JP4395195B1 - Echo canceller - Google Patents

Abstract

A calculation amount of a filter for generating a pseudo echo is reduced.
An adaptive filter 106a generates a whitened pseudo echo rw (t) by calculating a linear prediction residual vector ε (t), which is a finite number of non-zero excitation pulse trains, with a tap coefficient H (t). . The inverse filter 106b whitens the digital transmission audio signal y (t). The adder 106c subtracts the whitened pseudo echo rw (t) from the whitened transmission audio signal yw (t) to obtain a whitened residual signal ew (t). The pseudo echo generation filter 106d calculates a linear prediction residual vector ε (t) using the tap coefficient H (t) output from the adaptive filter 106a to generate a whitened pseudo echo rw ′ (t). The synthesis filter 106e synthesizes the whitened pseudo echo rw ′ (t) using the linear prediction coefficient L (t) to generate the pseudo echo r (t). The adder 106f subtracts the pseudo echo r (t) from the digital transmission voice signal y (t) to obtain a residual signal e (t).
[Selection] Figure 1

Description

  The present invention relates to an echo canceller for canceling an echo generated when a sound output from a speaker is fed back to a microphone.

  FIG. 1 is a block diagram showing a configuration of a conventional echo canceller. In FIG. 1, an encoded digital signal (received signal) is decoded into a digital received voice signal x (t) by a voice decoding circuit (SPDEC) 11 and converted into an analog received voice signal by a digital / analog converter (DAC) 12. Then, it is output as sound from the speaker 13. The analog transmission voice signal input to the microphone 14 is converted into a digital transmission voice signal y (t) by an analog / digital converter (ADC) 15 and encoded digital signal (SPCOD) 17 by a voice encoding circuit (SPCOD) 17. (Transmission signal). At that time, when the sound output from the speaker 13 is fed back to the microphone 14, an echo is generated. When echoes occur, the voice you spoke is heard behind the speaker, which is very annoying. In hands-free telephones and the like, a specific frequency component in an echo is amplified and howling (oscillation phenomenon) may occur.

  The echo canceller 16 is a circuit for preventing echo and howling. The echo canceller 16 is adapted to generate a pseudo echo r (t) by calculating a digital reception voice signal x (t) at a discrete time t with a tap coefficient H (t). The filter 16a and an adder 16b that subtracts the pseudo echo r (t) from the digital transmission voice signal y (t) output from the analog / digital converter 15 and outputs a residual signal e (t). The residual signal e (t) is output to the transmission line and given to the adaptive filter 16a as a tap coefficient update signal.

In addition to the sound s (t), the background noise n (t) and the echo d (t) from the speaker 13 are input to the microphone 14, and these are output by the analog / digital converter 15 to the digital transmission sound signal y (t). Is converted to Assuming that the impulse response of the echo path of the echo d (t) can be approximated by an mth-order FIR (finite impulse response) filter, the tap coefficient H (t) of the adaptive filter 16a at the discrete time t, and the adaptive filter The digital reception voice signal x (t) input to 16a is expressed by the following equations (1) and (2), respectively. However, T represents transposition of a vector.
H (t) = [h1 (t), h2 (t),..., Hm (t)] ^T (1)
x (t) = [x (t), x (t−1),..., x (t−m + 1)] ^T (2)

Thereby, the pseudo echo r (t) and the residual signal e (t) are expressed as the following (3) and (4).
r (t) = H (t) ^TX (t) (3)
e (t) = y (t) -r (t) (4)

  That is, the residual signal e (t) is a signal obtained by subtracting the pseudo echo r (t) from the digital transmission voice signal y (t) and suppressing the echo component included in the digital transmission voice signal y (t).

The update of the tap coefficient H (t) is generally expressed as the following equation (5). Note that μ in the equation (5) is a step size for adjusting the convergence speed of the tap coefficient, and the update vector ΔH (t) varies depending on the type of the adaptive algorithm.
H (t + 1) = H (t) + μΔH (t) (5)

  However, in the echo canceller shown in FIG. 1, when the digital received speech signal x (t) input to the adaptive filter 16a is a colored signal such as speech (a signal with a nonuniform spectrum distribution), the convergence speed is remarkably high. There is a problem of lowering.

  As a solution, a method has been proposed in which an input signal is converted into a white signal (a signal having a uniform spectral distribution) by an inverse filter to increase the convergence speed (Patent Document 1).

  The echo canceller of Patent Document 1 includes a first inverse filter that whitens the voice of its own input to the microphone, and a second inverse filter that whitens the voice of the other party reproduced from the speaker. The filter generates a pseudo echo using the white signal output from the two inverse filters.

  Further, Non-Patent Document 1 discloses a residual excitation linear prediction (“Residual” using a linear prediction coefficient and a linear prediction residual vector as a speech encoding technique in an echo canceller provided in a digital communication apparatus including a speech decoding circuit. Excited Linear Predictive (RELP) ”), instead of obtaining the tap coefficient of the filter using the whitened signal of the received speech signal (see Figure 1), pseudo echo generation based on the linear prediction residual vector (See Figure 2 on page 1334, "III. ECHO CANCELLATION" and "IV. VOICE CODEC AND ECHO CANCELLER INTEGRATION").

  The echo canceller of Non-Patent Document 1 eliminates the need to add an inverse filter for whitening the received speech signal because the whitened linear prediction synthesized sound source signal is input to the adaptive filter. Compared to a canceller, the configuration can be simplified and the convergence speed can be greatly improved. Further, the echo canceller of Non-Patent Document 1 omits the calculation for the sample of the value 0 when the input signal to the adaptive filter becomes a sparse pulse train when the information compression of the sound source is performed using multi-pulse encoding or the like. Therefore, the calculation amount of the adaptive filter can be reduced as compared with the echo canceller disclosed in Patent Document 1.

JP 2002-94419 A

WILSON PJ et al., 'An integrated voice codec and echo canceller implemented in a single DSP processor', IEEE International Conference on Acoustics, Speech, and Signal Processing, USA, IEEE, April 1986, vol.11, pp.1333 -1336

  However, the conventional echo canceller disclosed in Non-Patent Document 1 cannot reduce the calculation amount of the pseudo echo generation filter because the input signal to the pseudo echo generation filter is synthesized speech.

  The present invention has been made in view of such a point, and an object thereof is to provide an echo canceller capable of reducing the amount of calculation of a filter.

  The echo canceller of the present invention receives a linear prediction coefficient and a linear prediction residual vector obtained by linear prediction analysis, drives a synthesis filter with the linear prediction coefficient, synthesizes the linear prediction residual vector, and receives received speech. An echo canceller provided in a digital communication apparatus having a speech decoding circuit for obtaining a signal, wherein the tap coefficient setting means obtains a filter tap coefficient based on a signal obtained by whitening a transmission speech signal and the linear prediction residual vector And echo removing means for generating a pseudo echo using the tap coefficient, the linear prediction coefficient, and the linear prediction residual vector, and suppressing an echo included in the transmission voice signal using the pseudo echo. The structure to do is taken.

  According to the present invention, when sound source information compression is performed using multi-pulse coding or the like, the input signal to the pseudo-echo generation filter is also a sparse pulse train, and the calculation for the sample of value 0 can be omitted. Therefore, the calculation amount of the filter can be reduced as compared with the conventional echo canceller.

Block diagram showing the configuration of a digital communication apparatus including a conventional echo canceller 1 is a block diagram showing a configuration of a digital communication apparatus including an echo canceller according to Embodiment 1 of the present invention. The figure which shows an example of the synthetic | combination filter which concerns on Embodiment 1 of this invention. The figure which shows an example of the inverse filter which concerns on Embodiment 1 of this invention The figure which shows the state of the input signal of each filter of the conventional echo canceller The figure which shows the state of the input signal of each filter of the echo canceller which concerns on Embodiment 1 of this invention. FIG. 3 is a block diagram showing a configuration of a digital communication apparatus including an echo canceller according to Embodiment 2 of the present invention. The figure which shows the state of the input signal of each filter of the echo canceller which concerns on Embodiment 2 of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(Embodiment 1)
FIG. 2 is a block diagram showing a configuration of the digital communication apparatus including the echo canceller according to Embodiment 1 of the present invention.

  As shown in FIG. 2, the digital communication apparatus includes a speech decoding circuit 101, a digital / analog converter (DAC) 102, a speaker 103, a microphone 104, an analog / digital converter (ADC) 105, and an echo canceller. 106 and a speech encoding circuit 107.

  The audio decoding circuit 101 includes a demultiplexer 101a, a sound source signal generation circuit 101b, and a synthesis filter 101c. The echo canceller 106 includes an adaptive filter 106a, an inverse filter 106b, an adder 106c, a pseudo echo generation filter 106d, a synthesis filter 106e, and an adder 106f. The speech encoding circuit 107 includes an analysis filter 107a, a sound source characteristic analysis circuit 107b, and a multiplexer 107c.

  Here, in a cellular phone system or a videophone system, in order to reduce the amount of data transmission, excitation encoding processing for compressing the information amount of the excitation and excitation decoding processing for expanding the information amount of the compressed excitation are always performed. Is called. In the sound source encoding process, the digital transmission speech signal is divided into frames of about 10 to 20 ms, linear prediction analysis is performed for each frame, linear prediction coefficients (LPC) and linear prediction residual signals are obtained, and linear prediction coefficients and linear prediction are calculated. This is a process of individually encoding the residual signal. The sound source decoding process is a process of obtaining a digital received speech signal by driving a synthesis filter with a linear prediction coefficient and synthesizing a linear prediction residual vector. Note that the linear prediction coefficient is a parameter representing the spectrum envelope of speech. The linear prediction residual vector is a signal obtained by segmenting and transmitting a linear prediction residual signal and returning it to the original continuous data stream.

  An encoded digital signal (received signal) transmitted from a communication partner apparatus is composed of a sound source characteristic parameter obtained by encoding a linear prediction residual vector and a linear prediction coefficient, and is input to the speech decoding circuit 101.

  The voice decoding circuit 101 decodes the encoded digital signal (received signal) to obtain a digital received voice signal x (t). The demultiplexer 101a separates the encoded digital signal (received signal) into a linear prediction coefficient L (t) and a sound source characteristic parameter. The sound source signal generation circuit 101b decodes the sound source characteristic parameter to obtain a linear prediction residual vector ε (t). The synthesis filter 101c synthesizes the linear prediction residual vector ε (t) using the linear prediction coefficient L (t) to obtain a digital received speech signal x (t). The synthesis filter 101c is usually an all-pole type lattice filter using PARCOR (partial auto-correlation coefficient) coefficients or an LSP synthesis filter using LSP (line spectrum pair) parameters.

  The digital / analog converter 102 converts the digital reception voice signal x (t) output from the voice decoding circuit 101 into an analog reception voice signal. The analog received audio signal is output as audio from the speaker 103.

  The analog / digital converter 105 converts the analog transmission voice signal input to the microphone 104 into a digital transmission voice signal y (t). In addition to the sound s (t), the background noise n (t) and the echo d (t) from the speaker 103 are input to the microphone 104.

  The echo canceller 106 subtracts the pseudo echo r (t) from the digital transmission voice signal y (t) output from the analog / digital converter 105 to obtain a residual signal e (t). A detailed description of each component inside the echo canceller 106 will be described later.

  The voice encoding circuit 107 converts the residual signal e (t) output from the echo canceller 106 into an encoded digital signal (transmission signal). The analysis filter 107a performs linear prediction analysis on the residual signal e (t) to obtain a linear prediction coefficient and a linear prediction residual signal. The sound source characteristic analysis circuit 107b encodes the linear prediction residual signal to obtain a sound source characteristic parameter by performing pitch extraction processing and multi-pass encoding, vector quantization, or the like. By encoding the linear prediction residual signal, the data transmission amount can be compressed. The multiplexer 107c multiplexes the linear prediction coefficient and the sound source characteristic parameter to obtain a coded digital signal (transmission signal). The analysis filter 107a is usually an all-zero lattice filter.

  The encoded digital signal (transmission signal) is transmitted to the communication partner device and used for speech synthesis in the communication partner device.

  It should be noted that the demultiplexer 101a, the sound source signal generation circuit 101b, the synthesis filter 101c, the analysis filter 107a, the sound source characteristic analysis circuit 107b, and the multiplexer 107c in FIG. Regardless of whether or not the form of echo canceller is used, the digital communication apparatus is originally provided.

  Next, each component inside the echo canceller 106 will be described in detail.

  The linear prediction residual vector ε (t) output from the sound source signal generation circuit 101b is input to the adaptive filter 106a and the pseudo echo generation filter 106d. When the excitation information is compressed using multi-pass coding or the like, the linear prediction residual vector ε (t) is a finite number of non-zero excitation pulse sequences (excitation pulse sequences in which only a small number of samples have values other than 0). It becomes.

  The adaptive filter 106a calculates the linear prediction residual vector ε (t) at the discrete time t with the tap coefficient H (t) to generate the whitened pseudo echo rw (t). The adaptive filter 106a updates the tap coefficient H (t) so that the whitening residual signal ew (t) output from the adder 106c becomes an optimum value, and the updated tap coefficient H (t). Is output to the pseudo echo generation filter 106d.

  As described above, by using the linear prediction residual vector ε (t), which is a finite number of non-zero excitation pulse trains, as the input signal of the adaptive filter 106a, it is possible to omit the calculation for the zero-value sample. Compared with the case where the whitened received audio signal is used as the input signal of the adaptive filter (Patent Document 1), the calculation amount of the adaptive filter 106a can be reduced.

For example, consider a case where the sampling frequency is 10 kHz, the basic frequency of voiced sound is 100 Hz / the basic period is 10 ms (the number of samples in the basic period is 100), and the sound source signal for one period is a combination of ten non-zero pulses. In this case, the amount of calculation is reduced to 10/100 = 1/10 compared to when an adaptive algorithm that requires an amount of calculation in the order of N with respect to the tap length N of the filter is used. Further, in the adaptive algorithm which requires the calculation amount of the order of ^{N 2,} the amount of computation is reduced to ^(10/100) 2 = 1/100.

  In general, an LMS (Minimum Mean Square) algorithm, an NLMS (Normalized LMS) algorithm, a projection method, an RLS (Sequential Least Square) algorithm, or the like is used as an adaptive algorithm of the adaptive filter 106a. These are adaptive algorithms that perform sequential calculation each time a new signal sample value is input and gradually converge the tap coefficient to an optimum value. In addition, since the amount of calculation can be greatly reduced as described above, block processing using an input / output signal of a finite length is performed on the adaptive algorithm of the adaptive filter 106a, and the optimum tap coefficient is obtained by a single calculation. A least squares method can be used.

  The inverse filter 106b has the inverse characteristics of the synthesis filter 101c, and whitens the white speech signal y (t) output from the analog / digital converter 105 using the linear prediction coefficient L (t). A transmission voice signal yw (t) is obtained. The inverse filter 106b uses an all-pole lattice filter when the synthesis filter 101c is an all-zero lattice filter, and an all-pole type when the synthesis filter 101c is an LSP synthesis filter. A filter using the LSP parameter is used.

  The adder 106c subtracts the whitened pseudo echo rw (t) output from the adaptive filter 106a from the whitened transmission audio signal yw (t) output from the inverse filter 106b, and then the whitened residual signal ew (t )

  The pseudo echo generation filter 106d is a simple FIR filter, and calculates the linear prediction residual vector ε (t) output from the sound source signal generation circuit 101b using the tap coefficient H (t) output from the adaptive filter 106a. Thus, the whitened pseudo echo rw ′ (t) is generated.

  The synthesis filter 106e synthesizes the whitened pseudo echo rw '(t) using the linear prediction coefficient L (t) to generate the pseudo echo r (t).

  The adder 106f subtracts the pseudo echo r (t) output from the synthesis filter 106e from the digital transmission speech signal y (t) output from the analog / digital converter 105 to obtain a residual signal e (t). obtain. Thereby, the echo component contained in the digital transmission voice signal y (t) can be suppressed.

  Next, an example of the synthesis filter 101c and the inverse filter 106b will be described.

  FIG. 3 is a diagram illustrating an example when the synthesis filter 101c is an LSP synthesis filter using LSP parameters, and FIG. 4 is an inverse filter 106b when the LSP synthesis filter illustrated in FIG. 3 is used as the synthesis filter 101c. FIG.

3 and 4 are examples of the sixth order. In FIG. 3 and FIG. 4, z ⁻¹ represents a sample delay, and C _i is a filter coefficient obtained from the LSP parameter ω _i by the following equation (6).

The synthesis filter 101c shown in FIG. 3 has a transfer coefficient H _S (z) of the following equation (7). Further, the inverse filter 106b shown in FIG. 4 has a transfer coefficient H _A (z) of the following equation (8). Note that the constant P in the equations (7) and (8) is the order of the filter, and the equations (7) and (8) show the case where the order P is an even number.

The transmission coefficient of the transmission coefficient _H S (z) and inverse filter 106b of synthesis filter 101c _H A (z), have a relationship of the following equation (9).

  Next, the state of the input signal of each filter of the echo canceller according to Embodiment 1 of the present invention will be described using FIG. 5 in comparison with the prior art.

  FIG. 5A is a diagram illustrating a state of an input signal of each filter of a conventional echo canceller (Non-Patent Document 1). FIG. 5B is a diagram showing a state of an input signal of each filter of the echo canceller according to the present embodiment.

  As shown in FIGS. 5A and 5B, when a sound source characteristic parameter whose information amount is compressed is decoded by the sound source signal generation circuit, a linear prediction residual vector is obtained. This linear prediction residual vector is a white signal and a sound source pulse train (sparse pulse train) in which only a small number of samples have a value other than 0. Further, when a synthesis filter is driven by a linear prediction coefficient to synthesize a linear prediction residual vector, synthesized speech (received speech signal) is obtained.

  As shown in FIG. 5A, in the echo canceller of Non-Patent Document 1, since the input signal to the adaptive filter is a sparse pulse train, the calculation amount of the adaptive filter can be reduced. However, in the echo canceller of Non-Patent Document 1, since the input signal to the pseudo echo generation filter is synthesized speech, the amount of calculation of the pseudo echo generation filter cannot be reduced.

  On the other hand, as shown in FIG. 5B, in the echo canceller according to the present embodiment, both the input signals of the adaptive filter 106a and the pseudo echo generation filter 106d are sparse pulse trains, and the order is large (the tap length is long). After the process of the pseudo echo generation filter 106d, the process of the synthesis filter 106e having a small order (short tap length) is performed. Therefore, in the echo canceller according to the present embodiment, the calculation amount of the adaptive filter 106a and the pseudo echo generation filter 106d can be reduced.

  Note that the echo canceller of the present embodiment has an increased number of synthesis filters 106e compared to the echo canceller of Non-Patent Document 1, but the generation of pseudo echoes rather than an increase in the amount of computation due to the addition of the synthesis filter 106e. The amount of calculation reduced in the filter 106d for use is much larger. This is because, usually, the duration of the echo to be canceled, that is, the tap length of the pseudo echo generation filter 106d is very large, so that the filter order of the pseudo echo generation filter 106d is much larger than that of the synthesis filter 106e. Because it becomes. For example, if the sampling frequency is 10 kHz, the filter order of the pseudo-echo generation filter 106d having an echo length of 50 ms is 500th, whereas the filter order of the synthesis filter 106e is about 10-18.

  Thus, according to the present embodiment, when sound source information compression is performed using multi-pulse encoding or the like, the input signal to the pseudo echo generation filter also becomes a sparse pulse train, and in the pseudo echo generation filter Since the calculation for the sample having the value 0 can be omitted, the calculation amount of the filter can be reduced as compared with the conventional echo canceller.

(Embodiment 2)
FIG. 6 is a block diagram showing a configuration of a digital communication apparatus including an echo canceller according to Embodiment 2 of the present invention. In FIG. 6, the same components as those in FIG. 2 are denoted by the same reference numerals as those in FIG. 2, and detailed description thereof is omitted.

  In the digital communication apparatus shown in FIG. 6, the internal configuration of the echo canceller 206 is different from that of the echo canceller 106 of the digital communication apparatus in FIG. The echo canceller 206 has a configuration in which the pseudo echo generation filter 106d and the synthesis filter 106e are deleted from the echo canceller 106, and a synthesis filter 206a is added.

  The synthesis filter 206a synthesizes the whitened pseudo echo rw (t) generated by the adaptive filter 106a using the linear prediction coefficient L (t) to generate a pseudo echo r (t).

  FIG. 7 is a diagram showing the state of the input signal of each filter of the echo canceller according to the present embodiment.

  As shown in FIG. 7, in the echo canceller according to the present embodiment, since the input signal to the adaptive filter 106a is a sparse pulse train, the calculation amount of the adaptive filter 106a can be reduced. Further, in the echo canceller according to the present embodiment, the pseudo filter is generated using the whitened pseudo echo output from the adaptive filter 106a by the synthesis filter 206a having a small order (short tap length). The calculation amount of the filter can be reduced with respect to one echo canceller.

  Thus, according to the present embodiment, when sound source information compression is performed using multi-pulse coding or the like, the whitened pseudo echo obtained by the filter operation using a sparse pulse train is used in the adaptive filter. Therefore, the amount of calculation of the filter can be reduced as compared with the conventional echo canceller.

  In the first and second embodiments, a synthesis filter having a small order is added, so that input signals to all the filters (adaptive filter and pseudo echo generation filter) having a large order are converted into linear prediction residuals. It can be a vector, and has a common effect that the amount of calculation of the filter can be reduced when the sound source information is compressed using multi-pass coding or the like.

  Hereinafter, the calculation amount of the echo canceller is compared for each method of Non-Patent Document 1, Embodiment 1 and Embodiment 2. Here, the tap length of the adaptive filter and the pseudo echo generation filter is N, and the linear prediction analysis synthesis order is P order. The calculation amount of the adaptive filter is k times the tap length (ie, kN), the calculation amount of the pseudo echo generation filter is N equal to the tap length, and the calculation amounts of the linear prediction synthesis filter and the inverse filter are P equal to the order. . In addition, the compression ratio when the information compression of the linear predictive synthesized sound source is performed by multipulse coding is α. Note that the number of pulses of the synthesized sound source signal when synthesizing speech for one period of M samples is αM.

<Non-Patent Document 1>
Adaptive filter αkN
Pseudo echo generation filter N
Inverse filter P
-------------------------
Total (αk + 1) N + P

<Embodiment 1>
Adaptive filter 106a αkN
Pseudo echo generation filter 106d αN
Inverse filter 106b P
Synthesis filter 106e P
-------------------------
Total (αk + α) N + 2P

<Embodiment 2>
Adaptive filter 106a αkN
Inverse filter 106b P
Synthesis filter 206a P
-------------------------
Total αkN + 2P

  For example, when the tap length N = 500, the order P = 16, the sound source information compression ratio α = 0.1, and the coefficient k = 3, the calculation amount is as follows. Note that the value of k varies depending on the adaptive algorithm used. k = 3 is a value when the NLSM algorithm is used.

<Non-Patent Document 1>
Calculation amount = (αk + 1) N + P
= (0.1 × 3 + 1) × 500 + 16
= 666

<Embodiment 1>
Calculation amount = (αk + α) N + 2P
= (0.1 × 3 + 0.1) × 500 + 2 × 16
= 232

<Embodiment 2>
Calculation amount = αkN + 2P
= 0.1 × 3 × 500 + 2 × 16
= 182

  As is clear from the above results, the amount of calculation can be significantly reduced in the methods of the first and second embodiments than in the non-patent document 1. It can also be seen that the method of Embodiment 2 has the least amount of computation.

  In the present invention, the linear prediction coefficient output of the analysis filter 107a and the linear prediction coefficient input of the synthesis filter 101c in FIG. 2 are short-circuited, the sound source characteristic parameter output of the sound source characteristic analysis circuit 107b and the sound source characteristic of the sound source signal generation circuit 101b. The present invention can also be applied to a howling canceller of a loudspeaker having a configuration in which a parameter input is short-circuited.

  The present invention is suitable for use in an echo canceller, a howling canceller, a howling canceller for a loudspeaker, a howling canceller for a hearing aid, etc. in a bidirectional communication system (wireless telephone, wired telephone, interphone, video conference system, etc.).

DESCRIPTION OF SYMBOLS 101 Speech decoding circuit 101a Demultiplexer 101b Sound source signal generation circuit 101c Synthesis filter 102 Digital / analog converter 103 Speaker 104 Microphone 105 Analog / digital converter 106, 206 Echo canceller 106a Adaptive filter 106b Inverse filter 106c Adder 106d For pseudo echo generation Filters 106e and 206a Synthesis filter 106f Adder 107 Speech encoding circuit 107a Analysis filter 107b Sound source characteristic analysis circuit 107c Multiplexer 206a Synthesis filter

Claims (3)

Speech decoding that receives a linear prediction coefficient and a linear prediction residual vector obtained by linear prediction analysis, drives a first synthesis filter with the linear prediction coefficient, and synthesizes the linear prediction residual vector to obtain a received speech signal An echo canceller provided in a digital communication device provided with a circuit,
An adaptive filter that computes the linear prediction residual vector with a tap coefficient to generate a first whitened pseudo echo, and updates the tap coefficient according to a given whitened residual signal;
An inverse filter having a reverse characteristic to the first synthesis filter and using the linear prediction coefficient to whiten a transmission voice signal and obtain a whitened transmission voice signal;
A first adder for subtracting the first whitened pseudo echo from the whitened transmission audio signal to obtain a whitened residual signal;
A pseudo-echo generation filter that calculates the linear prediction residual vector using the tap coefficient updated by the adaptive filter to generate a second whitened pseudo-echo;
A second synthesis filter for synthesizing the second whitened pseudo echo using the linear prediction coefficient to generate a pseudo echo;
A second adder that subtracts the pseudo echo from the transmission voice signal to obtain a residual signal;
An echo canceller comprising: Speech decoding that receives a linear prediction coefficient and a linear prediction residual vector obtained by linear prediction analysis, drives a first synthesis filter with the linear prediction coefficient, and synthesizes the linear prediction residual vector to obtain a received speech signal An echo canceller provided in a digital communication device provided with a circuit,
An adaptive filter that computes the linear prediction residual vector with a tap coefficient to generate a whitened pseudo echo, and updates the tap coefficient according to a given whitened residual signal;
An inverse filter having a reverse characteristic to the first synthesis filter and using the linear prediction coefficient to whiten a transmission voice signal and obtain a whitened transmission voice signal;
A first adder that subtracts the whitened pseudo echo from the whitened transmission audio signal to obtain a whitened residual signal;
A second synthesis filter that has the same characteristics as the first synthesis filter, synthesizes the whitened pseudo echo using the linear prediction coefficient, and generates a pseudo echo;
A second adder that subtracts the pseudo echo from the transmission voice signal to obtain a residual signal;
An echo canceller comprising: The echo canceller according to claim 1 or 2 , wherein the linear prediction residual vector is a finite number of non-zero excitation pulse trains.

Images