JPH0766758A - Sound encoding/decoding type echo canceler - Google Patents

Abstract

PURPOSE:To quickly and accurately estimate the impulse response of an echo line with a small amt. of calculation. CONSTITUTION:A code encoded by a CELP system is decoded by a decoder 13 and supplied to a pseudo echo line 24 to be sent to an echo line 23 after D/A conversion. A signal from the echo line 23 is A/D converted, an output from the pseudo echo line 24 is subtracted from the A/D converted signal by an erasing circuit 18 and the result is supplied to a encoder 19. An excitation signal decoded by the decoder 13 is supplied to an impulse response estimating part 25 through a band width enlarging synthetic filter 54. The filter coefficient of the synthesizing filter 54 is set up to a value corresponding to a decoded spectrum envelope parameter, the spectrum envelope of the excitation signal whose frequency characteristic is almost flat is made slightly rugged to form a moderate whitened signal. The signal and the output of the circuit 18 are used for obtaining the estimation of an impulse response by a method similar to a conventional one.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller for canceling echo signals which are a cause of howling and a hearing loss in a loudspeaker telephone conference communication system, a two-wire to four-wire conversion system, etc. For the signal to
An encoder and a decoder that perform high-efficiency voice encoding and decoding are provided.

[0002]

2. Description of the Related Art FIG. 6A shows a loudspeaker type communication terminal device provided with a high-efficiency voice encoding / decoding device of this type. Input terminal 1
1 is decoded into a baseband signal by a transmission line decoder 12, and the baseband signal is decoded by a voice decoder 13 into a coded voice signal, for example, a telephone band voice signal. Further, it is converted into an analog signal by the D / A converter 14. This analog audio signal is supplied to the speaker 15 and is emitted as an acoustic signal. On the other hand, the voice signal received by the microphone 16 is converted into a digital signal by the A / D converter 17, the echo signal is eliminated by the erasing circuit 18, and the voice signal is supplied to the voice coder 19 for high efficiency voice encoding. The encoded voice signal is encoded by the transmission line encoder 21 into a signal on the transmission line and transmitted from the output terminal 22 to the transmission line. In order to prevent the acoustic signal emitted from the speaker 15 from being captured by the microphone 16 and transmitted as an echo signal, the pseudo echo path 2 that imitates the echo path 23 that connects the speaker 15 and the microphone 16 is simulated.
4 is connected to the input side of the speaker 15, the signal to the speaker 15 is branched and supplied to the pseudo echo path 24, and the output passing therethrough is supplied to the cancellation circuit 18 and subtracted from the signal from the microphone 16, that is, the echo. The signal is allowed to cancel. Input signal of speaker 15 and erase circuit 18
And the output signal of the input signal is input to the impulse response estimation unit 25, the impulse response of the echo path 23 is estimated, the estimated impulse response characteristic is set in the pseudo echo path 24, and the signal input to the pseudo echo path 24 is It is designed to fold the impulse response.

Similarly, in a 4-wire / 2-wire conversion system, FIG.
6A, the output side of the D / A converter 14 and the input side of the A / D converter 17 are connected to the 4-wire side terminal of the hybrid transformer 26, as indicated by the same reference numerals. The 2-wire type transmission line 27 is connected to the 2-wire side terminal of the hybrid transformer 26. There is an echo path 28 in which the output signal of the D / A converter 14 leaks from the hybrid transformer 26 and reaches the A / D converter 17 side, and the echo signal passing through this echo path 28 is canceled by the canceling circuit 18 in the case of FIG. 6A. Similarly, it is canceled.

Further, as shown in FIG. 7, in a mobile radio communication base station 29, a digital voice signal from an analog network 31 is encoded by a voice encoder 19 and further encoded by a transmission line encoder 21. In the mobile terminal device 32, the signal of the base station 29 is transmitted to the mobile terminal device 32 via a wireless line, and the signal of the base station 29 is converted into a baseband signal by the transmission path decoder 33, and further decoded into a voice signal by the voice decoder 34
The audio signal is converted into an analog signal by the D / A converter 14 and supplied to the speaker 15. The audio signal from the microphone 16 is converted into a digital signal by the A / D converter 17,
The voice encoder 35 performs high-efficiency encoding, and the encoded output is converted into a code signal on the transmission line by the transmission line encoder 36 and transmitted to the base station 29 via a wireless line. In the base station 29, the received signal is decoded into a baseband signal in the transmission line decoder 12, and the baseband signal is decoded into a digital voice signal in the voice decoder 13 to obtain the analog network 31.
Sent to. Also in this case, the echo path 23 from the speaker 15 to the microphone 16 is formed, and the cancellation of the echo signal through the echo path 23 is performed by the voice encoder 19 of the base station 29.
Is performed by the pseudo echo path 24, the erasing circuit 18, and the impulse response estimation unit 25 provided between the input side of the ∘ and the output side of the speech decoder 13.

The speech coder and speech decoder shown in FIGS. 6A, 6B and 7 are for highly efficient coding and decoding of speech signals using linear prediction. For example, CELP (Code) is used.
Excited Linear Predictio
n: code-excited linear prediction) coding method is used. Briefly speaking, the input voice signal is L as shown in FIG. 8A.
The PC analysis unit 41 performs LPC analysis to obtain a spectrum envelope parameter for each block, and this parameter is set as a filter coefficient in the linear prediction synthesis filter 42. The excitation signal selected from the excitation source 43 is given a gain in the gain section 44 and is supplied to the linear prediction synthesis filter 42 as an excitation signal. The distortion evaluation unit 45 performs selection of the excitation signal of the excitation source 43 and gain control given to the gain unit 44 so that the distortion of the synthesized signal synthesized by the synthesis filter 42 with respect to the input speech signal is minimized. A code indicating an excitation signal (vector) in which an audio signal is selected in block units, a code indicating a set gain, and a spectrum envelope parameter are output as a coded signal.

As shown in FIG. 8B, the decoder for decoding this coded signal extracts the spectrum envelope parameter in the spectrum envelope decoder 47, sets it as the filter coefficient in the linear prediction synthesis filter 48, and also sets the excitation source. The excitation signal is selectively decoded by the decoder 49, and the excitation signal is provided with the gain decoded by the gain unit 51 and is input to the linear prediction synthesis filter 48 as the excitation signal.
The audio signal is restored and output from 8.

The echo cancellers shown in FIGS. 6 and 7 have different requirements for echo signal cancellation, but the principle of echo signal cancellation is the same. The acoustic echo canceller of FIG. 6A will be described below as an example. For the estimation of the echo response of the echo path, there is a method in which noise in a wide band is emitted from the speaker 15 before voice communication is started and the signal input to the microphone 16 is used. In this method, since the frequency band of the acoustic signal emitted from the speaker 15 is wide, an accurate impulse response can be estimated in a short time. However, there is a drawback in that it cannot follow the impulse response fluctuation based on the fluctuation of the echo path 23.

Apart from this method, there is a method of successively correcting the estimation of the impulse response of the echo path 23 while using the voice signal during communication. The problems of this method are the speed following the fluctuation of the impulse response, the estimation accuracy, and the calculation amount. For example, when a simple learning fixing method is used as this estimation method, when the input signal is a signal that is biased to a low frequency region such as voice, the speed of following the fluctuation of the impulse response is extremely reduced. Various methods have been attempted to solve this problem, but practical problems such as an increase in the amount of calculation have not been sufficiently solved.

[0009]

SUMMARY OF THE INVENTION An object of the present invention is to provide an echo canceller capable of accurately and quickly estimating the impulse response of an echo path with a relatively simple structure.

[0010]

The present invention is premised on an encoder for encoding / decoding a transmission / reception signal for an echo path with high efficiency using linear prediction, and an echo canceller provided with a decoder. According to the invention of claim 1, the decoded excitation signal of the decoder or the encoded excitation signal of the encoder is taken out and supplied to the impulse response estimation means, and the impulse response estimation is performed by this excitation signal and the output of the cancellation circuit. Done.

According to the invention of claim 2, the decoded spectrum circuit parameter or the coded spectrum envelope parameter is set as a filter coefficient in the bandwidth expansion synthesis filter,
The excitation signal is subjected to bandwidth expansion synthesis by this synthesis filter and supplied to the impulse response estimation means. According to the third aspect of the invention, the decoded speech signal from the decoder is passed through the linear prediction synthesis filter of the decoder and the filter having the inverse characteristic, and is supplied to the impulse response estimating means.

According to the fourth aspect of the invention, the input speech signal of the encoder is passed through the linear prediction synthesis filter of the encoder and a filter having an inverse characteristic, and is supplied to the impulse response estimating means. In each of the third and fourth aspects of the invention, the inverse characteristic filter may be a bandwidth expansion filter. The frequency characteristic of the excitation signal used for high-efficiency speech coding is almost flat, that is, it is almost white signal, and the frequency characteristic of the speech signal passed through the inverse characteristic filter is almost flat, that is, whitening. To be done. Therefore, the impulse response can be estimated in a short time.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows an embodiment of the invention as claimed in claims 1, 2 and 3, and the same reference numerals are given to the portions corresponding to those in FIGS. In this embodiment, the decoded excitation signal from the excitation source decoder 49 is branched and supplied to the bandwidth expansion synthesis filter 54, and the filter coefficient of the bandwidth expansion synthesis filter 54 is converted into the decoded spectrum envelope parameter from the spectrum envelope decoder 47. Correspondingly set. That is, when the transfer function of the linear prediction synthesis filter 48 is A (z), the transfer function of the bandwidth expansion synthesis filter 54 is A (γz), and γ is called a bandwidth expansion coefficient and is 1 or less, for example, 0. It is a constant of about 0.5. The decoded excitation signal is usually a whitened signal having a substantially flat frequency characteristic, and the linear prediction synthesis filter 48 makes the spectral envelope of the input excitation signal uneven according to the decoded spectral envelope parameter. In the expansion synthesis filter 54, the excitation signal is provided with unevenness weaker than the linear prediction synthesis filter 48 in its spectral envelope. Therefore, the output of the bandwidth expansion synthesis filter 54 becomes a gently whitened signal.

The gently whitened signal is supplied to the impulse response estimating section 25. The impulse response estimation unit 25 estimates the impulse response of the echo path 23 (28) from the gentle white signal and the output of the erasing circuit 18 in the same manner as the conventional method. Since a signal that has been gently whitened in this way is used to estimate the impulse response, there is not much frequency bias like a voice signal, and the estimation of the impulse response is more accurate than when using the voice signal as it is. And can be performed at high speed, and the echo path 23 (2
It is possible to adapt the characteristics of the pseudo echo path 24 by quickly and highly accurately following the fluctuation of the impulse response of 8).

It is also possible to omit the bandwidth expansion synthesis filter 54 and supply the decoded excitation signal directly to the impulse response estimation unit 25. In this case, the white signal is used for impulse response estimation, and similarly, it can be estimated at high speed and with high accuracy. However, although the decoding process is performed for each frame, the estimation process is performed at a cycle that is about 10 times longer than the frame. Therefore, when viewed from the estimation process cycle, the excitation signal is subjected to the bandwidth expansion process as described above. In many cases, the whitened state is obtained and the estimated speed and accuracy are better when the bandwidth is widened than when the bandwidth widening process is not performed.

As shown by the dotted line in FIG. 1, the decoded speech signal from the speech decoder 13 is expanded by the inverse bandwidth filter 55.
Then, the characteristics of the bandwidth expansion inverse filter 55 are controlled by the decoded spectrum envelope parameter so that the characteristics are the inverse characteristics of the linear prediction synthesis filter 48 and the bandwidth is expanded as described above. This band widening inverse filter 55
May be supplied to the impulse response estimation unit 25.
In this case, the supply of the excitation signal to the impulse response estimation unit 25 may be omitted or may be supplied simultaneously. The synthesized speech signal that has passed through the band widening inverse filter 55 has its spectrum envelope unevenness weakened and becomes a gently whitened signal, so that the impulse response can be estimated at high speed and with high accuracy. Also in this case, the filter 55 may have a characteristic exactly opposite to that of the linear prediction synthesis filter 48 without expanding the bandwidth, and the filter output may be a substantially perfect white signal.

FIG. 2 shows another embodiment of the present invention. The decoding excitation signal is supplied to the impulse response estimation unit 25 and the white signal is used for the impulse response estimation, which is the same as part of the description of FIG. 1. However, in this embodiment, the decoding excitation signal is not the decoding speech signal. The pseudo echo path 24 is supplied. In response to this, the digital signal from the echo path 23 (28) side from the A / D converter 17 is passed through the linear prediction synthesis filter 48 and the inverse filter 56 having the inverse characteristic, and the echo signal is also inverse filter 5.
It is whitened by 6 and supplied to the erasing circuit 18, and echoes are erased in the whitened series. The signal to be transmitted is also the inverse filter 5
6, the output of the elimination circuit 18 is passed through the linear prediction synthesis filter 57 having the same characteristics as the linear prediction synthesis filter 48 to remove the influence of the inverse filter 56 and the speech coder 1
Supply to 9.

As shown in FIG. 3, the decoded excitation signal is supplied to the pseudo echo path 24 and the impulse response estimation unit 25, and the output of the pseudo echo path 24 is simulated through the linear prediction filter 58 having the same characteristics as the linear prediction synthesis filter 48. The echo signals are combined and supplied to the cancellation circuit 18, and the output of the cancellation circuit 18 is supplied to the impulse response estimation unit 25 through the linear prediction filter 48 and an inverse filter 59 having an inverse characteristic.

An example in which the inventions of claims 2 and 4 are applied to the echo canceller shown in FIG. 7 is shown in FIG. 4, FIG. 1 and FIG.
Parts corresponding to those in FIG. 8 are designated by the same reference numerals. In this case, the coded excitation signal which is the output of the gain unit 44 in the voice encoder 19 is supplied to the impulse response estimation unit 25 as a signal that is gently whitened through the bandwidth expansion synthesis filter 54. The output coded excitation signal of the gain section 44 is a substantially white signal as in the case of the decoded excitation signal, and the same effect as in the case of FIG. 1 is obtained. In this case as well, the encoded excitation signal is directly
It may be supplied to the impulse response estimation unit 25. Further, as shown by the dotted line, the input speech signal of the speech encoder 19 is passed through the bandwidth expansion inverse filter 55 and the impulse response estimation unit 2
5 may be supplied. The inverse filter 55 may have a characteristic reverse to that of the linear prediction synthesis filter 42.

In FIG. 4, the reverberant signal passes through the reverberation path 23 via the speech encoder 19 and the decoder 34,
Further, it passes through the voice encoder 35 and the decoder 13 again to become an echo signal. For this reason, the quantization noise generated in the process of voice encoding and decoding double mixes into the impulse response of the echo signal, which is supposed to be estimated, as a cause of disturbance. Therefore, when this quantization noise cannot be ignored, the output coded signal of the voice encoder 19 is decoded by the local decoder 61 to obtain a voice signal as shown in FIG. It is supplied to the impulse response estimation unit 25. In this way, the output of the local decoder 61 becomes the same as the output of the speech decoder 34 of the mobile terminal device 32, so that the quantization error is mixed only once, and the impulse response can be easily estimated.

[0021]

According to the present invention, the estimation of the impulse envelope and the variation of the impulse response can be performed accurately and at high speed by directly using the processing of the spectrum envelope estimation and the whitening which are originally used in the speech coding. It is possible to follow the characteristics of the pseudo echo path.

An FIR filter having 512 taps is used as the pseudo echo path 24, and the block length is 160 samples.
The sampling frequency is 8 kHz and the step size is 1.
0, linear prediction order is 10, bandwidth expansion coefficient γ is 0.5
As a result, the results of obtaining the echo cancellation rate (dB) by simulation in the conventional learning identification method, the conventional projection method, and the embodiment shown in FIG. 1 are shown below.

[0023]

Table 1 Elapsed time 20 [ms] 200 [ms] 2 [s] 4 [s] Learning identification method 13.5 15.4 20.8 23.4 Projection method 21.0 21.0 26.0 27. 7. The present invention 15.6 20.5 25.6 28.1 This is a combination with voice coding in which a normal voice is input and an echo signal obtained by convolving a normal room impulse response is processed block by block. The erasure rate is the ratio of the echo signal and the residual echo energy accumulated in a specified time expressed in decibels.

From this result, the method of the present invention has performance equivalent to that of the conventional projection method, but the amount of calculation is almost the same as that of the learning identification method and is smaller than that of the projection method. This is because it is simpler to whiten a block once as in the present invention, as compared with the projection method in which whitening is sequentially performed for each sample, and the voice decoding process can be used. As a real-time processing device, voice coding and an echo canceller can be integrated into one signal processing LSI to be economical.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the invention of claims 2 and 3.

FIG. 2 is a block diagram showing an embodiment of the invention of claim 1;

FIG. 3 is a block diagram showing another embodiment of the invention of claim 1;

FIG. 4 is a block diagram showing an embodiment of the invention of claims 2 and 4.

FIG. 5 is a block diagram showing still another embodiment of the invention of claims 2 and 4.

6A is a block diagram showing a conventional acoustic echo canceller in a loudspeaker type communication terminal, and FIG. 6B is a block diagram showing a conventional line echo canceller.

FIG. 7 is a block diagram showing a conventional configuration for canceling a remote echo.

FIG. 8A is a block diagram showing an example of a speech encoder 19.
B is a block diagram showing an example of the speech decoder 13.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Shoji Shimada 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation

Claims (4)

[Claims] 1. A coder for highly efficient coding / decoding using linear prediction, and a decoder for coding / decoding a voice signal transmitted / received to / from an echo path, and a signal to the echo path and erasure. From the output of the means, the impulse response of the echo path is estimated by the impulse response estimation means, and the estimated impulse response is convolved with the signal to the echo path in the pseudo echo path, and the pseudo echo path In the echo canceller for subtracting the output signal from the signal from the echo path by the canceling means, the decoded excitation signal of the decoder or the encoded excitation signal of the encoder is extracted and supplied to the impulse response estimation means. An echo canceller combined with voice encoding / decoding, which comprises: 2. A means for expanding and combining the extracted excitation signal with the decoded spectrum envelope of the decoder or the coded spectrum envelope of the encoder and supplying it to the impulse response estimation means. An echo canceller with combined voice encoding and decoding according to claim 1. 3. A coder for highly efficient coding / decoding using linear prediction, and a decoder for coding / decoding a voice signal transmitted / received to / from an echo path, and a signal to the echo path and erasure. From the output of the means, the impulse response of the echo path is estimated by the impulse response estimation means, and the estimated impulse response is convolved with the signal to the echo path in the pseudo echo path, and the pseudo echo path In the echo canceller which subtracts the output signal from the signal from the echo path by the erasing means, the decoded voice signal from the decoder is input, has the inverse characteristic of the linear prediction synthesis filter of the decoder, and outputs. A voice canceller / combiner type echo canceller, comprising: a filter means for supplying the impulse response estimating means. 4. A coder for highly efficient coding / decoding using linear prediction, and a decoder for coding / decoding a voice signal transmitted / received to / from an echo path, and a signal to the echo path and erasure. From the output of the means, the impulse response of the echo path is estimated by the impulse response estimation means, and the estimated impulse response is convolved with the signal to the echo path in the pseudo echo path, and the pseudo echo path In the echo canceller for subtracting the output signal from the signal from the echo path by the canceling means, the input voice signal of the encoder is input, has the inverse characteristic of the linear prediction synthesis filter in the encoder, and outputs A voice canceller / combiner type echo canceller, comprising: a filter means for supplying the impulse response estimating means.