JP2003284184A - Echo canceller and echo canceling method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a small-sized echo canceller for canceling an echo at low cost while ensuring a maximum echo path length longer than a prescribed length in a two-way communication system for transceiving a stereophonic signal with a high sampling frequency. <P>SOLUTION: First and second band split means 4, 6 apply band division to a sound signal with a mid-component adopting the M-S (mid-side) system among stereophonic signals with a prescribed sampling frequency, first and second frequency conversion means 8, 10 decrease the sampling frequency of the mid-component sound signal at a low frequency band, an adaptive filter 13 processes the processed sound signal, and third frequency conversion means 20, 21 increase the frequency of an error signal outputted from the adaptive filter 13 to have the original sampling frequency. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo canceller and an echo canceling method for an audio signal, and more particularly to a method suitable for processing a stereo audio signal having a high sampling frequency.

[0002]

2. Description of the Related Art A system is becoming widespread in which persons who are located far from each other can communicate with each other bidirectionally by voice through a network. An example of such an interactive communication system is a video conference system.

FIG. 1 shows a device for voice transmission / reception used in an existing video conference system. At the point A, a monaural microphone 101, a speaker 102, an AD converter 103, a DA converter 104, an echo canceller 105, a voice codec 106, and a network interface 107 are provided.

A microphone 101, a speaker 102, and an AD converter 10 are also provided at a point B apart from the point A.
3, DA converter 104, echo canceller 105,
An audio codec 106 and a network interface 107 are provided.

The sampling frequency of the AD converter 103 is a relatively low frequency of 16 kHz (twice the necessary frequency band of 8 kHz, which is the required band of the voice signal for everyday conversation).

Network interface 10 at point A
7 and the network interface 107 at the point B,
It is connected by a network 108 using a telephone line or the like.

At point A, the speech of the conference attendees is
The audio codec 1 is input to the microphone 101, digitally converted by the AD converter 103 at a sampling frequency of 16 kHz, and then passed through the echo canceller 105.
It is encoded (compressed) in 06 and sent from the network interface 107 to the network interface 107 at the point B via the communication network 108. And
At the point B, the audio codec 106 decodes (decompresses)
Then, after passing through the echo canceller 105, the DA converter 1
The analog conversion is performed at 04, and the analog output is output from the speaker 102.

At point B as well, the voices of participants in the conference are input to the microphone 101 and the AD converter 10 is used.
3 is digitally converted at a sampling frequency of 16 kHz and then encoded by the voice codec 106 through the echo canceller 105, and then the network interface 1
07 to the network interface 107 at the point A via the communication network 108. Then, at the point A, it is decoded by the audio codec 106, passed through the echo canceller 105, converted into analog by the DA converter 104, and output from the speaker 102.

FIG. 2 shows the configuration of a conventional echo canceller 105 in such a video conference system. The echo canceller 105 is composed of an adaptive filter and includes a digital filter 111, an adder / subtractor 112, and a controller 113.

The echo is a sound sent from the other party's point, output from the speaker 102 at his / her own point, and then passes through an unknown system (echo path) 114 including the room space and the room wall. This is a phenomenon in which the user jumps into the microphone 101 and is returned to the other party's point.

The digital filter 111 estimates the transfer function Hk (z) of the echo path 114. Then, the audio signal x (k) with a sampling frequency of 16 kHz sent from the other party's point to the own point is used as a reference signal, and this transfer function Hk (z) and this reference signal x
(K) time series vector [x (k), x (k−1), ...
x (k−N−1)] is convoluted with a target signal to generate a replica y ′ (k) of the input audio signal from the microphone 101.

The adder / subtractor 112 passes through the microphone 101, the AD converter 103, and the echo canceller 105.
The echo is canceled by subtracting the replica y '(k) from the audio signal (target signal) y (k) having the sampling frequency of 16 kHz input to the.

The controller 113 uses the learning identification method according to the magnitude of the error signal e (k) output from the adder / subtractor 112 (the remaining signal obtained by subtracting the replica y '(k) from the target signal y (k)). By updating the tap coefficient of the digital filter 111 by a mathematical method such as the affine projection method or the least square method, the estimation accuracy of the transfer function Hk (z) is improved.

The number of taps of the digital filter 111 is set to the maximum echo path length for a monaural audio signal having a sampling frequency of 16 kHz (which indicates how many milliseconds after the output from the speaker, the audio jumped into the microphone can be canceled as an echo). It corresponds to the number.

The echo canceller 105 is constructed by using, as actual hardware, a DSP (digital signal processor) in which software for operating as an adaptive filter is installed, and a RAM.

As the adaptive filter having the number of taps corresponding to the maximum echo path length for a monaural voice signal having a sampling frequency of 16 kHz like the echo canceller 105, specifically, those using two DSPs and six RAMs respectively. Existing.

[0017]

By the way, the conventional video conference system has a low sampling frequency of an audio signal and transmits and receives a monaural audio signal as shown in FIG. 1. In the future, the video conference system will be described. However, it is conceivable that the sampling frequency of the audio signal will be increased in order to improve the sound quality, and that the stereo audio signal will be transmitted and received in order to obtain a realistic sensation.

Further, in the future, as a new type of interactive communication system for entertainment system, for example, a concert venue and a home will be connected by a network, and the performance sound of a musical instrument or the voice of a vocalist will be sent from the concert venue to the home. There is a possibility that a two-way communication system will appear that sends applause and cheering voices to the concert venue.

Alternatively, a plurality of points are connected by a network, and people at those points experience the same virtual space (for example, the jungle to be explored) as a group, and voices are sent from each point to all the remaining points. There is a possibility that a two-way communication system that allows each person to communicate with each other in the virtual space will appear.

In such a new type of interactive communication system, considering that the required band of the voice signal is 20 kHz which is the same as the required band of the acoustic signal for hi-fi,
Double the sampling frequency of the audio signal, 40 kHz
It is necessary to do the above, and it is necessary to transmit and receive a stereo audio signal in order to obtain a realistic sensation.

However, when the echo canceller 105 having the structure shown in FIG. 2 is used in the two-way communication system for transmitting and receiving the stereo audio signal having the high sampling frequency, the echo canceller becomes large in size and high in cost. There is an inconvenience that it causes

There are two reasons. That is, as a first reason, in an echo canceller using an adaptive filter such as the echo canceller 105 in FIG. 2, since a FIR filter is generally used as a digital filter to identify the system, the maximum echo path length is shown in FIG. As shown
[The number of taps (the number of taps that can actually perform multiplication on each sample signal with the number of taps of the digital filter being the upper limit)] x [1 / FS] (FS is the sampling frequency of the audio signal).

A DSP used as an adaptive filter
Since the processing speed of RAM and RAM is limited, when the echo canceller 105 performs real-time processing, when the sampling frequency FS becomes high, [the number of taps that can actually multiply each sample signal] is proportional to it. Decrease.

Further, as the sampling frequency FS increases, [1 / FS] also decreases in proportion thereto.

Therefore, in an adaptive filter having the same processing capacity (using the same number of DSPs and RAMs having the same specifications), the maximum echo path length decreases in proportion to the square of the sampling frequency FS (for example, sampling). When the frequency FS is increased from 16 kHz to 48 kHz, which is three times, the maximum echo path length becomes 1/9 short).

However, when the maximum echo path length becomes short,
Only the sound that jumps into the microphone after a short time after being output from the speaker can be canceled as an echo, so that the echo cannot be canceled in a slightly wider room. Therefore, in the echo canceller as a product, the maximum echo path length must be a certain length (echo path length of the corresponding room, for example, 300 milliseconds) or more.

Even if the sampling frequency FS becomes high as described above, the processing capacity of the echo canceller 105 must be increased in order to secure the maximum echo path length of a certain length or more. For example, sampling frequency FS
If the maximum echo path length of the same length is to be secured even when the frequency is increased from 16 kHz to 48 kHz, the processing capacity of the echo canceller 105 is increased by 9 times (when a DSP or RAM of the same specifications is used, It is necessary to increase the number of RAMs and RAM 9 times).

Next, as a second reason, when processing the stereo audio signal with the echo canceller 105 of FIG. 2, as shown in FIG. 4, the left channel speaker 122 is used.
The phenomenon that the sound L output from L jumps into the left channel microphone capsule 121L in the left and right two-channel stereo microphone, left channel speaker 1
The phenomenon that the sound L output from 22L jumps into the right channel microphone capsule 121R in this stereo microphone, the phenomenon that the sound R output from the right channel speaker 122R jumps into this left channel microphone capsule 121L, the right channel speaker 12
It is necessary to treat a total of four phenomena in which the sound R output from 2R jumps into the right channel microphone capsule 121R as echoes.

Therefore, in contrast to the case where only the phenomenon in which the sound output from one speaker jumps into one monaural microphone is treated as an echo, the echo canceller 1
4 times the processing capacity of 05 (DSP and R of the same specification)
When AM is used, the number of DSPs and RAMs needs to be quadrupled.

Further, the processing for discriminating between the sound L from the left channel speaker 122L and the sound R from the right channel speaker 122R among the sounds jumped into the left channel microphone capsule 121L, and the right channel microphone capsule 121R are processed. It is also necessary to perform a process of distinguishing the sound L from the left channel speaker 122L and the sound R from the right channel speaker 122R among the jumped sounds.

A technique for making this determination is disclosed in, for example, Japanese Patent Laid-Open No. 10-190848, but it is necessary to add more circuits to the echo canceller 105.

For these two reasons, when processing a stereo audio signal with a sampling frequency of 48 kHz by the echo canceller 105, for example, the sampling frequency 16
The processing capacity of the echo canceller 105 is increased to 9 × 4 = 36 times or more compared to the case of processing a kHz monaural audio signal (when using DSPs or RAMs of the same specifications, the number of DSPs or RAMs is 36 times or more). Will be required).

Specifically, the sampling frequency is 16 kHz.
Echo canceller 10 for processing z monaural audio signal
As described above, 5 and 2 are used for DSP and RAM respectively.
There are 6 and 6 units, so when processing a stereo audio signal with a sampling frequency of 48 kHz, 2 × 36 = 7 DSPs and RAMs with the same specifications are used.
It is necessary to use a large number of two, 6 × 36 = 216 or more. Therefore, the echo canceller becomes very large and expensive.

In view of the above points, the present invention is a two-way communication system for transmitting and receiving a stereo audio signal with a high sampling frequency, while canceling echoes at a small size and at low cost while ensuring a long maximum echo path length of a certain length or more. It is an object of the present invention to provide an echo canceller and an echo canceling method that can be performed.

[0035]

In the present invention, when an adaptive filter is used in an echo canceller, attention is paid to the following two characteristics of voice and echo.

As a first characteristic, as the frequency band becomes higher, the sound has a larger attenuation coefficient in propagation and also a larger absorption coefficient when reflected on a wall of a room or the like. Therefore, most of the low-frequency band component is contained in the sound that is output from the speaker and is delayed by a certain time (for example, 100 milliseconds) and then jumps into the microphone.

From the first feature, it is not necessary to secure a long maximum echo path length of, for example, 300 milliseconds for voices in all frequency bands, but at least for voices in a low frequency band. Become.

As a second feature, the sound output from the left and right two-channel speakers and jumping into the stereo microphones becomes different from the left-channel sound and the right channel as the distance of the echo path between those speakers and the stereo microphone becomes longer. The difference from the channel sound is reduced. Therefore, the sound that is output from those speakers and then reflected by the wall of the room and jumps into the stereo microphone is closer to monaural sound rather than stereo sound. In the case of the stereo system of the microphone, such a sound close to a monaural sound corresponds to a sound of a mid component which is a component in the front direction of the microphone in the MS (mid-side) system.

From the second characteristic, even in a two-way communication system for transmitting and receiving a stereo audio signal,
The processing in the echo canceller does not need to be performed for the audio signals of the respective channels in the left and right two-channel system, but may be performed mainly for the audio signal of the mid component in the MS system.

Therefore, in the echo canceller according to the present invention, of the stereo audio signals of the predetermined sampling frequency sent to the speaker, the audio signal of the MS mid component is converted into the audio signal of the low frequency band and the audio signal of the high frequency band. First band dividing means for band-dividing into an audio signal and first frequency converting means for lowering a sampling frequency of a mid-component audio signal of a low frequency band, which is band-divided by the first band dividing means, at a predetermined rate. And, of the stereo audio signal of this predetermined sampling frequency input from the microphone,
Second band division means for band-dividing the M-S system mid-component voice signal into the low-frequency band voice signal and the high-frequency band voice signal, and the second band division means. The second frequency conversion means for lowering the sampling frequency of the low frequency band mid component audio signal at the predetermined rate, and the low frequency band mid component audio signal output from the first frequency conversion means as reference signals. While being supplied, the low-frequency band mid-component audio signal output from the second frequency conversion means is supplied as a target signal, a replica of this target signal is created from this reference signal, and this replica is created from this target signal. The adaptive filter that outputs the error signal after subtracting and the sampling frequency of the error signal output from this adaptive filter are And so and a third frequency converting means for increasing the sampling frequency.

In this echo canceller, of the stereo audio signals of a predetermined sampling frequency sent to the speaker, the audio signal of the M-S mid component is mixed with the audio signal of the low frequency band and the high frequency by the first band dividing means. The band is divided into a band voice signal and a band voice signal. Then, the audio signal of the low frequency band mid component is input to the adaptive filter as a reference signal after the sampling frequency is lowered by the first frequency conversion means.

Of the stereo audio signals of this predetermined sampling frequency input from the microphone, M-
The S-system mid-component audio signal is band-divided into a low-frequency band audio signal and a high-frequency band audio signal by the second band-dividing means, as in the first band-dividing means. Then, the audio signal of the mid component in the low frequency band is input to the adaptive filter as a target signal after the sampling frequency is lowered by the second frequency converting means at the same rate as that of the first frequency converting means.

In the adaptive filter, a replica of this target signal is created from this reference signal and an error signal obtained by subtracting this replica from this target signal is output in the same manner as described for the echo canceller 105 in FIG. .

At this time, the processing capacity of this adaptive filter required to secure a constant maximum echo path length becomes lower in proportion to the square of the degree of decrease in the sampling frequency for the above-mentioned first reason. There is.

Further, since only the audio signal of the low frequency band mid component is processed, it is not necessary to increase the processing capacity of this adaptive filter by four times as described above as the second reason.

The error signal output from this adaptive filter is output from this echo canceller after the sampling frequency is raised to the same sampling frequency as the input audio signal to the echo canceller by the third frequency conversion means.

According to this echo canceller, even if the stereo audio signal is processed, and even if the sampling frequency of the stereo audio signal is high, the echo canceller maintains a certain level or more while using an adaptive filter having a low processing capability. A long maximum echo path length (for example, 300 milliseconds) can be secured.

Therefore, in a two-way communication system for transmitting and receiving a stereo audio signal having a high sampling frequency, it is possible to cancel an echo at a small size and at low cost while ensuring a long maximum echo path length of a certain value or more.

In this echo canceller, as an example, of the stereo audio signals of the predetermined sampling frequency sent to the speaker, the audio signal of the MS side component is converted into the audio signal of the low frequency band and the audio signal of the high frequency band. Third band dividing means for band-dividing the audio signal into the audio signal of the second audio signal, and fourth frequency conversion for lowering the sampling frequency of the audio signal of the side component of the low frequency band divided by the third band dividing means at a predetermined rate. Means, and of the stereo audio signals of the predetermined sampling frequency input from the microphone, the audio signal of the MS side component is band-divided into the audio signal of the low frequency band and the audio signal of the high frequency band. Fourth band dividing means and sampler of the side signal audio signal of the low frequency band band-divided by the fourth band dividing means Fifth frequency conversion means for reducing the frequency at this predetermined rate, and the audio signal of the side component of the low frequency band output from the fourth frequency conversion means are supplied as a reference signal and the fifth frequency conversion means. The second adaptation which outputs the error signal obtained by subtracting this replica from the target signal by supplying the side signal audio signal of the low frequency band output from the target signal as the target signal. A filter, a sixth frequency conversion means for increasing the sampling frequency of the error signal output from the second adaptive filter to the predetermined sampling frequency, an error signal output from the sixth frequency conversion means, and a third frequency It is preferable to further include addition means for adding the error signal output from the conversion means.

Alternatively, the audio signal of the mid component of the high frequency band which is band-divided by the first band dividing means is supplied as a reference signal, and the mid component of the high frequency band which is band-divided by the second band dividing means. Audio signal is supplied as a target signal, a replica of this target signal is created from this reference signal, and an error signal obtained by subtracting this replica from this target signal is output, and this third adaptive filter It is also preferable to further include an adding unit that adds the error signal output from the third frequency converting unit and the error signal output from the third frequency converting unit.

As described above, the second and third audio signals for processing the side component audio signal of the low frequency band and the mid component audio signal of the high frequency band other than the audio signal of the low frequency band mid component are processed.
By providing the adaptive filter of (2) so to speak, the echo can be canceled even for such a voice signal, so that the echo can be canceled even better.

The maximum echo path length that should be ensured for such a side component voice in the low frequency band and a mid component voice in the high frequency band is secured for the mid component voice in the low frequency band from the characteristics of the voice and echo described above. It may be shorter than the maximum echo path length to be used.

Therefore, since the processing capacity of these auxiliary adaptive filters is low, echo can be canceled in a small size and at low cost.

In this echo canceller, as an example, the left and right two-channel stereo audio signals sent to the speaker are input from the first stereo system conversion means for converting into a stereo system audio signal of the MS system and the microphone. Left and right two-channel stereo audio signal, M-
A second stereo method converting means for converting to an S method stereo audio signal is further provided, and the mid-component audio signal outputted from the first stereo method converting means is band-divided by the first band dividing means. It is preferable that the mid-component audio signal output from the second stereo conversion unit is band-divided by the second band-dividing unit.

As a result, even in a two-way communication system provided with a stereo microphone of an ordinary method (2-channel left / right method) instead of the MS method, the echo can be canceled in a small size and at low cost.

Further, in this echo canceller, as an example, the adaptive filter calculates the time average energy of the reference signal, the target signal and the error signal, and the tap coefficient is updated according to the magnitude of the time average energy. Is preferably included.

In this way, by controlling the update of the tap coefficient according to the magnitude of the time-averaged energy of each of the reference signal, the target signal, and the error signal, the change in the echo path (for example, the change in the direction in which the stereo microphone is directed). ) Or a momentary noise sound (eg door opening / closing sound)
It is possible to cancel the echo by appropriately responding to the occurrence of.

The echo canceller further includes, as an example, an attenuator for attenuating an error signal output from the adaptive filter and equal to or less than a predetermined threshold, and the sampling frequency of the output signal of the attenuator is the third frequency converter. It is preferable to raise by.

As a result, the error signal of a minute level which cannot be canceled by the adaptive filter is attenuated, so that the echo can be canceled more effectively.

Next, in the echo canceling method according to the present invention, of the stereo audio signals of a predetermined sampling frequency sent to the speaker, the audio signal of the MS mid component is converted into the audio signal of the low frequency band. The first step of band-dividing into a high-frequency band audio signal and the low-frequency band mid-component audio signal band-divided in this first step are reduced in sampling frequency at a predetermined rate and supplied to the adaptive filter as a reference signal. Of the stereo audio signal of the predetermined sampling frequency input from the microphone, the audio signal of the MS mid component is converted into the audio signal of the low frequency band and the audio signal of the high frequency band. A third step of band splitting,
In the fourth step of lowering the sampling frequency at the predetermined rate and supplying the target signal to the adaptive filter, the reference signal is used in the reference filter in the low frequency band mid-component audio signal band-divided in the third step. A fifth step of creating a replica of this target signal from the signal and outputting an error signal obtained by subtracting this replica from this target signal, and raising the sampling frequency of the error signal output from this adaptive filter to this predetermined sampling frequency. And 6 steps.

According to this echo canceling method,
Just as the echo canceller according to the present invention described above, in a two-way communication system in which the sampling frequency of a voice signal is high and a stereo voice signal is transmitted and received, a small maximum size and low cost are ensured while ensuring a long maximum echo path length above a certain level. You will be able to cancel the echo.

Also in this echo canceling method, as an example, of the stereo audio signals of the predetermined sampling frequency sent to the speaker, the audio signal of the MS side component is converted into the audio signal of the low frequency band. And a step of band-dividing into a high-frequency band audio signal, and the band-divided low-frequency band side component audio signal,
Lowering the sampling frequency at a predetermined rate and supplying it as a reference signal to another adaptive filter,
Of the stereo audio signal of the predetermined sampling frequency inputted from the microphone, a step of band-dividing the audio signal of the MS side component into the audio signal of the low frequency band and the audio signal of the high frequency band, The step of supplying the audio signal of the side component of the band-divided low frequency band to the other adaptive filter as a target signal by lowering the sampling frequency at the predetermined rate and from the reference signal in the other adaptive filter. Creating a replica of this target signal and outputting an error signal obtained by subtracting this replica from this target signal;
It is preferable to further include a step of adding the error signal output from the other adaptive filter and the error signal whose sampling frequency is increased in the sixth step.

Alternatively, the mid-component audio signal of the high frequency band which is band-divided in the first step is supplied to another adaptive filter as a reference signal, and the mid-component audio signal of the high frequency band which is band-divided in the third step is supplied. The step of supplying this other adaptive filter as a target signal, and the step of creating a replica of this target signal from this reference signal in this another adaptive filter and outputting an error signal obtained by subtracting this replica from this target signal, It is also preferable to further include a step of adding the error signal output from the other adaptive filter and the error signal whose sampling frequency is increased in the sixth step.

As a result, in the same manner as the above-described echo canceller according to the present invention, the echo is canceled even for the audio signal of the side component in the low frequency band and the audio signal of the mid component in the high frequency band. In addition to being cancelable, the processing capacity of these other adaptive filters (auxiliary adaptive filters) is low, so that echoes can be canceled in a small size and at low cost.

[0065]

BEST MODE FOR CARRYING OUT THE INVENTION An example in which the present invention is applied to a two-way communication system for transmitting and receiving a stereo audio signal having a sampling frequency of 48 kHz will be described below with reference to the drawings.

FIG. 5 shows a device for voice transmission / reception used in a two-way communication system to which the present invention is applied, and devices common to FIGS. 1 and 4 are designated by the same reference numerals.

At point A, left and right two-channel stereo microphones 121, speakers 122L, 122R,
An AD converter 51, a DA converter 52, an echo canceller 1, a voice codec 106 and a network interface 107 are provided.

A stereo microphone 121, speakers 122L, 122 are also provided at a point B apart from the point A.
An R / AD converter 51, a DA converter 52, an echo canceller 1, a voice codec 106 and a network interface 107 are provided.

The sampling frequency of the AD converter 51 is 48 kHz (twice or more of 20 kHz which is the necessary band of the acoustic signal for hi-fi).

Network interface 10 at point A
7 and the network interface 107 at the point B,
It is connected by a network 108 using a telephone line or the like.

At the point A, the stereo microphone 1
When voice is input to 21, the stereo microphone 121
The left and right two-channel audio signals from are digitally converted by the AD converter 51 at a sampling frequency of 48 kHz, then encoded (compressed) by the audio codec 106 via the echo canceller 1, and then transmitted from the network interface 107 to the point via the communication network 108. B network interface 107. Then, at the point B, it is decoded (decompressed) by the audio codec 106, passes through the echo canceller 1, and then the DA converter 5
Analog converted by 2 and speaker 122L, 122R
Output from.

Also at the point B, when voice is input to the stereo microphone 121, the stereo microphone 12
The left and right two-channel audio signals from 1 are digitally converted by the AD converter 51 at a sampling frequency of 48 kHz, and then coded by the audio codec 106 via the echo canceller 1 to obtain the network interface 1
07 to the network interface 107 at the point A via the communication network 108. Then, at the point A, it is decoded by the audio codec 106, passed through the echo canceller 1, converted into analog by the DA converter 52, and output from the speakers 122L, 122R.

FIG. 6 shows the configuration of the echo canceller 1. The echo canceller 1 includes the MS converters 2 and 3,
Band division filters 4 to 7, decimators 8 to 11, echo cancellation blocks 12 to 15, and expander /
Gate circuits 16 to 19 and interpolators 20 and 21
And an adder 22, 23 and an LR converter 24.

The left and right two-channel audio signals xL and xR having a sampling frequency of 48 kHz, which are sent from the other party's point to the own point and are input from the audio codec 106 (FIG. 5) to the echo canceller 1, are supplied to the MS converter 2. Together with this, it is output from the echo canceller 1 as it is and sent to the DA converter 52 (FIG. 5).

The sampling frequency 48 input to the echo canceller 1 from the stereo microphone 121 (FIG. 5) at the user's point via the AD converter 51 (FIG. 5).
The left and right channel audio signals yL and yR of kHz are M
It is supplied to the S converter 3.

The MS converters 2 and 3 have the following equations, respectively.
By performing the operations of (1) and (2), the left and right channel audio signals L, R (here, audio signals xL, yL, x
R, yR) is a circuit for converting an audio signal M, S of a mid component and a side component of the MS (mid-side) stereo system. M = (L + R) / 2 (1) S = (LR) / 2α (where 0 <α <1) (2)

Audio signal x converted by the MS converter 2
M and xS are supplied to the band division filters 4 and 5, respectively. Also, the audio signal y converted by the MS converter 3
M and yS are supplied to the band division filters 6 and 7, respectively.

The band division filters 4 to 7 respectively convert the audio signal into the low frequency band audio signal of less than 3 kHz and the audio signal of 3 kHz.
The audio signal of the above high frequency band is divided.

FIG. 7 shows the configuration of the band division filters 4 to 7. In the band division filters 4 to 7, the linear phase type FI is used.
An LP filter (low-pass filter) 31 is composed of an R filter, and the LPF 31 obtains an audio signal in a low frequency band from an input audio signal.

Also, the input voice signal is delayed by the delay circuit 32 to LP.
By delaying the processing time in F31 and subtracting the output signal of the LPF 31 from the output signal of the delay circuit 32 in the adder / subtractor 33, an audio signal in a high frequency band is obtained.

The band-dividing filters 4 to 7 are not configured to perform band-dividing by using LPFs or HPFs (high-pass filters) configured by IIR filters, but are configured as shown in FIG. In order to faithfully reproduce the original audio signal in consideration of reproducing the original audio signal by re-synthesizing the audio signal and the high frequency band audio signal with adders 22 and 23 as described later. This is because

As shown in FIG. 6, the audio signal xM of the mid component of the high frequency band which is band-divided by the band-division filter 4.
H is sent as a reference signal x (k) to the echo cancellation blocks 12 and 14, respectively.

The low-frequency band mid-component audio signal xML band-divided by the band-division filter 4 is decimator 8
Sent to.

The sound signal xSH of the side component of the high frequency band, which has been band-divided by the band-division filter 5, is supplied to the echo cancel blocks 12 and 14 respectively as the reference signal x
Sent as (k).

The audio signal xSL of the side component of the low frequency band, which has been band-divided by the band-division filter 5, is decimator 9
Sent to.

The mid-component audio signal yMH in the high frequency band, which has been band-divided by the band-division filter 6, is sent to the echo cancellation block 12 as the target signal y (k).

The low-frequency band mid-component audio signal yML band-divided by the band-division filter 6 is decimator 1
Sent to 0.

The side component audio signal ySH of the high frequency band, which has been band-divided by the band-division filter 7, is sent to the echo cancellation block 14 as the target signal y (k).

The audio signal ySL of the side component of the low frequency band, which has been band-divided by the band-division filter 7, is the decimator 1
Sent to 1.

The decimators 8 to 11 are circuits that reduce the sampling frequency of the digital signal at a rate of 1/8.

Set the sampling frequency to 48 with the decimator 8.
The low-frequency band mid-component audio signal xML lowered to 1/8 of 6 kHz (2 times 3 kHz, which is the boundary between the low-frequency band and the high-frequency band) is supplied to the echo cancellation blocks 13 and 15 as reference signals. x (k)
Sent as.

The sampling frequency is set to 6k by the decimator 9.
Audio signal of side component of low frequency band lowered to Hz x
SL is sent to the echo cancellation blocks 13 and 15 as a reference signal x (k), respectively.

The decimator 10 sets the sampling frequency to 6
The low-frequency band mid-component audio signal yML lowered to kHz is sent to the echo cancel block 13 as the target signal yML.
Sent as (k).

The decimator 11 sets the sampling frequency to 6
The audio signal ySL of the side component of the low frequency band lowered to kHz is sent to the echo cancellation block 15 by the target signal ySL.
Sent as (k).

FIG. 8 shows the echo cancel block 12-.
The structure of 15 is shown. Echo cancellation block 12-1
5 is an adaptive filter, like the echo canceller 105 of FIG. 2, and includes a digital filter 41, an adder / subtractor 42, and a controller 43.

The digital filter 41 estimates the transfer function Hk (z) of the echo path in the same manner as described for the digital filter 111 of FIG.
(Z) and reference signal x (k) time series vector [
x (k), x (k−1), ... X (k−N−1)] and the replica y ′ of the target signal y (k).
Create (k).

At this time, the ratio of creating the replica y '(k) using the mid-component audio signal as the reference signal x (k) and the side-component audio signal as the reference signal x
As (k), it is made larger than the ratio of creating the replica y ′ (k).

Specifically, the echo cancellation block 1
3, the audio signal xM of the low frequency band mid component
Replica y'where L is the reference signal x (k)
Creation of (k) and audio signal x of side component of low frequency band
Replica y ′ with SL as reference signal x (k)
(K) is created at a ratio of about 10: 1.

The adder / subtractor 42 cancels the echo by subtracting the replica y '(k) from the target signal y (k).

The controller 43 will be described below according to the magnitude of the error signal e (k) (the remaining signal obtained by subtracting the replica y '(k) from the target signal y (k)) output from the adder / subtractor 42. As described above, the reference signal x (k) and the target signal y (k) are based on the ES affine projection method.
And a method of monitoring the energy of the error signal e (k),
The tap coefficient of the digital filter 41 is updated.

In the general affine projection method, the tap coefficient Bk (z) of the digital filter 41 is updated as in the following equation (3).

[0102]

[Equation 1]

In the ES affine projection method, by further considering the coefficient a (z) of the acoustic property of the echo in this affine projection method, the tap coefficient Bk (z) becomes
It is updated as in (4).

[0104]

[Equation 2]

This coefficient a (z) generally decreases exponentially as z increases. That is, the larger the delay amount, the smaller the update amount of the tap coefficient Bk (z).

In the updating algorithm by the controller 43, the reference signal x (k), the target signal y (k) and the error signal e are based on the ES affine projection method.
A coefficient b (k) (0 ≦ b (k) ≦ 1), which is a coefficient that is variably set by monitoring the energy of (k), is introduced.

Then, using this coefficient b (k), the tap coefficient Bk (z) of the digital filter 41 is expressed by the following equation (5).
It is updated like this.

[0108]

[Equation 3]

FIG. 9 shows the setting process of the coefficient b (k) executed by the controller 43 in each sampling cycle.

In this processing, first, the reference signal x (k), the target signal y (k), the time average energy Xe (k), Ye (k), and Ee (k) of the error signal e (k) are respectively obtained. Calculate (step S1).

Here, the exponential time-averaged energy (the signal at the current time point is weighted more than the signal at the subsequent time point) is calculated with the product of the number of taps and the sampling period as the average time.

However, other than this, as a method of calculating the time average energy, a method of using an average time other than the product of the number of taps and the sampling period, and a method of calculating an even (non-weighted) time average energy Can also be mentioned.

Which of these calculation methods is used is
For each individual two-way communication system to which the present invention is applied, the entire system (frequency band of voice signal,
It may be determined in consideration of the sampling period of the audio signal, the characteristics of the microphone and the speaker, etc.).

Following step S1, the reference signal x
The time average energy Xe (k) of (k) is the threshold Xth.
It is determined whether or not (however, Xth is a value between −20 dB and −6 dB) (step S2).

When the level of the audio signals (audio signals xL, xR in FIG. 5) sent from the other party's point to the own point is low, the volume output from the speakers 122L, 122R and jumping into the stereo microphone 121 is reduced. Therefore, in the input sound to the stereo microphone 121, the ratio of the sound other than the sound jumping from the speakers 122L and 122R is dominant.

When the tap coefficient Bk (z) of the digital filter 41 is updated in such a state, the digital filter 4
1 will not converge to a normal value.

Therefore, in step S2, Xe (k)>
By determining whether or not Xth, the speaker 1
It is determined whether or not the sound volume jumping into the stereo microphone 121 from the 22L and 122R becomes a certain level or more.

Here, Xth is changed from -20 dB to -6 dB.
The value of Xth may be determined in consideration of the entire system as in the method of calculating the time average energy described above.

If NO in step S2, the value of b (k) in the sampling period is set to 0, and
Set the count value CC of the internal counter of the controller 43 to 0
(Step S7). Then, the process ends.

By determining the value of b (k) to 0 in this way, the tap coefficient Bk (z) is not updated in the sampling period from the above equation (5).

On the other hand, if YES in step S2, then the product of the time average energy Xe (k) of the reference signal x (k) and the predetermined coefficient k1 is the target signal y.
It is determined whether or not it is larger than the time average energy Ye (k) of (k) (step S3).

Even if the volume jumping into the stereo microphone 121 from the speakers 122L and 122R is more than a certain level, the volume higher than that generated at the user's point is input to the stereo microphone 121 at the same time (for example, his own). At the point, a person uses a stereo microphone 12
1 or you may be making a lot of noise at your point).

When the tap coefficient Bk (z) of the digital filter 41 is updated in such a double talk state, the digital filter 41 also does not converge to a normal value.

Therefore, in step S3, k1.Xe
By determining whether or not (k)> Ye (k), the volume jumping into the stereo microphone 121 from the speakers 122L and 122R is higher than the volume generated at the user's point and being input to the stereo microphone 121. Judge whether it is large or not.

K1 is a coefficient corresponding to the amount by which the sound output from the speakers 122L and 122R is attenuated by the time it jumps into the stereo microphone 121.
The value is set to a value between 0 dB and -3 dB (the value of k1 may be determined in consideration of the entire system).

If NO in step S3, step S3
Proceed to 7. As a result, the tap coefficient Bk (z) is not updated in the sampling period from the above equation (5).

On the other hand, if Yes in step S3, then the time average energy Ee of the error signal e (k)
It is determined whether (k) is larger than a predetermined threshold Eth (step S4).

When the value of Ee (k) is equal to or less than a certain value, it may be considered that the digital filter 41 is optimized (locked). In step S4, Ee (k)> E
By determining whether or not th, it is determined whether or not the digital filter 41 is locked.

If NO in step S4 (if the digital filter 41 is locked), the value of b (k) in the sampling cycle is set to k3 (where k3 is a value between 0.3 and 0.7). ) Is set (step S10). Then, the process ends.

When the digital filter 41 is locked, the error signal e (k) generally becomes smaller as the update amount of the tap coefficient is made smaller.

Therefore, in step S10, the value of b (k) is reduced. Thus, from equation (5) above,
In the sampling cycle, the update amount of the tap coefficient Bk (z) becomes small.

Although k3 is set to a value between 0.3 and 0.7 here, the value of k3 may be determined in consideration of the entire system.

On the other hand, if YES in step S4, the count value CC of the internal counter is the threshold value CCth (however, CC
This count value CC is incremented by 1 on condition that th is less than 256 to 512) (step S5).

Subsequently, it is determined whether or not the count value CC of the internal counter has reached this threshold value CCth (step S6).

Time average energy E of error signal e (k)
When e (k) is larger than a certain value (when the digital filter 41 is not locked), the cause is that the echo path has changed (for example, the orientation of the stereo microphone 121 has changed), and the momentary change. There are two types of noise noises (for example, door opening / closing noise).

However, which of these causes is not distinguishable unless the value of Ee (k) is monitored over a certain period (a period considerably longer than the sampling period).

Therefore, in steps S5 and S6, an internal counter is used to determine whether or not the digital filter 41 remains unlocked for a certain period or longer.

If NO in step S6, the value of b (k) in the sampling cycle is set to 0 (step S9). Then, the process ends.

As a result, the updating of the tap coefficient Bk (z) is stopped from the above equation (5) until the digital filter 41 remains unlocked for a certain period or longer.

On the other hand, if the result of step S6 is YES, the value of b (k) in the sampling cycle is k2 · b (k-
1) + (1-k2) (where k2 is 0.1 to 0.001)
(Value between) (step S8). Then, the process ends.

That is, if the digital filter 41 remains unlocked for a certain period or longer, it is determined that the echo path has changed, and each sampling is performed with k2 as a time constant until the digital filter 41 is locked. The value of b (k) in the cycle is exponentially approximated to 1.

As a result, from the above equation (5), the update amount of the tap coefficient Bk (z) in each sampling cycle gradually increases until the digital filter 41 is locked.

In this way, when the echo path changes, the tap coefficient Bk (z) is gently updated at first, but if the digital filter 41 is still not locked, the digital filter 41 is promptly changed. In order to lock, the update amount of the tap coefficient Bk (z) is increased.

Here, CCth and k2 are 256 respectively.
The values are between 512 and 512 and between 0.1 and 0.001, but these values may be determined in consideration of the entire system.

In the echo cancel block 12, the tap number of the digital filter 41 is set to the sampling frequency 4
The maximum echo path length for a monaural audio signal of 8 kHz is set to about several tens of milliseconds.

In the echo cancel block 13, the tap number of the digital filter 41 is set to the sampling frequency 6
The maximum echo path length for a monaural audio signal of kHz is set to about 300 milliseconds.

In the echo cancel block 14, the tap number of the digital filter 41 is set to the sampling frequency 4
The maximum echo path length for a monaural audio signal of 8 kHz is set to about several tens of milliseconds.

In the echo cancel block 15, the tap number of the digital filter 41 is set to the sampling frequency 6
The maximum echo path length for a monaural audio signal of kHz is about 50 milliseconds.

The echo cancel blocks 12 to 15 are
As the actual hardware, a DSP installed with software for operating as an adaptive filter
And RAM.

As shown in FIG. 6, the output signal of the echo cancel block 12 (sampling frequency 48 k for the mid component of the high frequency band output from the adder / subtractor 42)
The Hz error signal e (k)) is sent to the expander / gate circuit 16.

Output signal of the echo cancel block 13 (error signal e having a sampling frequency of 6 kHz for the low frequency band mid component output from the adder / subtractor 42)
(K)) is sent to the expander / gate circuit 17.

Output signal of the echo cancel block 14 (error signal e having a sampling frequency of 48 kHz for the side component of the high frequency band output from the adder / subtractor 42)
(K)) is sent to the expander / gate circuit 18.

Output signal of the echo cancel block 15 (error signal e having a sampling frequency of 6 kHz for the side component of the low frequency band output from the adder / subtractor 42)
(K)) is sent to the expander / gate circuit 19.

The expander / gate circuits 16 to 19 are
This is a circuit used to reduce residual noise and increase the dynamic range in professional audio equipment, and as shown in FIG. 10, it is a nonlinear input that attenuates and outputs an input signal equal to or less than a threshold th. It has output characteristics.
In the figure, D is the amount of gain reduction.

The expander / gate circuits 16 to 19 are
Instead of changing the gain instantaneously, the gain is gradually changed with a time constant as shown in the following equation (6) (G
(T) is a gain, A is a set value, and k is a time constant). That is, the gain is exponentially changed with respect to the set value, like the coefficient b (k) in step S8 of FIG. 9 described above. G (t) = A * k + G (t-1) * (1-k) ...
(6)

Here, the set value when the gain is lowered (when the expander / gate circuit is turned on) is about 10
When gain is increased for 0 ms (signal level is threshold th
(When it returns to the linear input / output characteristic because it exceeds the limit), the setting value is about 5 milliseconds, and the gain reduction amount is 6d.
Although it is set to a value between B and 12 dB, these may be determined in consideration of the entire system.

The high-frequency band mid-component audio signal xMH band-divided by the band-division filter 4 and the high-frequency side component audio signal xSH band-divided by the band-division filter 5 are expanded / gated. Circuit 1
The signals 6 and 18 are also given as control signals.

The decimator 8 lowers the sampling frequency to 6 kHz, and the low frequency band mid-component audio signal xML, and the decimator 9 changes the sampling frequency to 6 kHz.
The audio signal xSL of the side component of the low frequency band lowered to kHz is also given as a control signal to the expander / gate circuits 17 and 19, respectively.

In the expander / gate circuits 16 to 19, based on these control signals, for example, when an audio signal is being sent from the other party's point to his / her own point (when no echo occurs), the input signal is input. When the voice signal is output as it is from the other party's point to the other party's point (when an echo is generated), the input signal is attenuated by the input / output characteristic of FIG. 9 and then output. .

As shown in FIG. 6, the output signal of the expander / gate circuit 16 (error signal having a sampling frequency of 48 kHz for the mid component of the high frequency band) is sent to the adder 22.

The output signal of the expander / gate circuit 17 (error signal having a sampling frequency of 6 kHz for the low frequency band mid component) is sent to the interpolator 20.

The output signal of the expander / gate circuit 18 (error signal having a sampling frequency of 48 kHz for the side component in the high frequency band) is sent to the adder 23.

The output signal of the expander / gate circuit 19 (error signal having a sampling frequency of 6 kHz for the side component in the low frequency band) is sent to the interpolator 21.

The interpolators 20 and 21 are circuits for increasing the sampling frequency of the digital signal at a rate of 8 times.

The error signal for the low frequency band mid component whose sampling frequency is raised to 48 kHz, which is 8 times the 6 kHz frequency, by the interpolator 20 is sent to the adder 22 and the error signal for the high frequency band mid component is sent. Is added.

The error signal for the side component in the low frequency band whose sampling frequency has been raised to 48 kHz by the interpolator 21 is sent to the adder 23 and added to the error signal for the side component in the high frequency band.

The output signal of the adder 22 (error signal for the mid component) and the output signal of the adder 23 (error signal for the side component) are sent to the LS converter 24.

The LS converter 24 uses the following equations (7), (8)
By performing the calculation of, the audio signals M and S (here, error signals) of the mid and side components of the MS stereo system are converted into the left and right audio signals L and R of the left and right two-channel stereo system. It is a circuit to convert. L = M + αS (where 0 <α <1) (7) R = M-αS (8)

The error signals eL and eR for the left and right channels converted by the LS converter 24 are
The audio codec 106 is output from the echo canceller 1.
(Fig. 5).

Next, how the echo canceller 1 cancels the echo will be described.

[0171] The left and right two-channel audio signals xL and xR with a sampling frequency of 48 kHz, which are sent from the other party's point to their own point and are input from the audio codec 106 to the echo canceller 1, are transmitted by the MS converter 2 to the MS stereo system mid signal. Component, side component audio signal xM, xS
Is converted to.

Of these, the mid component audio signal xM is
Audio signal xML in the low frequency band of less than 3 kHz and audio signal xM in the high frequency band of 3 kHz or more by the band division filter 4.
The band is divided into H and H. The low frequency band mid-component audio signal xML is input to the echo cancellation block 13 as the reference signal x (k) after the sampling frequency is lowered to 6 kHz by the decimator 8.

Also, the left and right two-channel audio signals yL and yR having a sampling frequency of 48 kHz input from the stereo microphone 121 at the user's point are converted by the MS converter 3 into the mid-side audio signal yM of the MS stereo system and the side-audio signals yM and yR. converted to yS.

Of these, the mid-component audio signal yM is
Audio signal yML in the low frequency band of less than 3 kHz and audio signal yM in the high frequency band of 3 kHz or more by the band division filter 6.
The band is divided into H and H. Then, the audio signal yML of the mid component in the low frequency band is input to the echo cancel block 13 as the target signal y (k) after the sampling frequency is lowered to 6 kHz by the decimator 10.

In the echo cancel block 13, a replica y '(k) of this target signal y (k) is created from this reference signal x (k), and this replica y' (k) is created from the target signal y (k). By subtracting, the echo is canceled at the maximum echo path length of about 300 milliseconds for the low frequency band mid-component audio signal.

Here, voice and echo have the following two characteristics. First, as a first feature, as the frequency band of a sound increases, the attenuation coefficient in propagation increases and the absorption coefficient in reflection on a wall of a room also increases. Therefore, most of the low-frequency band component of the sound that is output from the speakers 122L and 122R and that is delayed by a certain time (for example, 100 milliseconds) or more and then jumps into the stereo microphone 121.

From the first feature, it is not necessary to secure a long maximum echo path length of, for example, 300 milliseconds for voices in all frequency bands, but at least for voices in a low frequency band. Become.

As a second feature, the speaker 122
L, 122R output from stereo microphone 121
The difference between the left-channel sound and the right-channel sound in the sound that jumps in becomes smaller as the distance of the echo path between the speakers 122L and 122R and the stereo microphone 121 becomes longer. Therefore, the speaker 122
The sound output from L and 122R and reflected by the wall of the room and jumping into the stereo microphone 121 is closer to monaural sound rather than stereo sound. In the case of the stereo system of the microphone, such a sound close to a monaural sound corresponds to a mid-component sound which is a front component of the microphone in the MS (mid-side) system, as shown in FIG.

From the second feature, even in a two-way communication system for transmitting and receiving a stereo audio signal,
The processing in the echo canceller does not need to be performed for the audio signals of the respective channels in the left and right two-channel system, but may be performed mainly for the audio signal of the mid component in the MS system.

In the echo canceller 1, paying attention to these characteristics of voice and echo, the echo cancel block 13 cancels the echo of the voice signal of the mid component in the low frequency band with the maximum echo path length of about 300 milliseconds. I am trying.

At this time, as described above, the controller 43 causes the time average energy Xe of each of the reference signal x (k), the target signal y (k), and the error signal e (k).
By controlling the update of the tap coefficient Bk (z) according to the magnitudes of (k), Ye (k), and Ee (k), it is possible to accurately change the echo path and generate a momentary noise sound. Correspondingly, the echo can be canceled.

Further, as described above, the replica y '(k) having the audio signal xSL of the side component of the low frequency band as the reference signal x (k) is also created at a ratio of about 1/10, so that the mid The side component echo, which remains when only the component echo is canceled, is also canceled to some extent.

Further, as described above, the error signal output from the echo cancellation block 13 that is equal to or less than the threshold value th is attenuated by the expander / gate circuit 17, so that the minute level error signal that cannot be canceled by the adaptive filter is attenuated. This makes it possible to cancel the echo better.

Further, in addition to the low-frequency mid-component audio signal, the high-frequency mid-component audio signal, the low-frequency side component audio signal, and the high-frequency side component audio signal are, so to speak, auxiliary. The echo canceling blocks 12, 14 and 15 provided in the above cancel the echoes with the maximum echo path lengths of about tens of milliseconds, tens of milliseconds and 50 milliseconds, respectively, and then expander / gate circuits 16, 18, 19 respectively. Since the minute level error signal is attenuated by this, the echo can be canceled more effectively from that point as well.

The error signal for the mid component of the low frequency band output from the expander / gate circuit 17 has a sampling frequency of 48 at the interpolator 20.
The frequency is raised to kHz, sent to the adder 22, added with the error signal for the mid component of the high frequency band, and converted to the error signal for the left channel and the right channel together with the error signal for the side component by the LS converter 24. , Output from the echo canceller 1.

In this way, this echo canceller 1
Then, in a two-way communication system that transmits and receives a stereo sound signal with a sampling frequency of 48 kHz (a two-way stereo microphone 121 is provided as a microphone, a two-way communication system), a low frequency band mid-component sound signal is mainly used. , The echo is canceled at the maximum echo path length of about 300 milliseconds.

Next, the scale and cost of the hardware of the echo canceller 1 are set to the sampling frequency of 48 kHz.
The description will be made in comparison with the case where the echo canceller 105 as shown in FIG.

The first item in the [Problems to be solved by the invention] column
As described above, the processing capacity of the adaptive filter required to secure a constant maximum echo path length is low in proportion to its square when the sampling frequency is low.

Comparing the echo cancel block 13 and the echo canceller 105, the maximum echo path lengths are both about 300 milliseconds, but the echo cancel block 13 has a sampling frequency of 6 kHz, while the echo canceller 105 has a sampling frequency of about 6 ms. 1
Since it is 6 kHz, the sampling frequency of the echo cancellation block 13 is lower by 6/16.

Further, since the echo cancellation block 13 processes only the audio signal of the low frequency band mid component,
As described as the second reason in the [Problems to be solved by the invention] column, it is not necessary to increase the processing capacity of the echo cancellation block 13 four times.

Therefore, the echo cancel block 1
3 is 36/256 compared with the echo canceller 105
Since it requires only a low processing capacity, a small number of DSP and RA
It is made compact and low cost by using M.

Also, the echo cancel block 15 is
The maximum echo path length is 1 in echo cancel block 13.
Since it is about 50 milliseconds of / 6, the processing capacity is further lower than that of the echo cancellation block 13, and therefore the size and cost of the echo cancellation block 13 are further reduced.

Also, the echo cancel block 12, 1
The sample No. 4 has a sampling frequency of 48 kHz and is three times as high as the echo canceller 105. Therefore, if the maximum echo path length that is the same as that of the echo canceller 105 is secured, the processing capacity of the echo cancel blocks 12 and 14 must be 9 times higher than that of the echo canceller 105.

However, the echo cancel block 12,
The maximum echo path lengths of 14 are several tens of milliseconds, which is about 1/10 to a fraction of the echo canceller 105. Therefore, the echo cancellation block 1
Since 2 and 14 also require processing power equal to or lower than that of the echo canceller 105, they are also small in size and low in cost using a small number of DSPs and RAMs.

On the other hand, the sampling frequency is 48 kHz.
When the echo canceller 105 is used in a two-way communication system for transmitting and receiving a stereo audio signal of z, the second problem is set in the [Problems to be solved by the invention] column.
As mentioned above, the processing capacity of the echo canceller 105 is increased to 36 times or more (a DSP or R having the same specifications).
When AM is used, the number of DSPs and RAMs needs to be 36 times or more), resulting in extremely large size and high cost.

Therefore, this echo canceller 1
Compared with the case where the echo canceller 105 is used in a two-way communication system that transmits and receives a stereo audio signal having a sampling frequency of 48 kHz, even if the portions other than the echo cancellation blocks 12 to 15 are included,
Much smaller and less expensive to construct.

As described above, according to the echo canceller 1, in a two-way communication system for transmitting and receiving a stereo audio signal having a sampling frequency of 48 kHz, a long maximum echo path length of about 300 milliseconds is ensured, and the size and cost are low. You can cancel the echo.

Therefore, for example, when the two-way communication system is constructed as a video conference system (in the video conference system, the sampling frequency of the audio signal is increased to improve the sound quality, and the stereo audio signal is transmitted / received in order to obtain a sense of presence. In this case, while ensuring a long maximum echo path length,
The echo can be canceled in a small size and at low cost.

Alternatively, when this interactive communication system is constructed as a new interactive communication system of entertainment type as described in the section [Problems to be solved by the invention],
It is possible to cancel the echo at a small size and at low cost while ensuring a long maximum echo path length.

In the above example, the MS converters 2 and 3 are provided in the echo canceller 1. However, when the echo canceller 1 is used in a two-way communication system provided with an MS stereo microphone, M
The S converters 2 and 3 may be omitted.

In the above example, the audio signal is divided into a low frequency band of less than 3 kHz and a high frequency band of 3 kHz or more. However, it is divided into a low frequency band and a high frequency band with a frequency other than 3 kHz as a boundary. Good.

Alternatively, the audio signal may be transmitted in three or more bands (for example, a band of 12 kHz or more, a band of 6 kHz or more and 12 k).
Band below 3 Hz, band above 3 kHz and below 6 kHz, 3
The frequency band may be divided into four bands of less than kHz), and the lower the frequency band, the larger the sampling frequency and the longer the maximum echo path length to cancel the echo.

By doing so, it becomes possible to cancel the echo more finely.

Further, although the present invention is applied to the two-way communication system in the above example, the present invention may be applied to cancel howling at a concert hall or the like.

Further, the present invention is not limited to the above examples, and it goes without saying that various other configurations can be adopted without departing from the gist of the present invention.

[0206]

As described above, according to the present invention, in a two-way communication system for transmitting and receiving a stereo audio signal having a high sampling frequency, a small maximum size and low cost echo can be ensured while ensuring a certain maximum long echo path length. The effect of being able to cancel is obtained.

Further, by canceling the echoes of the side component audio signal of the low frequency band and the audio signal of the mid component of the high frequency band by the auxiliary adaptive filter,
There is an effect that the echo can be canceled better and the echo can be canceled more compactly and at low cost.

Also, by converting the stereo audio signal of the left and right two-channel system into the stereo audio signal of the M-S system, the two-way communication system provided with the stereo microphone of the normal system (two-channel system of the left and right) is small. Moreover, the effect that the echo can be canceled at low cost can be obtained.

Further, by controlling the updating of the tap coefficient according to the magnitude of the time average energy of each of the reference signal, the target signal and the error signal, it is also suitable for the change of the echo path and the generation of a momentary noise sound. It is possible to obtain the effect that the echo can be canceled in a reliable manner.

Further, by attenuating the error signal output from the adaptive filter, which is equal to or less than the predetermined threshold value, the effect that the echo can be canceled more effectively can be obtained.

[Brief description of drawings]

FIG. 1 is a diagram showing a device for voice transmission / reception used in an existing video conference system.

FIG. 2 is a diagram showing a configuration of a conventional echo canceller.

FIG. 3 is a diagram showing a maximum echo path length of an echo canceller using an adaptive filter.

FIG. 4 is a diagram showing a phenomenon that needs to be treated as an echo when a stereo audio signal is processed by a conventional echo canceller.

FIG. 5 is a diagram showing a device for voice transmission / reception used in a two-way communication system to which the present invention is applied.

6 is a diagram showing a configuration of the echo canceller of FIG.

7 is a diagram showing a configuration of the band division filter of FIG.

8 is a diagram showing a configuration of an echo cancel block of FIG.

9 is a flowchart showing a process executed by a controller in the echo cancel block of FIG.

10 is a diagram showing input / output characteristics of the expander / gate circuit of FIG.

FIG. 11 is a diagram showing an MS method of a stereo microphone.

[Explanation of symbols]

1 Echo canceller, 2, 3 MS converter,
4 to 7 band division filter, 8 to 11 decimator,
12-15 Echo cancellation block, 16-19
Expander / gate circuit, 20, 21 interpolator, 22, 23 adder, 24 LR converter, 41 digital filter, 42 adder / subtractor, 4
3 controller, 51 AD converter, 52
DA converter, 105 audio codec, 107
Network interface, 108 network, 121 stereo microphone, 122L, 1
22R speaker

Claims (9)

[Claims] 1. A stereo audio signal of a predetermined sampling frequency sent to a speaker, an audio signal of a mid component of the MS (mid-side) system, an audio signal of a low frequency band and an audio signal of a high frequency band. A first band-dividing means for band-dividing into a plurality of bands, first frequency-converting means for lowering a sampling frequency of a mid-component audio signal of a low-frequency band band-divided by the first band-dividing means at a predetermined rate, Of the stereo audio signal of the predetermined sampling frequency input from the second band for dividing the audio signal of the M-S mid component into the audio signal of the low frequency band and the audio signal of the high frequency band. Dividing means and the sampling frequency of the audio signal of the mid component of the low frequency band divided by the second band dividing means at the predetermined rate. The second frequency conversion unit and the low-frequency band mid-component audio signal output from the first frequency conversion unit are supplied as a reference signal, and the low-frequency band output signal from the second frequency conversion unit is supplied. An audio signal of a mid component is supplied as a target signal, a replica of the target signal is created from the reference signal, an adaptive filter that outputs an error signal obtained by subtracting the replica from the target signal, and the adaptive filter output from the adaptive filter. An echo canceller, comprising: a third frequency conversion means for increasing the sampling frequency of the error signal to the predetermined sampling frequency. 2. The echo canceller according to claim 1, wherein among the stereo audio signals of a predetermined sampling frequency sent to the speaker, an audio signal of a side component of the MS system is used as an audio signal of a low frequency band. Third band division means for performing band division into an audio signal of a high frequency band, and a fourth frequency reduction means for lowering a sampling frequency of the side component audio signal of the low frequency band divided by the third band division means at a predetermined rate. Of the stereo audio signals of the predetermined sampling frequency input from the microphone, the audio signal of the side component of the MS system is converted into the audio signal of the low frequency band and the audio signal of the high frequency band. Fourth band dividing means for band-dividing, and a sample of the audio signal of the side component of the low frequency band band-divided by the fourth band dividing means Fifth frequency conversion means for lowering the frequency at a predetermined rate, an audio signal of a side component of a low frequency band output from the fourth frequency conversion means is supplied as a reference signal, and the fifth frequency conversion means is provided. A second frequency component side audio signal output from the means is supplied as a target signal, a replica of the target signal is created from the reference signal, and an error signal obtained by subtracting the replica from the target signal is output. An adaptive filter; sixth frequency conversion means for increasing the sampling frequency of the error signal output from the second adaptive filter to the predetermined sampling frequency; and the error signal output from the sixth frequency conversion means, And an adder that adds the error signal output from the third frequency converter. And an echo canceller. 3. The echo canceller according to claim 1, wherein the audio signal of the mid component of the high frequency band, which is band-divided by the first band-dividing means, is supplied as a reference signal, and the second band-division is performed. A high-frequency mid-component audio signal band-divided by the means is supplied as a target signal, creates a replica of the target signal from the reference signal, and outputs an error signal obtained by subtracting the replica from the target signal; And the error signal output from the third adaptive filter,
An echo canceller further comprising: an addition unit that adds the error signal output from the third frequency conversion unit. 4. The echo canceller according to claim 1, further comprising: first stereo system conversion means for converting left and right two-channel stereo sound signals sent to the speaker into M-S system stereo sound signals. It further comprises a second stereo method converting means for converting the left and right two-channel stereo audio signals inputted from the microphone into a M-S method stereo audio signal, and the mid component of the mid component outputted from the first stereo method converting means. An echo characterized in that an audio signal is band-divided by the first band-dividing means, and a mid-component audio signal output from the second stereo-system converting means is band-divided by the second band-dividing means. Canceller. 5. The echo canceller according to claim 1, wherein the adaptive filter calculates a time average energy of each of the reference signal, the target signal, and the error signal, and has a magnitude of the time average energy. Echo canceller according to the present invention. 6. The echo canceller according to claim 1, further comprising an attenuator for attenuating the error signal output from the adaptive filter and having a predetermined threshold value or less, wherein the sampling frequency of the output signal of the attenuator is the first An echo canceller characterized by being raised by the frequency conversion means of 3. 7. A stereo audio signal of a predetermined sampling frequency sent to a speaker, an audio signal of a mid component of the MS (mid-side) system, an audio signal of a low frequency band and an audio signal of a high frequency band. A first step of band-dividing into a band, and a second step of lowering the sampling frequency of the low-frequency band mid-component audio signal band-divided in the first step at a predetermined rate and supplying it to the adaptive filter as a reference signal, A third step of band-dividing an audio signal of a mid component of the MS system from the stereo audio signal of the predetermined sampling frequency inputted from the microphone into the audio signal of the low frequency band and the audio signal of the high frequency band. And a sampling frequency of the low-frequency mid-component audio signal band-divided in the third step A fourth step of reducing the rate and supplying it as a target signal to the adaptive filter; and, in the adaptive filter, creating a replica of the target signal from the reference signal and outputting an error signal obtained by subtracting the replica from the target signal. And a sixth step of increasing the sampling frequency of the error signal output from the adaptive filter to the predetermined sampling frequency. 8. The echo canceling method according to claim 7, wherein among the stereo audio signals having a predetermined sampling frequency sent to the speaker, an audio signal of a side component of the MS method is used as an audio signal in a low frequency band. A step of band-splitting the signal into a high-frequency band audio signal, and lowering the sampling frequency of the band-divided low-frequency band side component audio signal at a predetermined rate, and referring to an adaptive filter different from the adaptive filter. Supplying as a signal, of the stereo audio signals of the predetermined sampling frequency input from the microphone, the audio signal of the side component of the MS system, the audio signal of the low frequency band and the audio signal of the high frequency band. Band-dividing the audio signal of the side component of the low frequency band Lowering the pulling frequency at the predetermined rate and supplying it as a target signal to the another adaptive filter; and in the other adaptive filter, creating a replica of the target signal from the reference signal, and replicating the target signal from the target signal. And a step of adding the error signal output from the another adaptive filter and the error signal whose sampling frequency has been increased in the sixth step. Echo canceling method. 9. The echo canceling method according to claim 7, wherein the audio signal of the mid component of the high frequency band, which is band-divided in the first step, is supplied to an adaptive filter different from the adaptive filter as a reference signal. Together with the step of supplying, as a target signal, the mid-component audio signal of the high frequency band that has been band-divided in the third step, in the another adaptive filter, a replica of the target signal from the reference signal. And outputting an error signal obtained by subtracting the replica from the target signal, adding the error signal output from the another adaptive filter, and the error signal increased in sampling frequency in the sixth step. Echo canceling method, further comprising: .