JPH07154307A - Acoustic echo removing device - Google Patents

Abstract

PURPOSE:To provide the acoustic echo removing device for which the performance degradation of operational stability caused by the difference of update frequency between respective blocks in the case of coefficient correction amount dicided update processing is improved. CONSTITUTION:In the case of divided update processing in which a coefficient sequence stored in a relevant pseudo impulse response register 9 is divided into N pieces and the entire coefficient sequence is updated M times as the number of total steps, a large interpolation correction loop gain alpha0 interpolated to a relevant coefficient correction amount arithmetic circuit 7 is applied to the block with high update frequency. An extremely small interpolation correction loop gain alphaM-1 is applied to the block to be updated at the frequency of 1/M times. Then, an interpolation correction loop gain alpham smaller than the interpolation correction loop gain alpha0 applied to the block with high update frequency but larger than the interpolation correction loop gain alphaM-1 applied to the block to be updated at the frequency of 1/M times is applied to the block to be updated more than 2/M times and less than M/M times.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used in a communication line, a room sound field control device, and a high-quality voice communication conference device, and an acoustic echo component in which a signal on a receiving path appears in a transmitting path via an acoustic echo path. The present invention relates to an acoustic echo canceller that removes noise.

[0002]

2. Description of the Related Art Generally, an acoustic echo canceller is used in communication hygiene and long-distance telephone lines using a submarine cable.
It can be roughly classified into one that removes reflection caused by impedance mismatch of the line-to-four-line converter and one that removes reverberation due to acoustic coupling of speaker's voice in a loudspeaker such as a video conference system. It is composed of a variable coefficient filter and a subtraction circuit that generate pseudo-acoustic echo. The basic operation of the acoustic echo canceller will be described below.

FIG. 9 is a diagram showing the basic structure of an acoustic echo canceller. The reception signal input terminal 1 is connected to the reception signal output terminal 2, and the reception signal of the reception signal input terminal 1 is branched and supplied to the variable coefficient filter 3 to generate pseudo acoustic echo. The transmission signal from the transmission signal input terminal 4 and the pseudo-acoustic echo that is the output of the variable coefficient filter 3 are input to the subtraction circuit 5, the acoustic echo component in the transmission signal is removed, and the output of the subtraction circuit 5 is It is output to the transmission signal output terminal 6. The output of the transmission signal output terminal 6 and the signal of the reception signal input terminal 1 are input to the correction amount calculation circuit 7, and the filter coefficient of the variable coefficient filter 3 is corrected by the output of the coefficient correction amount calculation circuit 7. The reception signal is input to the reception signal input register 8 in the variable coefficient filter 3, and the sum of products of the reception signal of the reception signal input register 8 and the pseudo impulse response of the pseudo impulse response register 9 is obtained by the sum of products circuit 10. Sum of products circuit 10
Is output as a pseudo acoustic echo. The reception signal output terminal 2 and the transmission signal input terminal 4 are connected to a two-wire to four-wire converter in the case of a long-distance telephone line, and to a speaker and a microphone in the case of a public telephone system.

Assuming that the signal propagation characteristic of the acoustic echo path is linear and represented by an FIR type digital filter, if the impulse response h (t) and the input received signal x (t) are used, the sampling time interval is Is T, and the acoustic echo y _{k at} time kT is represented by y _k = h′x _k (1). However, h = [h ₁ , h ₂ , ..., H _N ] '(2) x = [x _k-1 , ..., x _kn ]'': transposition of the vector.

On the other hand, the estimated value of h at time kT is h
If s _k , the estimated value ys _k of y _k is given by ys _k = hs _k 'x _k (3). In the acoustic echo canceller, when there is a voice signal in the reception signal input terminal 1 and there is no voice signal in the transmission signal input terminal 4 and only acoustic echo exists, the echo elimination operation is performed as an adaptive operation state. A learning identification method is generally adopted for this adaptive operation algorithm. Successive correction of hs _k by the learning identification method is performed by _{hs k + 1 = hs k +} α (x k e k) / x k 'x k (4). However, the _{e k = y k -ys k,} 0 < a α ≦ 1 (5) ek called the residual acoustic echo. Such a calculation operation is processed in the coefficient correction amount calculation circuit 7. The contents of the pseudo impulse response register 9 include the variable coefficient series h.
s _k is stored. α is a correction loop gain for determining the sensitivity of estimation, and a larger correction amount can be given as the value is closer to 1.0, and high-speed acoustic echo removal is possible. However, in actual use, near-end noise and It is necessary to change the setting depending on the line status. At present, the actual determination of the modified loop gain α depends on an empirical rule.

When the acoustic reverberation characteristic in the loud sound field is expressed by the FIR type digital filter in this way, several hundreds
~ It becomes a long configuration with several thousand taps, and the amount of calculation related to updating the modification amount of the variable coefficient sequence hsk becomes enormous and it cannot be realized by a small-scale hardware. Therefore, the variable coefficient sequence hsk is divided into several stages. A method of reducing the update calculation amount in one step is adopted. As an example, the case of the simplest two-division processing in the division updating method will be described. When the total number of variable coefficient sequences stored in the pseudo impulse response register 9 is N, the division contents of the coefficient sequence can be expressed as follows.

Hs1 _k : 0 to N / 2 hs2 _k : (N / 2) +1 to N The update algorithm applies the above-mentioned division range to formula (4)
More, expressed as _{hs1 k + 1 = hs1 k +} α (x k e k) / x k 'x k (6) hs2 k + 1 = hs2 k + α (x k e k) / x k' x k (7) things can be, M is 2, that is, an adaptive algorithm to update the entire variable coefficient series hs _k in two steps. Therefore, the calculation amount in one step can be reduced to 1/2, and of course, if the division number N is increased, the calculation amount can be reduced to 1 / N in proportion to the increase. However, although the amount of calculation can be reduced, the convergence time for eliminating a certain amount of acoustic reverberation becomes long. In order to improve the convergence time, each divided block is not updated in order, but each block is weighted to change the update frequency. As a result, the number of steps M for updating the entire coefficient series
However, it is possible to significantly improve the convergence speed. The impulse response characteristic of the sound field is used to weight each divided block. FIG. 2 shows an example of impulse response characteristics of the sound field observed using the maximum period sequence code. It can be seen that the characteristic exhibits a damping characteristic. By utilizing this characteristic, the convergence speed can be improved by preferentially performing the update process in the first half of the impulse response where power is concentrated.

[0008]

When a division update method with different update frequencies is used for updating the coefficient correction amount, if the update frequencies between the divided blocks are significantly different, it is difficult to update the low weighted blocks. Become. At this time, if a block having a low update frequency is updated with a large correction loop gain, a large input / output error is attempted to be corrected at once and a large correction error is generated. Further, as can be seen from FIG. 2, the coefficient of the portion where the acoustic echo delay time is large is extremely small, and the update frequency of the divided block corresponding to this portion is quite low. Therefore, the coefficient is updated with a large correction loop gain. For example, it gives a coefficient with a large error and increases noise. For this reason, when performing division processing with different update frequencies, there are problems that noise on the communication line is increased and the risk of howling is increased, which lowers operation stability. It was

The present invention has been made in view of the above points, eliminates the above-mentioned problems, is excellent in operation stability, and has high followability with respect to changes in acoustic echo paths, and high-speed audio. An object of the present invention is to provide an acoustic echo canceller that realizes echo canceling characteristics and performs acoustic echo control of a sound field while always maintaining a large acoustic echo canceling amount.

[0010]

SUMMARY OF THE INVENTION The present invention is to solve these problems and comprises a reception signal input terminal, a reception signal output terminal, a transmission signal input terminal, and a transmission signal output terminal. A variable coefficient digital filter that receives a received signal input from the received signal input terminal, a pseudo impulse response register that stores a coefficient sequence of the variable coefficient digital filter, the contents of the pseudo impulse response register, and the received signal input. Sum-of-products arithmetic circuit for performing convolution integration operation with the input signal from the terminal, and subtraction for obtaining a difference value between the pseudo-acoustic echo generated by the sum-of-products arithmetic circuit and the acoustic echo input from the transmission signal input terminal A circuit and a coefficient correction for adding a correction amount to the coefficient sequence of the pseudo impulse response register so that the variable coefficient digital filter supplies an approximate value of the acoustic echo. The amount calculation circuit and the coefficient sequence of the pseudo impulse response register are divided into N blocks, and each block is sent to the coefficient correction amount calculation circuit so that the entire coefficient sequence is automatically updated with the total number of steps M times. In an acoustic echo canceller composed of a coefficient block selection circuit which sends a command for sequentially selecting and performing a coefficient update operation, the coefficient correction amount calculation circuit is interpolated according to the update frequency number set in each block. To provide an acoustic echo canceller in which the corrected corrected loop gain is set.

[0011]

The present invention prevents the deterioration of operational stability such as an increase in background noise on the communication line and an increased risk of howling due to the different update frequency of the coefficient correction amount division update by the above-mentioned means, and the update is performed. It is possible to realize fast and stable acoustic echo removal while significantly reducing the amount of computation involved, and to ensure high tracking even when the acoustic echo path characteristics fluctuate arbitrarily, so a steady state Since it has an excellent convergence speed to, it is possible to maintain a high level of acoustic echo cancellation at all times, and high-performance acoustic control becomes possible.

[0012]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a first acoustic echo canceller of the present invention. As shown in FIG. 1, according to the present invention, the conventional reception signal input terminal 1, reception signal output terminal 2, variable coefficient digital filter 3, transmission signal input terminal 4, subtraction circuit 5, transmission signal output terminal 6, coefficient correction. Quantity calculation circuit 7,
A coefficient block selection circuit 11 is added to a device having the same configuration as the acoustic echo canceling device that employs a learning identification method as an adaptive algorithm, which is composed of a reception signal input register 8, a pseudo impulse response register 9, and a product-sum operation circuit 10. It has been configured.

The receiving signal input terminal 1, the receiving signal output terminal 2, the transmitting signal input terminal 4, the transmitting signal output terminal 6, and the receiving signal input from the receiving signal input terminal 1 are input. The variable coefficient digital filter 3, the pseudo impulse response register 9 storing the coefficient series of the variable coefficient digital filter 3, the contents of the pseudo impulse response register 9 and the input signal from the reception signal input terminal 1. Sum-of-products arithmetic circuit 10 for performing the convolution-integral arithmetic operation of, and the subtraction circuit for taking the difference value between the pseudo-acoustic echo generated by the sum-of-products arithmetic circuit 10 and the acoustic echo input from the transmission signal input terminal 4. 5 and the variable coefficient digital filter 3
The coefficient correction amount calculation circuit 7 for adding a correction amount to the coefficient series of the pseudo impulse response register 9 so as to supply the approximate value of the acoustic echo, and the pseudo impulse response register 9
The coefficient sequence of is divided into N blocks, and the total number of steps M
A coefficient block selection circuit 11 that sends a command to the coefficient correction amount calculation circuit 7 to sequentially select each block and to perform a coefficient update operation so that the entire coefficient series is automatically updated in each time. In the echo canceller, the coefficient sequence stored in the pseudo impulse response register 9 is divided into N pieces and the entire coefficient series is updated M times. For a block in which a large interpolation correction loop gain α ₀ interpolated in the coefficient correction amount calculation circuit 7 is applied and the update frequency is set low so that it is updated every 1 / M times, A small interpolation correction loop gain α _m-1 is applied, and is updated every time for the n-th block whose update frequency is set to be updated at a rate of 2 / M times or more and less than M / M times. Interpolation set in the block An interpolation correction loop gain α _m smaller than the positive loop gain α _{0 and} larger than the interpolation correction loop gain α _M−1 set in the block updated every 1 / M times is adapted.

The magnitude relation of the above-mentioned interpolation correction loop gain can be expressed as follows.

0 <α _M-1 <... <α _m <... <α ₀ ≦ 1 (8) Then, when the interpolation correction loop gain is used to construct the successive update algorithm, the equation (4) is obtained. From this, it can be shown as in Expression (9). However, 1 <n <N.

[0016] _{hs1 k + 1 = hs1 k +} α0 (x k e k) / x k 'x k hsn k + 1 = hsn k + α m (x k e k) / x k' x k (9) hsN k + _{1 = hsN k + α M-} 1 (x k e k) / x k 'x variable coefficients of each n-divided blocks in a sequential update algorithm shown in _k equation (9) sequence matrix hsn (n = 1,2, ··
, N) can be expressed as in Expression (10) by arranging the blocks in ascending order of update frequency from the block with the smallest number of updates.

HsN <... <hsn <... <hs1 (10) Of course, the update frequency of each block may be the same due to the setting scheduling of division update. In such a case, whether or not to adopt the interpolation correction loop gain having the same value may be determined depending on how the operation characteristics are given.

An acoustic echo characterized by performing a division process for updating the coefficient correction amount using each interpolation correction loop gain stored in the coefficient correction amount calculation circuit 7 in accordance with the setting update frequency under the above conditions. Removal device.

FIG. 3 and FIG. 4 show the pseudo impulse responses of the one in which the interpolation correction loop gain is set in accordance with the update frequency of the division processing according to the present invention and the one in which the fixed interpolation correction loop gain of the related art is set. It is an observation result of the coefficient series of the register 9. The result of using the adaptive interpolation correction loop gain according to the present invention of FIG. 3 shows that the impulse response exhibits the damping characteristic, while the result of using the fixed interpolation correction loop gain of FIG. 4 shows a coefficient of a long delay time. The part has a large value,
The impulse response does not exhibit damping characteristics as a whole. Each impulse response is observed at the same step on the simulation with the same reference signal as an input. FIG. 5 and FIG. 6 are the results of obtaining the electric powers of FIG. 3 and FIG. 4, respectively. It can be seen that in the case of using the fixed interpolation correction loop gain of FIG. 6, the power is distributed over the entire coefficient sequence. As for the power distribution of the actual impulse response of the sound field, the coefficient power of the long delay time portion is extremely smaller than the coefficient power of the short delay time portion as shown in FIG. From this fact as well, it is understood that the present invention is effective in the division processing in which the intermittent update is performed.

FIG. 7 and FIG. 8 are observations of the power displacement of the error signal when the actual voice of an adult female is input as a reference signal. The one using the adaptive interpolation correction loop gain shown in FIG. 7 conspicuously attenuates the error signal, that is, the acoustic echo, while the one using the fixed interpolation correction loop gain in FIG. It can be seen that the erroneous erasure has occurred. The result of FIG. 8 is a reverberant sound that is very annoying to the ear.

[0021]

As described above, according to the present invention, the following effects can be obtained.

(1) Since the deterioration of the convergence speed of the acoustic echo canceling characteristic due to the division processing of updating the coefficient correction amount can be corrected, the acoustic echo removal can be speeded up.

(2) Without degrading the acoustic echo canceling performance,
Since the internal calculation amount of the adaptive algorithm can be greatly reduced, hardware can be realized with a small configuration,
The cost can be reduced.

(3) The variation factor of the echo path characteristic is the spatial movement of a person or an object close to the microphone and the speaker. That is, in the present invention in which the low-order tap of the impulse response is updated at a high frequency and a large interpolated correction loop gain is adapted, in the present invention, since the rising speed of the acoustic echo cancellation characteristic is high, the followability to the echo path variation is high. Is very strong and can quickly create a steady state of the communication line.

(4) It is possible to reduce the amount of calculation involved in updating the variable coefficient sequence to half or less while ensuring high quality of the communication line.

(5) Since the amplitude fluctuation of the error signal due to erroneous erasure is very small, the quasi-steady state is maintained, and the relatively large level of the residual echo signal does not exist on the communication line, so that the bidirectional communication can be easily detected. As a result, voice deterioration such as the cut off of the head of the transmitted voice is eliminated, and high sound quality is secured.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of an acoustic echo canceller according to the present description.

FIG. 2 is a diagram showing an example of impulse response characteristics of a sound field observed by the maximum period sequence code used in this description.

FIG. 3 is a diagram showing an example of a coefficient sequence stored in a pseudo impulse response register according to the present invention used in this description.

FIG. 4 is a diagram showing an example of a coefficient sequence stored in a pseudo impulse response register according to the conventional technique used in this description.

FIG. 5 is a diagram showing an example of coefficient sequence power stored in a pseudo impulse response register according to the present invention used in the present description.

FIG. 6 is a diagram showing an example of coefficient sequence power stored in a pseudo impulse response register according to a conventional technique used in the present description.

FIG. 7 is a diagram showing an example of a transition of error signal power when an adult female voice according to the present invention used as a reference signal is used in the present description.

FIG. 8 is a diagram showing an example of a transition of error signal power when a voice of an adult female is used as a reference signal according to the conventional technique used in the present description.

FIG. 9 is a block diagram showing an example of a basic configuration of an acoustic echo canceller using a conventional general learning identification method.

[Explanation of symbols]

 1 reception signal input terminal 2 reception signal output terminal 3 variable coefficient filter 4 transmission signal input terminal 5 subtraction circuit 6 transmission signal output terminal 7 correction amount calculation circuit 8 reception signal input register 9 pseudo impulse response register 10 sum of products calculation circuit 11 Coefficient block selection circuit

Claims (1)

[Claims] Claims: 1. A reception signal input terminal, a reception signal output terminal, a transmission signal input terminal, a transmission signal output terminal, and a variable coefficient digital filter having the reception signal input from the reception signal input terminal as an input. A pseudo impulse response register that stores a coefficient sequence of the variable coefficient digital filter; a product-sum operation circuit that performs a convolution integral operation of the contents of the pseudo impulse response register and an input signal from the reception signal input terminal; A subtraction circuit for obtaining a difference value between the pseudo acoustic echo generated by the product-sum operation circuit and the acoustic echo input from the transmission signal input terminal, and the variable coefficient digital filter supplying an approximate value of the acoustic echo And a coefficient correction amount calculation circuit for adding a correction amount to the coefficient sequence of the pseudo impulse response register, and the coefficient sequence of the pseudo impulse response register to N blocks. Coefficient block, which sends a command for sequentially selecting each block to the coefficient correction amount calculation circuit and performing a coefficient update operation so that the entire coefficient sequence is automatically updated with the total number of steps M times. In an acoustic echo canceller including a selection circuit, a correction loop gain inserted in the coefficient correction amount calculation circuit is set according to the number of update frequencies set in each block. Removal device.