference signal (near-end signal), when the absolute maximum value of the adaptive filter coefficient is large, the output easily overflows a value determined by the number of bits of the DSP. . Therefore, the absolute value sum of the adaptive filter coefficients is used as information. In this case, the sum of the absolute values of the adaptive filter coefficients is set so that the output of the adaptive filter does not overflow the value determined by the number of bits of the DSP and the accuracy of the adaptive filter coefficients is adjusted to a scale that can be used to the maximum in the fixed-point DSP. Information.

  When considering a 16-bit fixed-point DSP, the value of the sum of absolute values of the adaptive filter needs to be 32767 or less. Therefore, the scale value is determined so that the sum of the absolute values is 32767 or less, as in the scale value correction based on the maximum absolute value of the adaptive filter described above. This method can use the same method as the scale value correction by the maximum absolute value of the adaptive filter, and a method of calculating a multiplication value by division or obtaining a shift amount is suitable.

Hreg_sum is the sum of absolute values of adaptive filter coefficients.
Hreg_sum = Sum (abs (Hreg [n])) (n = 0 ~ L_hreg) (5)
For example, assume that information of Hreg_sum = 32000 is obtained. If the sum Hreg_sum of the absolute values of the adaptive filter coefficients is x and the target maximum adaptive filter value is target,
delta_s = target / x = 32000 / (32000) = 1.0
In this case, the scale value s_new after the scale adjustment is not changed according to the equation (3), and the scale amplifying unit 14 and the scale attenuating unit 15 continue the operation at the scale so far.

  In this method, the adaptive filter coefficient may become extremely small. This is because the scale needs to be adjusted so that the absolute value sum of the adaptive filter coefficients is equal to or less than the maximum value that can be expressed by the fixed-point DSP, and the filter length of the adaptive filter is long (the number of additions in the sum is large). The scale adjustment may reduce the adaptive filter coefficient, and the coefficient accuracy may be extremely reduced. If you want to avoid this problem, refer to the adaptive filter output value described below, and try to maximize the adaptive filter coefficient so that the adaptive filter output value does not overflow at that time. It is good to adopt.

(3) When using the convolutional output value of the adaptive filter The output value of the adaptive filter can be determined by actually inputting an input signal to the adaptive filter unit 11 and observing the output signal. This value is denoted as Hreg_out, and the signal sequence input to the adaptive filter unit 11 is denoted as Xreg.
Hreg_out = Sum (Hreg [n] * Xreg [n]) (n = 0 ~ L_hreg) (6)
If Hreg_out = 10000 is obtained,
delta_s = 32000 / (10000) = 3.2
However, with this method, there is a possibility that the adaptive filter output instantaneously overflows a value determined by the number of bits of the DSP. Therefore, it is necessary to change the processing method so that the value determined by the number of bits of the DSP does not overflow during the convolution operation of the adaptive filter. For example, there are methods such as keeping the bit length of the convolution sum buffer larger, or multiplying by a coefficient that does not overflow when adding convolutions. In this case, the convolution output value of the adaptive filter does not overflow the value determined by the number of bits of the DSP of the adaptive filter, and the accuracy of the adaptive filter coefficient is adjusted to a scale that can be used to the maximum in the fixed-point DSP. Information.

  The (1) maximum absolute value of adaptive filter coefficients, (2) the sum of absolute values of adaptive filter coefficients, and (3) adaptive filter output values described above are almost always different. When all methods are considered, a method in which delta_s becomes a minimum value is adopted. When only Hreg_max is used, the accuracy of the adaptive filter coefficient is greatly improved, but overflow may occur during the adaptive filter convolution operation (adaptive filter output operation). Moreover, in Hreg_sum, it is possible to completely avoid overflow during adaptive filter convolution calculation (adaptive filter output calculation), but the adaptive filter accuracy may be significantly degraded. Furthermore, with Hreg_out, the accuracy of the adaptive filter can be improved to some extent, and overflow during the adaptive filter convolution operation (adaptive filter output operation) can be avoided to some extent.

  In this way, each method has different characteristics, which is not the best. When the present invention is actually used, it is properly used depending on the application. For example, when the result of the adaptive filter convolution operation can be stored in 32 bits instead of 16 bits, it may be considered that the overflow of the adaptive filter output hardly occurs. In such a program, better results can be obtained by determining the scale of the adaptive filter based on the value of Hreg_max and improving the accuracy of the adaptive filter.

  In addition, by setting the target maximum value of the adaptive filter to a small value such as 16384 instead of the maximum accuracy of 32768, the overflow of the adaptive filter is less likely to occur, so the value of Hreg_max can be reduced by sacrificing accuracy to some extent. It is also possible to obtain a good result by determining the scale value using. This is the aforementioned margin. Furthermore, when the adaptive filter coefficient can be set at 32 bits instead of 16 bits, the overflow of the output can be completely avoided by determining the scale of the adaptive filter using Hreg_sum.

  However, when performing the above-described calculations, the amount of calculations increases significantly, and may not be possible in implementation. In practice, there are many cases where Hreg is 16 bits and Xreg is 16 bits, and in such cases, the scale of the adaptive filter can be completely adjusted by determining the scale of the adaptive filter using Hreg_out. Compared with the case where it does not, it should be able to obtain a good result.

  It should be noted that the scale value determination timing described above is ideally performed for each sample of the acoustic signal, similarly to the scale determination information acquisition timing.

  The scale value obtained by the scale adjustment unit 13 passes to the scale attenuation unit 15 and the scale amplification unit 14. The scale attenuating unit 15 attenuates the error signal from the subtracting unit 12 to the adaptive filter unit 11 by the scale value, and the scale amplifying unit 14 amplifies the output signal from the adaptive filter unit 11 to the subtracting unit 12 by the reciprocal of the attenuation amount. To do.

  The output signal (pseudo acoustic echo signal) of the scale amplifying unit 14 is subtracted from the collected sound signal by the subtracting unit 12 to obtain an error signal. This error signal is a residual signal in which the acoustic echo is canceled with the pseudo acoustic echo signal. By using this as an output signal, a sound collection signal from which the influence of the acoustic echo is eliminated can be obtained.

  FIG. 3 shows the observation result of the adaptive filter coefficient whose scale value is not corrected. FIG. 4 shows the observation result of the adaptive filter coefficient with the scale corrected. The two adaptive filter coefficients in each figure are obtained by dividing the acoustic echo in the acoustic system into each frequency band by the band dividing unit 9 and performing echo cancellation by adaptive processing in each band. It is an adaptive filter coefficient. Comparing the left and right adaptive filter coefficients of FIGS. 3 and 4, it can be seen that there is a large difference in the size of the adaptive filter coefficient due to the different frequency bands.

  Comparing these adaptive filter coefficients, there is a difference in the maximum value among the filter coefficients when the scale is not adjusted. In this case, the maximum value of the coefficient of the band shown on the left side is small, and the coefficient accuracy is deteriorated. On the other hand, when looking at the adaptive filter coefficient after the scale adjustment, the maximum value of the adaptive filter coefficient between the frequency bands is almost constant, and there is no difference in the accuracy of the adaptive filter coefficient. As a result, it can be seen that there is no variation in the amount of acoustic echo cancellation in each frequency band by frequency band division, and the performance of the adaptive filter can be utilized.

  The parameters and the like described above are not limited to the contents described, but can be changed within a range that maintains the gist thereof.

The block diagram which shows embodiment of the audio | voice sound collection apparatus of this invention The figure which shows the detailed structure of the echo canceller with respect to one frequency band Diagram showing an example of adaptive filter coefficient observation without scaling Diagram showing an example of adaptive filter coefficient observation when the scale is adjusted

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Far end signal input terminal 2 Reproduction | regeneration part 3 Sound collection part 4 Sound collection signal output terminal 5 Adaptive filter part 6 Subtraction part 7 Scale adjustment means 8 Band division part 9 Band division part 10 Band synthesis part 11 Adaptive filter part 12 Scale subtraction part 13 Scale adjustment unit 14 Scale amplification unit 15 Scale attenuation unit 16 Scale adjustment means

Claims (6)

In a sound pickup device that uses a fixed-point DSP for the echo canceller,
An adaptive filter unit that generates a pseudo acoustic echo signal for canceling an acoustic echo that circulates from the reproducing unit that reproduces the far end signal to the sound collecting unit that collects the near end signal;
A scale value is determined by an output from the adaptive filter unit, and an error signal, which is a difference signal between the acoustic echo signal that has passed through the sound collection unit and the pseudo acoustic echo signal, is attenuated by the scale value. A sound pickup apparatus, wherein the level of the pseudo acoustic echo signal is automatically adjusted by providing a scale adjusting means for amplifying the pseudo acoustic echo signal by the reciprocal of the attenuation amount and providing it for the cancellation. The scale adjusting means includes
A scale value is determined by an output from the adaptive filter unit, and a scale adjustment unit that determines an attenuation amount and an amplification degree that is a reciprocal of the attenuation amount based on the scale value;
A scale attenuation unit for attenuating the error signal by the attenuation amount;
The sound pickup apparatus according to claim 1, further comprising a scale amplification unit that amplifies the pseudo acoustic echo signal output from the adaptive filter unit based on the amplification degree.   In the automatic adjustment, based on the maximum absolute value of the filter coefficient in the adaptive filter unit, the filter coefficient does not overflow the value determined by the number of bits of the DSP, and the accuracy of the filter coefficient is maximized in the DSP. 3. The sound collecting device according to claim 1, wherein the sound collecting device is used so as to be usable.   In the automatic adjustment, the output of the adaptive filter unit does not overflow the value determined by the number of bits of the DSP based on the absolute value sum of the filter coefficients in the adaptive filter unit, and the accuracy of the filter coefficient is increased. 3. The sound collecting apparatus according to claim 1, wherein the sound collecting apparatus is used so as to be maximally used in the DSP.   In the automatic adjustment, based on the convolution output value in the adaptive filter unit, the output of the adaptive filter unit does not overflow a value determined by the number of bits of the DSP, and the accuracy of the filter coefficient is maximized in the DSP. 3. The sound collecting device according to claim 1, wherein the sound collecting device is used so as to be usable. Band division means for dividing the far-end signal and the output signal of the sound collection unit into a signal of a limited frequency band is provided,
6. The sound pickup apparatus according to claim 1, wherein the adaptive filter unit and the scale adjusting unit operate in correspondence with the signal of the divided frequency band.

Images