JPH0795711B2 - Echo signal canceller - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo signal canceling apparatus for canceling echo signals generated by acoustic coupling between a speaker and a microphone and suppressing howling in a communication conference system and a loudspeaker telephone. Is.

2. Description of the Related Art In recent years, with the development of communication conference systems connecting remote locations,
In the above system, an echo signal canceller has been attracting attention as a means for canceling echo signals generated by acoustic coupling between a speaker and a microphone and suppressing howling. In the past, for example, the document “A design of canceller for broad band acoustic echo” (Furukawa, “A Design of Canceller for B
road Band Acoustic Echo, "Proceedings of Internatio
nal Teleconference Symposium, PP.232−239 (April 19
The echo signal canceller described in 84)) has been proposed.

Hereinafter, a conventional echo signal canceller as described above will be described with reference to the drawings.

FIG. 7 is a diagram showing a communication conference system using a conventional echo signal canceller. In FIG. 7, 1 and 2 are echo cancellers, 3, 4, 5 and 6 are amplifiers, and 7, 8 is a speaker, 9 and 10 are microphones, 11 is a speaker A, 12 is a speaker B, 13 is a conference room A, and 14 is a conference room B.

The voice spoken by the speaker A11 in the conference room A13 is transmitted from the microphone 9 to the speaker 8 via the amplifier 3 and the amplifier 6. The voice signal output from the speaker 8 in the conference room B14 is the speaker B.
At the same time as being transmitted to the speaker 12, it is returned to the speaker A11 as an echo signal delayed from the speaker 7 through the amplifier 5 and the amplifier 6 by the propagation delay time of the signal from the speaker 7. When such an echo signal returns, a natural conference call is disturbed. Further, the voice signal output from the speaker 7 also wraps around the microphone 9, so that the voice signal spoken by the speaker A forms a loop. When the loop gain exceeds 1, howling occurs, which greatly disturbs the conference call. Therefore, in the communication conference system shown in FIG. 7, echo signals are eliminated by using echo signal elimination devices 1 and 2, and howling is suppressed. That is, the echo signal of the speaker A11 is canceled by the echo signal canceller 2 on the far end side of the speaker A11, and the echo signal of the speaker B12 is canceled by the speaker B12.
By erasing with the echo signal erasing device 1 on the far end side of 12, it is possible to realize a communication conference system capable of two-way conference call with natural sound quality.

Considering a case where the sampling frequency is 16 KHz and the echo signal duration is 250 mS as an example of the echo signal canceller used in such a communication conference system, the number of taps of the filter is 4,000 taps. In order to process such a filter with a large number of taps in real time, it is necessary to divide the processing, and an echo signal canceller using a band division type echo canceller is proposed in the above-mentioned document. The advantage of using the band-division type echo canceller is that the same processing can be performed on the signals in each band by dividing the frequency band of the input signal and then converting it to a low frequency. .

FIG. 8 shows a configuration diagram of a conventional echo signal canceller using the above-described band division type echo cancellation. In FIG. 8, reference numeral 15 is a receiving side input signal terminal, 16 is a receiving side output signal terminal, 17 is a transmitting side input signal terminal, 18 is a transmitting side output signal terminal, and 19 is a plurality of transmitting side input signals (n pieces). A first band divider for dividing into frequency bands, 20 is a second band divider for dividing a receiving side input signal into n frequency bands having the same characteristics as the first band divider, and 211 to 21n are the above-mentioned The first band signal obtained by the first band divider and the first band signal obtained by the second band divider in the same band;
, An echo canceller for canceling the echo signal contained in each band signal obtained by the band divider, 22 is a transmission side output signal in which the echo signal is canceled by combining the band signals whose echo signals have been canceled by the echo canceller Is a band combiner. FIG. 9 is an internal block diagram of the echo canceller. In FIG. 9, 23 is a band signal terminal of a receiving side input signal, 24 is a band signal terminal of a transmitting side input signal, 25 is a band signal terminal of a transmitting side output signal, 26 is a band signal of the receiving side input signal and An adaptive filter comprising a variable coefficient transversal filter for creating an estimated band signal of an echo signal included in the band signal of the transmission side input signal based on the band signal of the transmission side input signal, 27 is the transmission side input It is a subtractor that subtracts the estimated band signal of the echo signal from the band signal of the signal to create the band signal of the transmission-side output signal from which the echo signal has been eliminated.

The operation of the echo signal canceller configured as described above will be described below.

First, the input signal on the transmitting side and the input signal on the receiving side are divided into n frequency bands by the first band divider 19 and the second band divider 20, respectively, to obtain band signals. Next is an echo canceller
Estimate the echo signal included in each band signal of the input signal on the transmission side using 211 to 21n, and cancel the echo signal by subtracting each estimated band signal of the echo signal from each band signal of the input signal on the transmission side Each band signal of the transmitted output signal is obtained. Then, the band signals of the transmission side output signal from which the echo signal has been eliminated are combined using the band combiner 22 to create a transmission side output signal from which the echo signal has been eliminated.

Next, the internal operation of the echo cancellers 211 to 21n will be described with reference to FIG. Echo Canceller 211-2
In 1n, using the adaptive filter 26, the band signal xj of the input signal on the receiving side and the band signal zj of the input signal on the transmitting side at time j are used.
Based on and, the impulse response (corresponding to the filter coefficient of the adaptive filter) between the speaker and the microphone (strictly speaking, between the band signal terminal 23 of the receiving side input signal and the band signal terminal 24 of the transmitting side input signal) h _ij (i , 1, 2, ..., N, N is the number of impulse response samples), and the estimated impulse response _ij (∧ represents an estimated value) and the band signal of the receiving side input signal An estimated band signal _j of the echo signal is created by performing convolution (filtering with the filter coefficient) as shown in equation (1).

Next, a subtracter 27 is used to subtract the estimated band signal of the reverberation signal from the band signal of the reception side input signal as shown in the equation (2), whereby the band signal of the transmission side output signal from which the reverberation signal has been eliminated. Create e _j .

e _j = z _j − _j (2) Impulse response is estimated when the input signal on the transmitting side is generated only by the echo signal of the input signal on the receiving side, that is, at the far end when viewed from the echo signal canceller. Only when someone is speaking. The impulse response is estimated, for example, in (3)
This is done by modifying the impulse response every sample period according to the learning identification algorithm shown in the equation.

However, α is a constant such that 0 <α <2.

As described above, in the communication conference system using the echo signal canceller, a natural conference call in which the echo signal is cancelled can be performed.

Problems to be Solved by the Invention However, the above-mentioned configuration has the following problems.

Generally, the ambient noise level in the room is high in the low frequency region and decreases as it goes to the high frequency region, as shown in FIG. 10 (a). In the learning identification algorithm, the echo signal cancellation amount A _CANC (dB) when the signal-to-noise ratio is S / N (dB) is given by equation (4).

A _CANC = S / N + 10log ₁₀ (2 / α-1) (4) Since ambient noise acts as noise, the echo signal level in a room with room noise characteristics shown in Fig. 10 (a) is eliminated. The characteristics as shown in FIG. 10 (b) are shown before and after the device is operated. That is,
In the conventional configuration, the amount of echo signal cancellation differs depending on the frequency, and the howling margin is determined by the minimum amount of echo signal cancellation.

Further, in the conventional echo signal canceller, howling may occur when the direction of the microphone is significantly changed and the acoustic coupling characteristic between the speaker and the microphone is drastically or drastically changed. Conventionally, these problems have been dealt with by limiting the usage method and usage environment.

In view of the above problems, the present invention provides a reverberation signal canceller that improves the howling margin against environmental changes to alleviate restrictions on the use environment and has a wide range of application.

Means for Solving the Problems In order to solve the above problems, the echo signal canceller according to the present invention divides a transmission side input signal into a plurality of frequency bands.
Band divider, a second band divider that divides the input signal on the reception side into a plurality of frequency bands having the same characteristics as the first band divider, and each band obtained by the first band divider. An echo canceller for canceling an echo signal contained in each band signal obtained by the first band divider based on the signal and each band signal of the same band obtained by the second band divider; An attenuator for attenuating each band signal in which the echo signal is canceled by the echo canceller, and a band combiner for combining the band signals attenuated by the attenuator to create a transmission-side output signal in which the echo signal is canceled A short-time power calculator for calculating the short-time power of each band signal obtained by the first band divider, and each band from the short-time power of each band signal obtained by the short-time power calculator Signal No. estimates the noise ratio, and a based on the estimated signal-to-noise ratio attenuation controller to control the attenuation amount of the attenuator.

Operation According to the present invention, the attenuation controller estimates the signal-to-noise ratio of each band signal from the short-time power of each band signal obtained by the first band divider, and the estimated signal By controlling the attenuation amount of the attenuator based on the noise-to-noise ratio, an echo signal cancellation amount adapted to the ambient noise level is given, and further, the attenuation amount controller causes howling from the short-time power of each band signal. If howling occurs by controlling the frequency and controlling the attenuation of the attenuator in the estimated band, howling can be suppressed immediately and the howling margin against environmental changes can be improved. it can.

Embodiment An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an echo signal canceller according to an embodiment of the present invention.

In FIG. 1, 28 is a reception side input signal terminal, 29 is a reception side output signal terminal, 30 is a transmission side input signal terminal, 31 is a transmission side output signal terminal, and 32 is a plurality of transmission side input signals (n). A first band divider for dividing the frequency band into 33, a second band divider 33 for dividing the input signal on the receiving side into n frequency bands having the same characteristics as the first band divider 32, and 341 to 34n are second band dividers. Based on each band signal obtained by the first band divider 32 and each band signal of the same band obtained by the second band divider 33, each band signal obtained by the first band divider 32 Echo cancellers for canceling the echo signals contained therein, 351-35n are attenuators for attenuating the band signals whose echo signals have been canceled by the echo cancellers, and 36 are the band signals attenuated by the attenuators. , A band combiner that creates a transmitter output signal with the echo signal removed, Reference numerals 371 to 37n are short-time power calculators for calculating the short-time power of each band signal obtained by the first band divider 32, and 38 is a short-time power of each band signal obtained by each time power calculator. Is an attenuation amount controller that controls the attenuation amount of each of the attenuators.

The operation of the echo signal canceller configured as described above will be described below.

First, the input signal on the transmitting side and the input signal on the receiving side are respectively divided into n frequency bands as shown in FIG. 2 by using the first band divider 32 and the second band divider 33 to obtain band signals. . In FIG. 2, CH1 to CHn are frequency-divided n
Represents a band signal.

FIG. 3 shows an internal configuration diagram of the first and second band dividers. In FIG. 3, 39 is an input signal terminal, 40 is a low-pass filter, 41 is an analog-digital converter (hereinafter referred to as A / D).
421 to 42n are band pass filters, 43
1 to 43n are low frequency converters, and 441 to 44n are band signal output terminals. The internal operation of the band divider will be described with reference to FIG.

An input signal of the band divider is band-limited to a low frequency by a low pass filter 40 and then converted into a digital signal by an A / D converter 41. Next, the input signal converted into a digital signal is passed through n band pass filters 421 to 42n having the characteristics shown in FIG. 2 to perform frequency band division, and the next low pass converters 431 to 43n. By converting to a low frequency and then down-sampling, n
, Band signals (to be precise, band signals converted to reduction) are obtained.

Next, in the echo cancellers 341 to 34n, the first band based on each band signal of the first band divider 32 and each band signal of the same band of the second band divider 33 obtained as described above. The echo signals included in the band signals of the divider 32 are estimated, and the band signals in which the echo signals are canceled by subtracting the estimated band signals of the echo signals from the band signals of the first band divider 32 are obtained. obtain. The internal operation of the echo canceller is the same as that described in the related art.

Next, in the attenuators 351-35n, the echo canceller 34
Each band signal from which the echo signal has been canceled by 1 to 34n is attenuated by the amount specified by the attenuation amount controller 38.

The band combiner 36 combines the band signals attenuated by the attenuators 351-35n to create a transmission-side output signal from which echo signals have been eliminated.

FIG. 4 shows an internal configuration diagram of the band synthesizer 36. In FIG. 4, 451 to 45n are band signal input terminals, 461 to 46n are high frequency converters, 471 to 47n are band pass filters having the same characteristics as the band pass filter of the band divider, 48 is an adder, and 49 is a digital signal. An analog converter (hereinafter referred to as a D / A converter), 50 is a low-pass filter, and 31 is a transmission side output signal terminal. Fourth
The internal operation of the band synthesizer will be described with reference to the drawings.

First, each band signal input to the band synthesizer 36 is upsampled by the high band converters 461 to 46n and then converted to a high band frequency, and the band signal components required by the next band pass filters 471 to 47n are converted. After the extraction, the adder 48 adds the band signals to combine the output signals on the transmission side. The D / A converter 49 converts the output signal on the transmission side of the digital signal synthesized as described above into an analog signal, removes unnecessary high-frequency signal components with the low-pass filter 50, and then outputs the output signal terminal on the transmission side. Output to 31.

On the other hand, the short-time power calculators 371 to 37n calculate the short-time power p _j of the band signal x _j at the time j obtained by the first band divider 32 according to the equation (5).

However, M is the number of samples in the section where the short-time power is calculated.

The attenuation amount controller 38 controls the attenuation amounts of the attenuators 351-35n based on the short-time power of each band signal obtained as described above.

The operation of the attenuation controller 38, that is, a specific method of controlling the attenuators 351 to 35n from the short-time power of each band signal will be described below.

The first method of controlling the attenuators 351 to 35n from the short-time power of each band signal is to estimate the signal-to-noise ratio of the band signal from the short-time power of each band signal, and calculate the estimated signal-to-noise ratio. Based on this, the attenuators 351 to 35n are controlled so that a band signal having a low signal-to-noise ratio is given a greater amount of attenuation than a band signal having a high signal-to-noise ratio.

A method for estimating the signal-to-noise ratio from the short-time power of a band signal on which noise is superimposed will be described with reference to FIG. In FIG. 5, (a) is a short-time power level of noise, (b) is a short-time power level of voice, (c) is a short-time power level of voice + noise, and each band signal. It is a figure which shows an example of the time change of.

As shown in FIG. 5 (b), since there is always a silent section in the voice, when the noise level is steady as shown in FIG. 5 (a), it is shown in FIG. 5 (c). The estimated value of the short-time power level of noise is the average value of a fixed number of the short-time power levels of a voice signal with noise superimposed over a fixed time (for example, 0.5 seconds to 2 seconds). _{If N} , the estimated value _S of the short-time power level of the signal component is given by the equation (6) from the short-time power level P _{S + N of} the signal on which noise is superimposed. _S = P _{S + N −N} (6) As described above, the signal-to-noise ratio of each band signal is estimated from the short-term power of each band signal, and the echo canceller has a low signal-to-noise ratio. By giving a large attenuation to the band signal with a small amount of echo signal cancellation,
It is possible to obtain an echo signal cancellation amount that is substantially uniform with respect to the frequency, and improve the howling margin.

The second method of controlling the attenuators 351 to 35n from the short-time power of each band signal is to estimate the frequency band in which howling is occurring from the short-time power of each band signal, and to estimate the attenuation amount of the estimated attenuator. It is controlled so as to increase, and is used in combination with the above method.

A method for estimating the frequency band in which howling is occurring will be described with reference to FIG. In FIG. 6, the horizontal axis represents frequency and the vertical axis represents short-time power level. When howling occurs, the short-time power level of each band signal has a large peak in the frequency band in which howling occurs, as shown in FIG. Therefore, by comparing the short-time power level of one band with the short-time power level of its adjacent band in order, and if the difference exceeds a certain threshold, it can be estimated that howling is occurring in that band. it can.

As described above, the frequency band in which howling is generated is estimated from the short-time power of each band signal, and the gain of the band signal that causes the housing is increased by increasing the attenuation amount of the attenuator in the estimated band. It can be reduced and howling can be immediately suppressed.

In the above embodiment, the rectangular window is used as the window for calculating the short-time power, but other windows may be used,
Alternatively, the short-time power may be obtained by using a filter and integrating with respect to time.

EFFECTS OF THE INVENTION As described above, the present invention inserts an attenuator into the band output signal of each echo canceller of a band division type echo canceller, and calculates the short time power of each band signal of the input signal on the transmission side. The signal-to-noise ratio of each band signal is estimated from the calculator and the short-time power of each band signal, and the attenuation amount of the attenuator is controlled based on the estimated signal-to-noise ratio. By estimating the frequency band that is present and providing an attenuator that controls the attenuator in the specified band to increase the amount of attenuation, a sufficient echo signal cancellation amount adapted to the frequency characteristics of the ambient noise level can be obtained. In addition, howling can be suppressed immediately when a housing is generated, and the howling margin with respect to environmental changes can be improved. Therefore, the restrictions on the usage method and usage environment can be remarkably alleviated, and the practical effects thereof are great.

[Brief description of drawings]

FIG. 1 is a block diagram of an echo signal canceller in one embodiment of the present invention, FIG. 2 is a frequency characteristic diagram of the same band divider, FIG. 3 is a block diagram of the same band divider, and FIG. FIG. 5 is a block diagram of a synthesizer, FIG. 5 is a characteristic diagram showing an example of temporal changes in short-time power level of noise, voice, and voice + noise band signals, and FIG. 6 is each band signal when howling occurs. FIG. 7 is a characteristic diagram showing a short-time power level of the device, FIG. 7 is a block diagram showing a communication conference system using an echo signal canceller, FIG. 8 is a block diagram of a conventional echo signal canceller, and FIG. 9 is an echo canceller. A block diagram and FIG. 10 are diagrams showing ambient noise level characteristics in a room and echo signal level characteristics when a conventional echo signal canceller is used in the room. 28 …… Reception side input signal terminal, 29 …… Reception side output signal terminal, 30 …… Transmission side input signal terminal, 31 …… Transmission side output signal terminal, 32 …… First band divider, 33 …… Second 2 band divider, 341-34n ... Echo canceller, 351-35n ... Attenuator, 36 ... Band combiner, 371-37n ... Short-time power calculator, 38 ... Attenuation controller.

Claims (1)

[Claims] 1. A first band divider for dividing a transmission side input signal into a plurality of frequency bands, and a second band divider for dividing a reception side input signal into a plurality of frequency bands having the same characteristics as the first band divider.
The band divider, and the first band divider based on the respective band signals obtained by the first band divider and the respective band signals of the same band obtained by the second band divider. The echo canceller for canceling the echo signal contained in each obtained band signal, the attenuator for attenuating each band signal whose echo signal is canceled by the echo canceller, and each band signal attenuated by the attenuator A band combiner that creates a transmission-side output signal that is synthesized to eliminate echo signals, a short-time power calculator that calculates the short-time power of each band signal obtained by the first band divider, and the short-circuit calculator. An attenuation amount that estimates the signal-to-noise ratio of each band signal from the short-time power of each band signal obtained by the time power calculator, and controls the attenuation amount of the attenuator based on the estimated signal-to-noise ratio Controller To be composed of echo signal canceller according to claim.