JP3420705B2 - Echo suppression method and apparatus, and computer-readable storage medium storing echo suppression program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an echo suppression method and apparatus and a computer-readable storage medium in which an echo suppression program is stored. For example, in a two-wire to four-wire conversion system and a voice communication system, howling The present invention relates to a technique of suppressing an echo signal that causes a cause and a hearing problem and emphasizing a speaker voice signal.

[0002]

2. Description of the Related Art First, the cause of such howling and an echo signal which is an auditory obstacle will be described with reference to a voice communication system shown in FIG. In this figure, reference numerals 1 and 3 are microphones for transmission, 2 and 4 are reception speakers,
5, 7 are transmission signal amplifiers, 6 and 8 are reception signal amplifiers, 9
Is a transmission line, 10 is a transmitter, and 11 is a receiver.

The transmitted voice uttered by the transmitter 10 is transmitted to the receiver 11 via the transmitting microphone 1, the transmitter signal amplifier 5, the transmission line 9, the receiver signal amplifier 8 and the receiver speaker 4. Unlike the conventional telephone call system, such a loud voice call system does not require a handset to be held in the hand, so that it is possible to make a call while working, and also to realize a natural face-to-face call. , Is widely used for communication conferences, videophones, loudspeakers, etc.

However, as a drawback of the voice call system,
The existence of echo is a problem. That is, the voice transmitted from the speaker 4 to the receiving side is received by the microphone 3, and is transmitted to the side of the speaker 10 via the transmitting signal amplifier 7, the transmission path 9, the receiving signal amplifier 6 and the speaker 2. For the speaker 10, this phenomenon is an echo phenomenon in which the voice uttered by the speaker 10 is reproduced from the speaker 2, and is called an acoustic echo or the like. This echo phenomenon causes a bad influence such as a call trouble or discomfort in a voice call system.

Further, the voice uttered from the speaker 2 is received by the microphone 1 and a closed loop of the voice signal is formed. When the loop gain of the closed loop is larger than 1, the howling phenomenon occurs and the call is disabled.

An echo canceller is used to overcome the problems of the voice communication system. FIG. 9 is a block diagram showing a configuration example of a conventional full-band echo canceller. In this figure, the same components as those in FIG. 8 are designated by the same reference numerals.

In the figure, reference numeral 21 is an echo canceller, 22 is a pseudo echo path convolution operation circuit, 23 is an echo path estimation circuit, 24 is a subtracter, and 25 is a nonlinear echo suppression processing section. Also, x (k) is the received signal, h
(K) is an echo path transfer characteristic (impulse response) between the receiving speaker 4 and the transmission microphone 3, y (k) is an echo signal, ye (k) is a pseudo echo signal, and he (k) is an echo path impulse. An estimated value of the response, b (k) represents an error signal, s (k) represents a near-end talker's speech signal, and z (k) represents a microphone output signal.

In the echo canceller 21, the echo path estimation circuit 23 estimates the impulse response of the echo path and transfers the estimated value he (k) to the pseudo echo path convolution operation circuit 22. Then, in the pseudo echo path convolution operation circuit 22, a convolution operation of the estimated value he (k) and the received signal z (k) is executed to synthesize the pseudo echo signal ye (k), and the subtractor 24 uses the microphone. The pseudo echo signal ye (k) is subtracted from the output signal of No. 3. In this case, if the echo path impulse response is well estimated, the echo signal y (k) and the pseudo echo signal ye (k) are substantially equal to each other, and the result of the subtraction is included in the output of the microphone 3. Echo signal y
(K) is removed.

Here, the pseudo echo path convolution operation circuit 22 needs to follow the temporal change of the echo path impulse response h (k). Therefore, the echo path estimation circuit 23 estimates the echo path impulse response h (k) using an adaptive algorithm. This estimation operation is the receiving state, that is, s (k) ≈0, and z (k) ≈y.
It is executed when it can be regarded as (k).

The adaptive algorithm is an algorithm for obtaining the estimated value he (k) of the impulse response in the pseudo echo path, which is the LMS method, the learning identification method or the ES.
The law is known. Here, the value of the pseudo echo path is close to the value of the true echo path, and the pseudo echo signal ye (k)
The state in which is almost equal to the echo signal y (k) is called “converged”. Further, the pseudo echo path convolution operation circuit 22 and the echo path estimation circuit 23 are collectively referred to as an adaptive filter here.

[0011]

By the way, the above-mentioned conventional echo canceller (that is, an adaptive filter) is used only in the receiving state, that is, when the near-end talker transmitted signal does not exist.
Estimate the echo path impulse response. In the double talk in which the far-end talker and the near-end talker talk at the same time, that is, when the echo signal and the transmission signal are input to the microphone 3 together, the estimation operation cannot be performed. For this reason, if there is a change in the echo path such as a person moving during double talk, the true echo signal cannot be estimated, so the echo cannot be canceled sufficiently, resulting in deterioration of the call quality. There is a point.

Further, since the adaptive filter estimates both the amplitude and the phase of the echo path impulse response, it is possible to sufficiently suppress the echo when converged, but it is necessary to estimate more information. Therefore, when the echo path changes as described above, the number of echoes that cannot be extinguished instantaneously (residual echo) increases, which causes annoyance. In order to suppress this residual echo, an echo canceller generally has a nonlinear echo suppression processing unit 25 such as an echo suppressor or a center clipper in addition to the adaptive filk described above.

The processing by the non-linear echo suppression processing section 25 is not affected by fluctuations in the echo path.
There is an advantage that echo can be suppressed robustly. However, if the transmitted voice is present at the same time as the echo, they are suppressed without distinction, which causes a problem that the transmitted voice is distorted or the sound is interrupted.

The present invention has been made in view of the above problems, and has the following objects. (1) With respect to variations in the echo path during double talk, the echo is sufficiently suppressed from the viewpoint of hearing without distorting or breaking the transmitted voice, and good speech quality is secured. (2) The residual echo that cannot be completely erased by only the adaptive filter is sufficiently suppressed from the auditory sense without distorting or breaking the transmitted voice to secure good speech quality.

[0015]

In order to achieve the above object, in the present invention, as a first means relating to an echo suppression method, an echo path transfer function h (k) in which a received signal x (k) varies with time. Of the echo signal b
The transmission signal y superimposed on the transmission signal s (k) as (k)
In the method of suppressing the echo signal b (k) superposed on the transmission signal y (k) with respect to (k), the reception signal x
(K) and the transmission signal y (k) are each converted into a short-time spectrum, and the short-time spectrum of the transmission signal y (k) and the reception signal x (k) in a state where the transmission signal s (k) is not included. Of the echo path transfer function h (k) is calculated by using the short-time spectrum of the echo path transfer function h (k), the short-time spectrum of the transmission signal y (k), and the short-time spectrum of the reception signal x (k). Is used to estimate the amplitude characteristic value of the short-time spectrum of the echo signal b (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) at the following times, and the echo characteristic b (k) of the echo signal b (k) is estimated. The echo suppression gain G (is calculated using an arithmetic expression based on the amplitude characteristic value of the short-time spectrum, the amplitude characteristic value of the short-time spectrum of the transmission signal s (k), and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k). ω Is calculated, the echo suppression gain G
The means for suppressing the echo signal b (k) by multiplying (ω) by the short-time spectrum of the transmission signal y (k) is adopted.

As a second means relating to the echo suppression method, the received signal x (k) propagates through the echo path of the echo path transfer function h (k) where the time changes, and the echo signal b (k)
As a transmission signal y (k) superimposed on the transmission signal s (k) as
With respect to the echo signal b superposed on the transmission signal y (k)
In the method of suppressing (k), the received signal x (k) and the transmitted signal y (k) are each converted into a short-time spectrum, the fluctuation of the echo path transfer function h (k) during double talk is detected, and the echo Reduced signal e (k) and received signal x
A pseudo echo path transfer function he (k), which is an estimated value of the echo path transfer function h (k), is calculated based on (k) and
The echo signal b (k) included in the transmission signal y (k) is suppressed by using the pseudo echo path transfer function he (k), and the residual echo signal bd (k) is included in the transmission signal s (k). If the fluctuation of the echo path transfer function h (k) during double talk is detected by generating the echo reduction signal e (k), the pseudo echo path transfer function he (k) before the fluctuation detection is generated.
Pseudo-amplitude frequency function | He (ω) | ² is calculated from Eq. ( ² ), and when the fluctuation of the echo path transfer function h (k) during double talk is not detected, transmission in the state where the transmission signal s (k) is not included The residual amplitude frequency function | Hd (ω) | ² is calculated by using the short-time spectrum of the signal y (k) and the short-time spectrum of the received signal x (k), and the pseudo amplitude frequency calculated alternatively Based on the function | He (ω) | ² or the residual amplitude frequency function | Hd (ω) | ² , the echo signal b
(K) or the amplitude characteristic value of the short-time spectrum of the residual echo signal bd (k) is estimated, and the amplitude characteristic value of the short-time spectrum of the echo signal b (k) or the residual echo signal bd (k) and the transmission signal The echo suppression gain G is calculated using an arithmetic expression based on the amplitude characteristic value of the short-time spectrum of s (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k).
(Ω) is calculated, and when the fluctuation of the echo path transfer function h (k) during double talk is not detected, the echo suppression gain G (ω) is multiplied by the short time spectrum of the echo reduction signal e (k). The residual echo signal bd
(K) is suppressed, and when the fluctuation of the echo path transfer function h (k) during double talk is detected, the echo suppression gain G (ω) is multiplied by the short-time spectrum of the transmission signal y (k). Therefore, a means of suppressing the echo signal b (k) is adopted.

As a third means relating to the echo suppressing method, in the second means, the reduction amount of the echo signal b (k) is obtained from the transmission signal y (k) and the echo reduction signal e (k), and the reception When the signal x (k) is present and the reduction amount is monotonically increasing, it is determined that the transmission signal s (k) is not included.

As a fourth means relating to the echo suppressing method, in any one of the first to third means, the amplitude characteristic value of the short time spectrum of the echo signal b (k) and the echo signal b one unit time before Of the amplitude characteristic values of the short-time spectrum of (k), the larger one is the echo signal b.
The means of using the estimated value of the amplitude characteristic value of the short-time spectrum of (k) is adopted.

As a fifth means relating to the echo suppression method, in any one of the above-mentioned first to fourth means, a value obtained by subjecting the amplitude characteristic value of the short-time spectrum of the echo signal b (k) to audibility correction is used as the echo signal. A means of using the estimated value of the amplitude characteristic value of the short-time spectrum of b (k) is adopted.

As a sixth means relating to the echo suppression method, in the fifth means, the audibility correction gain is set based on the masking level of the echo signal by the received voice, the human audible level, and the transmission delay amount, and the audibility correction gain is set. A means of multiplying the correction gain by the amplitude characteristic value of the short-time spectrum of the echo signal b (k) to correct the audibility is adopted.

As a seventh means relating to the echo suppressing method, in any one of the first to sixth means, the amplitude characteristic value of the short time spectrum of the echo signal b (k) and the transmission signal s one unit time before are transmitted. Based on the following equation (1) consisting of the amplitude characteristic value of the short-time spectrum of (k), the short-time spectrum of the transmission signal y (k), the smoothing factor β, and the half-wave rectification function P for extracting only the positive component of the signal. Transmission signal s
The means of calculating the estimated value of the amplitude characteristic value of the short-time spectrum of (k) is adopted.

[0022]

[Formula 19]

As the eighth means relating to the echo suppressing method, the means for setting the soothing factor β to a value within the range of 0.9 to 0.999 in the seventh means is adopted.

As a ninth means relating to the echo suppressing method, in any one of the first to eighth means, the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e
(Ω) | ² and the value G ^PE (ω) of the equation (2) is
The means applied to the echo suppression gain G (ω) is adopted.

[0025]

[Equation 20]

As a tenth means relating to the echo suppressing method, in any one of the above first to eighth means, the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e
A means of applying the value G ^W (ω) of the arithmetic expression (3) composed of (ω) | ² to the echo suppression gain G (ω) is adopted.

[0027]

[Equation 21]

As an eleventh means relating to the echo suppressing method, in any one of the above first to eighth means, the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e
(Omega) | value G ^ML of ² which consists of arithmetic expression (4) (omega),
The means applied to the echo suppression gain G (ω) is adopted.

[0029]

[Equation 22]

As a twelfth means relating to the echo suppression method, in any one of the first to eighth means, the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e
(Ω) | ² and transmission signal power | Y ^t (ω) | ^{2 is used} to calculate an intermediate variable υ (ω), and the intermediate variable υ (ω) and echo signal power | E ^t e (ω) | ² and transmission signal power | Y ^t (ω) | ² and the value G of the equation (6) consisting of the _0th and 1st modified Bessel functions I ₀ and I ₁ for the intermediate variable υ (ω) A means of applying ^EM (ω) to the echo suppression gain G (ω) is adopted.

[0031]

[Equation 23]

[0032]

[Equation 24]

On the other hand, in the present invention, as the first means relating to the echo suppressor, the received signal x (k) propagates through the echo path of the echo path transfer function h (k), which changes with time, and the echo signal b (k). ), The echo signal b (k) superimposed on the transmission signal y (k) is suppressed in the echo suppression device for the transmission signal y (k) superimposed on the transmission signal s (k). Receiving signal converting means for converting (k) into a short time spectrum, transmitting signal converting means for converting the transmission signal y (k) into a short time spectrum, and receiving signal x (k).
And a transmission signal y (k) based on the transmission signal y (k) and a one-sided speech state detection unit that detects a state in which the transmission signal s (k) is not included, and a transmission signal y in a state in which the transmission signal s (k) is not included.
The echo path transfer function h (k) is calculated using the short-time spectrum of (k) and the short-time spectrum of the received signal x (k).
Using the echo path transfer function calculating means for calculating the amplitude frequency function of, and the amplitude frequency function and the short time spectrum of the transmission signal y (k) and the short time spectrum of the received signal x (k), Echo signal b (k)
Echo signal amplitude frequency characteristic estimating means for estimating the amplitude characteristic value of the short-time spectrum of, the amplitude frequency function, the short-time spectrum of the transmission signal y (k), and the reception signal x.
A transmission signal amplitude frequency characteristic estimation means for estimating the amplitude characteristic value of the short time spectrum of the transmission signal s (k) at the subsequent time using the short time spectrum of (k), and the echo signal b (k). The echo suppression gain G is calculated by using an arithmetic expression based on the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k). An echo suppression gain determination unit for calculating (ω), and the echo suppression gain G (ω)
A means is provided that includes an echo suppression processing unit that multiplies the short-time spectrum of the transmission signal y (k) with the echo suppression processing unit that inversely converts the output of the echo suppression processing unit and outputs the result to the transmission path. To do.

As a second means relating to the echo suppressor, the received signal x (k) propagates through the echo path of the echo path transfer function h (k), which changes with time, and the echo signal b (k).
As a transmission signal y (k) superimposed on the transmission signal s (k) as
With respect to the echo signal b superposed on the transmission signal y (k)
An echo suppressor for suppressing (k), the received signal x
Reception signal converting means for converting (k) into a short time spectrum, transmission signal converting means for converting a transmission signal y (k) into a short time spectrum, reception signal x (k) and transmission signal y.
One-side utterance state detection unit that detects a state in which the transmission signal s (k) is not included based on (k) and echo path fluctuation that detects fluctuations in the echo path transfer function h (k) during double talk A pseudo echo path transfer function he (k), which is an estimated value of the echo path transfer function h (k), is generated using the detection unit and the adaptive filter, and the pseudo echo path transfer function he is generated.
The echo reduction signal e (k) in which the echo signal b (k) included in the transmission signal y (k) is suppressed by using (k) and the residual echo signal bd (k) is included in the transmission signal s (k). ) Is output, and when the fluctuation of the echo path transfer function h (k) during double talk is detected by the echo path fluctuation detecting unit, the pseudo echo path transfer function he before the fluctuation detection is detected.
(K) and an echo canceller that outputs the transmission signal y (k) as it is, and an echo path transfer function h during double talk by the echo path fluctuation detection unit.
When the variation of (k) is not detected, the transmission signal s (k)
Echo path transfer for calculating the residual amplitude frequency function | Hd (ω) | ² based on the short-time spectrum of the echo reduction signal e (k) and the short-time spectrum of the received signal x (k) The echo path transfer function h during double talk by the function calculating means and the echo path fluctuation detecting section
A second echo path transfer function calculation for calculating the pseudo amplitude frequency function | He (ω) | ² from the pseudo echo path transfer function he (k) output from the echo canceller when the fluctuation of (k) is detected Means and the pseudo-amplitude frequency function | He (ω) | ² which is alternatively input from the echo path transfer function calculating means or the second echo path transfer function calculating means.
Alternatively, the echo signal b (k) or the residual echo signal bd (k) is calculated based on the residual amplitude frequency function | Hd (ω) | ^2.
Echo signal amplitude frequency characteristic estimating means for estimating the amplitude frequency characteristic of the short time spectrum of, and the amplitude frequency characteristic and the short time spectrum of the echo reduction signal e (k) or the transmission signal y (k) and the received signal x (k). ), The echo signal amplitude frequency characteristic estimating means for estimating the amplitude frequency characteristic of the residual echo signal bd (k) or the short time spectrum of the echo signal b (k) at a subsequent time, and the amplitude frequency. Using the characteristics and the short time spectrum of the echo reduction signal e (k) or the transmission signal y (k) and the short time spectrum of the reception signal x (k), a short time of the transmission signal s (k) at a subsequent time. A transmission signal amplitude frequency characteristic estimating means for estimating the amplitude frequency characteristic of the spectrum, and the residual echo signal bd (k) or the echo signal b (k). Calculation based on the amplitude frequency characteristic of the short-time spectrum of the received signal, the amplitude frequency characteristic of the short-term spectrum of the transmission signal s (k), and the amplitude-frequency characteristic of the short-term spectrum of the echo reduction signal e (k) or the transmission signal y (k). And an echo suppression gain determination unit that calculates the echo suppression gain G (ω) using an equation, and the echo suppression gain G (ω)
Of the echo reduction signal e (k) or the transmission signal y (k), and an echo suppression processing unit for inversely converting the output of the echo suppression processing unit and outputting to the transmission path. And a means comprising means.

As a third means relating to the echo suppressor, in the second means, the reduction amount of the echo signal b (k) is obtained from the transmission signal y (k) and the echo reduction signal e (k), and the received signal is received. It is said that the one-sided speech state detection unit is configured to determine that the transmission signal s (k) is not included when the signal x (k) is present and the reduction amount is monotonically increasing. Adopt means.

As a fourth means relating to the echo suppressor, in any one of the first to third means, the amplitude frequency function, the short-time spectrum of the transmission signal y (k) and the short-term of the reception signal x (k) are used. An echo signal power calculator that estimates the amplitude characteristic value of the short-time spectrum of the echo signal b (k) at a subsequent time using the time spectrum;
The larger one of the first delay operator that outputs the amplitude characteristic value with a delay of one unit time and the output of the echo signal power calculation unit and the first delay operator is the echo signal b (k).
The means for constructing the echo signal amplitude frequency characteristic estimating means from the echo signal power determining section which is the estimated value of the amplitude characteristic value of the short-time spectrum is adopted.

As a fifth means relating to the echo suppressor, in any one of the above-mentioned first to fourth means, the echo signal amplitude frequency characteristic estimating means uses the amplitude characteristic value of the short-time spectrum of the echo signal b (k). A means of providing an echo signal audibility correction means for performing audibility correction is adopted.

As a sixth means relating to the echo suppressing apparatus, in the fifth means, the masking level of the echo signal by the received voice, the human audible level, and the auditory sense correction gain set based on the transmission delay amount are stored. The audibility correction gain storage unit and the audibility correction gain are stored in the echo signal b.
A means for constructing an echo signal audibility correction means from the audibility correction section for multiplying the amplitude characteristic value of the short-time spectrum of (k) is adopted.

As a seventh means relating to the echo suppressor, in any one of the above first to sixth means, a smoothing factor storage section for storing the smoothing factor β and a short-term spectrum of the transmission signal s (k) are stored. A second delay operator for delaying the amplitude characteristic value by one unit time, the output of the second delay operator, the amplitude characteristic value of the short time spectrum of the echo signal b (k) and the shortness of the transmission signal y (k). The estimated value of the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) is calculated based on the following equation (1) including the time spectrum, the smoothing factor β, and the half-wave rectification function P that extracts only the positive component of the signal. And a transmission signal power calculating section for constituting a transmission signal amplitude frequency characteristic estimation means.

[0040]

[Equation 25]

As the eighth means relating to the echo suppressor, the means for setting the soothing factor β to a value within the range of 0.9 to 0.999 in the seventh means is adopted.

As a ninth means relating to the echo suppressing device, in any one of the first to eighth means, the echo suppression gain G (ω) is used as the transmission signal power | S ^t e (ω).
The means for configuring the echo suppression gain determination unit is adopted so as to calculate the value G ^PE (ω) based on the arithmetic expression (2) consisting of | ² and the echo signal power | E ^t e (ω) | ² .

[0043]

[Equation 26]

As a tenth means relating to the echo suppressing apparatus, in any one of the above first to eighth means, as the echo suppressing gain G (ω), the transmission signal power | S ^t e (ω)
| ² and the echo signal power | E ^t e (ω) | ² is adopted, the echo suppression gain determination unit is configured to calculate the value G ^W (ω) based on the equation (3). To do.

[0045]

[Equation 27]

As an eleventh means relating to the echo suppressing device, in any one of the first to eighth means, the echo suppression gain G (ω) is set as the transmission signal power | S ^t e (ω).
| ² and the echo signal power | E ^t e (ω) | ² is used to calculate the value G ^ML (ω) based on the equation (4). .

[0047]

[Equation 28]

As a twelfth means relating to the echo suppressor, in any one of the first to eighth means, the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e
(Ω) | ² and transmission signal power | Y ^t (ω) | ^{2 is used} to calculate an intermediate variable υ (ω), and the intermediate variable υ (ω) and echo signal power | E ^t Based on e (ω) | ² and the transmission signal power | Y ^t (ω) | ² and the modified Bessel functions I ₀ and I _{1 of the 0th} and 1st order with respect to the intermediate variable υ (ω), A means of configuring the echo suppression gain determination unit so as to calculate the value G ^EM (ω) as the echo suppression gain G (ω) is adopted.

[0049]

[Equation 29]

[0050]

[Equation 30]

Further, according to the present invention, as the first means relating to the computer-readable storage medium in which the echo suppressing program is stored, the echo of the echo path transfer function h (k) in which the received signal x (k) is time-varying. Propagate the path,
An echo suppression program for suppressing the echo signal b (k) superimposed on the transmission signal y (k) for the transmission signal y (k) superimposed on the transmission signal s (k) as the echo signal b (k) In a computer-readable storage medium in which is stored, processing for converting the received signal x (k) and the transmitted signal y (k) into short-time spectra, and the transmitted signal s
A process of calculating the amplitude frequency function of the echo path transfer function h (k) using the short-time spectrum of the transmission signal y (k) and the short-time spectrum of the received signal x (k) in the state where (k) is not included. And the amplitude frequency function, the short-time spectrum of the transmission signal y (k), and the reception signal x.
A process of estimating the amplitude characteristic value of the short-time spectrum of the echo signal b (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) using the short-time spectrum of (k). Calculation based on the amplitude characteristic value of the short-time spectrum of the echo signal b (k), the amplitude characteristic value of the short-time spectrum of the transmission signal s (k), and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k). A process of calculating the echo suppression gain G (ω) using an equation and the echo signal b (k) is suppressed by multiplying the short time spectrum of the transmission signal y (k) by the echo suppression gain G (ω). The means for storing the echo suppression program consisting of

As a second means relating to a computer-readable storage medium in which the echo suppression program is stored, in the first means, the amplitude characteristic value of the short-time spectrum of the echo signal b (k) and one unit time before Of the amplitude characteristic values of the short-time spectrum of the echo signal b (k) of
A means of storing an echo suppression program including a process of setting the larger one as an estimated value of the amplitude characteristic value of the short-time spectrum of the echo signal b (k) is adopted.

As the third means relating to the computer-readable storage medium in which the echo suppressing program is stored, the echo signal b is used in the first or second means.
A means for storing an echo suppression program including a process of using a value obtained by subjecting the amplitude characteristic value of the short-time spectrum of (k) to the auditory sense as an estimated value of the amplitude characteristic value of the short-time spectrum of the echo signal b (k) is stored. adopt.

As a fourth means relating to a computer-readable storage medium in which the echo suppressing program is stored, the third means is based on the masking level of the echo signal by the received voice, the human audible level and the transmission delay amount. A means for storing an echo suppression program including a process of setting the audibility correction gain by applying the audibility correction gain and multiplying the audibility correction gain by the amplitude characteristic value of the short-time spectrum of the echo signal b (k) to correct the audibility.

As a fifth means relating to a computer-readable storage medium storing an echo suppression program, the amplitude characteristic value of the short-time spectrum of the echo signal b (k) in any one of the first to fourth means. And a half-wave rectification function P for extracting only the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) one unit time before, the short-time spectrum of the transmission signal y (k), the smoothing factor β, and the positive component of the signal. Based on the following equation (1)
The means for storing the echo suppression program including the processing for calculating the estimated value of the amplitude characteristic value of the short-time spectrum in (k) is adopted.

[0056]

[Equation 31]

As a sixth means relating to the computer-readable storage medium in which the echo suppressing program is stored, the smoothing factor β is included in the fifth means.
Is stored in the range of 0.9 to 0.999.

As a seventh means relating to a computer-readable storage medium in which the echo suppressing program is stored, in any one of the first to sixth means, the transmission signal power | S ^t e (ω) | ² Echo signal power ｜ E ^t e
(Ω) | ² and the value G ^PE (ω) of the equation (2) is
A means for storing an echo suppression program including processing applied to the echo suppression gain G (ω) is adopted.

[0059]

[Equation 32]

As an eighth means relating to a computer-readable storage medium in which the echo suppressing program is stored, in any one of the first to sixth means, the transmission signal power | S ^t e (ω) | ² Echo signal power ｜ E ^t e
A means of storing an echo suppression program including a process of applying the value G ^W (ω) of the arithmetic expression (3) consisting of (ω) | ² to the echo suppression gain G (ω) is adopted.

[0061]

[Expression 33]

As a ninth means relating to a computer-readable storage medium in which the echo suppression program is stored, in any one of the first to sixth means, the transmission signal power | S ^t e (ω) | ² Echo signal power ｜ E ^t e
(Omega) | value G ^ML of ² which consists of arithmetic expression (4) (omega),
A means for storing an echo suppression program including processing applied to the echo suppression gain G (ω) is adopted.

[0063]

[Equation 34]

As the tenth means relating to the computer-readable storage medium in which the echo suppression program is stored, in any one of the first to sixth means, the transmission signal power | S ^t e (ω) | ² Echo signal power ｜ E ^t e
(Ω) | ² and transmission signal power | Y ^t (ω) | ^{2 is used} to calculate an intermediate variable υ (ω), and the intermediate variable υ (ω) and echo signal power | E ^t e (ω) | ² and transmission signal power | Y ^t (ω) | ² and the value G of the equation (6) consisting of the _0th and 1st modified Bessel functions I ₀ and I ₁ for the intermediate variable υ (ω) A means of storing an echo suppression program including a process of applying ^EM (ω) to the echo suppression gain G (ω) is adopted.

[0065]

[Equation 35]

[0066]

[Equation 36]

[0067]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of an echo suppressing method and apparatus according to the present invention and a computer-readable storage medium in which an echo suppressing program is stored will be described below with reference to FIGS. . In addition, in the following description, the components already described in FIGS. 8 and 9 are denoted by the same reference numerals and the description thereof will be omitted.

[First Embodiment] First, a first embodiment of the present invention will be described with reference to FIGS. Figure 1
FIG. 3 is a block diagram showing a functional configuration of this embodiment. In this figure, reference numeral 31 is a fast Fourier transform unit (received signal transforming means), 32 is a fast Fourier transform unit (transmitted signal transforming means), 33 is a fast inverse Fourier transform unit (speech signal inverse transforming means), and 34 is one side. Utterance state detector 35, echo path coupling amount calculator 36, echo signal power calculator 3
Reference numerals 7 and 43 are delay operators, 38 is an echo signal power determination unit, 39 is a hearing correction gain storage unit, 40 is a hearing correction unit,
Reference numeral 41 is a smoothing factor storage unit, 42 is a transmission signal power calculation unit, 44 is an echo suppression gain determination unit, 45 is an echo suppression processing unit, 46 is a reception signal selection unit for calculating an echo path coupling amount, and 47 is an echo path coupling amount. It is a transmission signal selection unit for calculation. In this figure, the echo path transfer characteristic (impulse response) of the echo path between the receiving speaker 4 and the transmission microphone 3 is h (k).

Here, the echo path coupling amount calculator 35
The echo path coupling amount calculation reception signal selection unit 46 and the echo path coupling amount calculation transmission signal selection unit 47 constitute an echo path transfer function calculation unit. Echo signal power calculation unit 36, delay operator 37, echo signal power determination unit 38, auditory sense correction gain storage unit 39, and auditory sense correction unit 40.
And constitutes the echo signal amplitude frequency characteristic estimating means. Auditory correction gain storage unit 39 and audition correction unit 40
And constitutes the echo signal auditory sense correction means.
Further, the smoothing factor storage unit 41, the transmission signal power calculation unit 42, and the delay operator 43 constitute transmission signal amplitude frequency characteristic estimating means.

The fast Fourier transform unit 31 receives the received signal x
Fast Fourier transform of (k) to obtain short-time spectrum X
^It is output to the echo signal power calculation unit 36 and the echo path coupling amount calculation reception signal selection unit 46 as ^t (ω) (t: subscript indicating time).

The fast Fourier transform section 32 determines the transmission signal y.
Fast Fourier transform of (k) to obtain short-time spectrum Y
^It is output to the transmission signal power calculation unit 42, the echo suppression gain determination unit 44, the echo suppression processing unit 45, and the echo path coupling amount calculation transmission signal selection unit 47 as ^t (ω). The transmission signal y (k) is a transmission signal s (k) superposed with an echo signal b (k), that is, y (k) =
It is given as s (k) + b (k).

The one-sided utterance state detecting section 34 receives the received signal x
(K) and the transmitted signal y (k), the transmitted signal y
The state (time T) in which the transmission signal y (k) is composed only of the echo signal b (k) is detected and the detection result is echoed. It is output to the reception signal selection unit 46 for calculating the path coupling amount and the transmission signal selection unit 47 for calculating the echo path coupling amount.

Receiving signal selecting section 4 for calculating echo path coupling amount
6 (hereinafter, simply referred to as the reception signal selection unit 46), the short-time spectrum X ^{t of the} reception signal x (k) input from the fast Fourier transform unit 31 every moment based on the detection result of the one-sided speech state detection unit 34. Of (ω), the short-time spectrum X ^{t = T} (ω) at the time T is selected and output to the echo path coupling amount calculation unit 35.

Echo path coupling amount calculation transmission signal selection unit 4
7 (hereinafter, simply referred to as the transmission signal selection unit 47), the short-time spectrum Y ^{t of the} transmission signal y (k) that is input momentarily from the fast Fourier transform unit 32 based on the detection result of the one-sided speech state detection unit 34. Of (ω), the short time spectrum Y ^{t = T} (ω) at the time T is selected and output to the echo path coupling amount calculation unit 35.

The echo path coupling amount calculator 35 determines the short-time spectrum X of the received signal x (k) at time T.
^{t = T (ω)} and also based on the short-time spectrum Y ^{t = T} of the transmission signal y at time T (k) (ω), short-time spectrum ^{X t = T (ω),} Y t = T (ω ) Each amplitude characteristic value of the reception signal power | X ^t (ω) | ² and the transmission signal power | Y
^{t = T} (ω) | ² (= | E ^{t = T} (ω) | ² ) and the received signal power | X ^t (ω) | ² and transmission signal power | Y
^{From t = T} (ω) | ² and the amplitude frequency function of the echo path transfer characteristic h (k) is calculated as the squared value of the amplitude characteristic value, that is, the echo path coupling amount | He (ω) | ² , and the echo signal power is calculated. And outputs to the unit 36.

The echo signal power calculator 36 multiplies the received signal power | X ^t (ω) | ² at the current time by the above echo path coupling amount | He (ω) | ² to obtain the short echo signal b (k). The amplitude characteristic value of the time spectrum E ^t (ω) is calculated as the predicted echo signal power | E ^t e (ω) | ² and output to the delay operator 37 and the echo signal power determination unit 38.

The delay operator 37 delays the predicted echo signal power | E ^t e (ω) | ² by a unit time, and is 1 for the input of the predicted echo signal power | E ^t e (ω) | ^2.
Predicted echo signal power before unit time (t-1) | ^Et- 1e
(Ω) | ² is output to the echo signal power determination unit 38.

The echo signal power determination unit 38 compares the predicted echo signal power | E ^t e (ω) | ² with the predicted echo signal power | E ^t-1 e (ω) | ² one unit time ago. , The larger value is predicted echo signal power | E ' ^t e (ω)
| ² is determined and output to the auditory sense correction unit 40.

On the other hand, the audibility correction gain storage section 39 stores the audibility correction gain F (ω). The audibility correction gain F (ω) is a gain for correcting the physical (objective) frequency characteristic of the echo to the human subjective (audible audible) frequency characteristic. Predicted echo signal power above ｜
E ′ ^t e (ω) | ² is an objective quantity, which is different from the subjective quantity of the echo that is actually transmitted and loudened by the other party and detected by a person.

This is because the human auditory characteristics such as the voice being spoken by the other party and the voice sent from the other party are masked. Since this auditory characteristic changes according to the call state, the auditory sense correction gain F (ω) cannot be uniquely determined, but its representative value can be determined. For example, the auditory sense correction gain F (ω) can be calculated from the experimental value of the frequency characteristic of the required echo suppression amount of Japanese Patent Application No. 9-59549.

The audibility correction unit 40 multiplies the predicted echo signal power | E ′ ^t e (ω) | ² by the audibility correction gain F (ω), and obtains the obtained value F (ω) | E ′ ^t e ( ω) | ² is output to the transmission signal power calculation unit 42 and the echo suppression gain determination unit 44.

The smoothing factor storage unit 41 stores the smoothing factor β. For example, the smoothing factor storage unit 41 stores in the smoothing factor β a value within the range of 0.9 to 0.999,
Preferably, "0.99" is stored. This smoothing factor β is obtained by calculating the predicted transmission signal power | S ^t e (ω) | ² one unit time before the prediction transmission signal power | S as shown in Expression (1) described below. ^t-1 e (ω) |
It is a value that sets the degree of influence of ² . The larger the smoothing factor β, the more the past estimated value, that is, the predicted transmission signal power | S ^t-1 e (ω) | ² becomes dominant, and the prediction transmission signal power | S ^t e (ω) | ² is temporal. Will be averaged.

The transmission signal power calculation unit 42 receives the smoothing factor β, the value F (ω) | E ′ ^t e (ω) | ² input from the auditory sense correction unit 40, and the delay operator 43. Using the estimated transmission signal power | S ^t-1 e (ω) | ² and the short-time spectrum S ^t (ω) of the transmission signal s (k)
Is calculated as the predicted transmission signal power | S ^t e (ω) | ² and the predicted transmission signal power | S ^t e (ω) | ² is calculated by the delay operator 43 and the echo suppression gain determination unit 44. Output to.

The delay operator 43 calculates the predicted transmission signal power |
S ^t e (ω) | ² is delayed by a unit time, and when the predicted transmission signal power | S ^t e (ω) | ² is input, the prediction transmission one unit time before (t−1) Signal power ｜ S ^t-1 e
(Ω) | ² is output.

The echo suppression gain determination unit 44 calculates the predicted transmission signal power | S ^t e (ω) | ² and the value F (ω) | E ′ ^t e (ω) | ² input from the auditory sense correction unit 40. And the short-time spectrum Y input from the fast Fourier transform unit 32.
^The echo suppression gain G (ω) is calculated based on ^t (ω) and output to the echo suppression processing unit 45.

The echo suppression processing unit 45 multiplies the short time spectrum Y ^t (ω) by the echo suppression gain G (ω) and outputs the processed signal S ^t e (ω) in which the echo is suppressed to the fast inverse Fourier transform unit 33. Output.

Further, the fast inverse Fourier transform unit 33 performs an inverse Fourier transform on the processed signal Y ^t e (ω) to suppress the echo signal b (k) and emphasize the transmission signal s (k). The signal se (k) is output to the transmission line 9.

In the present embodiment, the echo suppressor having the above-described functional configuration is realized by incorporating an echo suppressor program into a computer device. In this case, the computer device receives the reception signal x (k) and the transmission signal y (k) as input signals, and uses the reception signal x (k) and the transmission signal y (k) based on the echo suppression program as shown in FIG. The time signal se (k) in which the echo signal b (k) is suppressed is generated by processing in the flow shown in FIG.

Next, the operation of the echo suppressor will be described in detail. First, the one-sided speech state detection unit 34 constantly monitors the reception signal x (k) and the transmission signal y (k), and detects the state of only the reception signal, that is, the state in which the transmission signal s (k) is not transmitted. Then, the detection time T is output to the reception signal selection unit 46. As a result, the reception signal selection unit 46
Is the short-time spectrum X ^t (ω) of the received signal x (k) at the time T.
Select ^{t = T} (ω) and output it to the echo path coupling amount calculation unit 35.

The echo path coupling amount calculator 35 receives the reception signal power | X ^t (ω) for the short time spectrum X ^{t = T} (ω) of the reception signal x (k) at the time T which is selectively input in this way. ω) | ² and the transmission signal power | Y ^{t = T} (ω) | ^{2 of} the transmission signal y (k), that is, the echo signal power | E ^{t = T} (ω) | ² is calculated, and the echo The echo path coupling amount | He (ω) | ² is calculated based on the signal power | E ^{t = T} (ω) | ² and the reception signal power | X ^{t = T} (ω) | ² .

The echo signal b (k) is the received signal x (k).
Is a signal transmitted through an echo path (transmission characteristic between the receiving speaker 4 and the transmission microphone 3) given as an echo path transfer characteristic (impulse response) h (k), that is, b
It is given as (k) = h (k) · x (k). Therefore, the echo path coupling amount | He (ω) | ² which is the power of the echo path transfer characteristic h (k) is equal to the echo signal power | E.
It is calculated by dividing ^{t = T} (ω) | ² by the reception signal power | X ^{t = T} (ω) | ² .

In the subsequent processing, the reception signal power │X ^{t = T} (ω) │ ² and the echo signal power │E in the absence of the transmission signal s (k) (that is, at time T)
^{t = T (ω)} | ² Metropolitan echo path coupling amount calculated from | the He
Various calculations are performed based on (ω) | ² .

First, the echo signal power calculation unit 36 calculates the reception signal power | X ^t (ω) | ² from the short time spectrum X ^t (ω) of the reception signal x (k) at the current time t, and the reception signal power | X ^t (ω) | The echo signal power | E ^t e (ω) | ² is calculated by multiplying the signal power | X ^t (ω) | ² by the echo path coupling amount | He (ω) | ² . This predicted echo signal power | E ^t e (ω) | ² is the transmission signal s (k) for the echo path transfer characteristic h (k) that changes from moment to moment.
And estimates the power of the echo signal at the current time t using ² | echo path binding amount was determined for the time T that but absent | He (ω).

The echo signal power deciding unit 38 makes one unit time for the echo signal power | E ^t e (ω) | ² and the echo signal power | E ^t e (ω) | ² thus estimated. The previous echo signal power | E ^t-1 e (ω) | ² is compared, and the larger value is predicted echo signal power | E ′ ^t e
(Ω) | Determine to ² .

This predicted echo signal power | E ′ ^t e (ω)
| ² is the hearing-sensation correction gain F
(Ω) is multiplied and the auditory sense correction value F (ω) | E ' ^t e (ω) |
Converted to ² . Then, in the transmission signal power calculation unit 42, the auditory sense correction value F (ω) | E ′ ^t e (ω) | ² , the smoothing factor β, and the predicted transmission signal power one unit time before | S ^t-1 e (Ω) | ² is used to calculate the predicted transmission signal power | S ^t e (ω) | ² . In this case, the transmission signal power calculator 42 calculates the predicted transmission signal power | S ^t e (ω) | ² based on the following equation (1). Here, P is a half-wave rectification function that extracts only the positive component of the signal.

[0096]

[Equation 37]

In this way, the predicted transmission signal power | S ^t e
When (ω) | ² is calculated, the echo suppression gain determination unit 4
4 is the predicted transmission signal power | S ^t e (ω) | ² , the perceptual correction value F (ω) | E ′ ^t e (ω) | ² and the transmission signal y.
The echo suppression gain G (ω) is calculated based on the short time spectrum Y ^t (ω) of (k). For example, the echo suppression gain determination unit 44 calculates G ^PE (ω) from the echo suppression gain G (ω) based on the following expression (2), and this G ^PE (ω)
Is an echo suppression gain G (ω).

[0098]

[Equation 38]

Further, as the echo suppression gain G (ω), it is possible to apply G ^W (ω) obtained based on the equation (3) and also obtained based on the equation (4). It is also conceivable to apply G ^ML (ω) to the echo suppression gain G (ω).

[0100]

[Formula 39]

[0101]

[Formula 40]

Further, as the echo suppression gain G (ω), the predicted transmission signal power | S ^t e (ω) | ² and the predicted echo signal power | E ′ ^t e (ω) | ² are used to obtain the equation (5) It is conceivable that the intermediate variable υ (ω) is calculated by (4), and the G ^EM (ω) calculated by the equation (6) is applied using the modified Bessel functions I ₀ and I ₁ of the 0th order and the 1st order.

[0103]

[Formula 41]

[0104]

[Equation 42]

In this way, the echo path coupling amount | He (ω) | ^{2 at} time T when the transmission signal s (k) does not exist
Predicted transmission signal power | S ^t e (ω) | ² obtained by using
And the auditory sense correction value F (ω) | E ′ ^t e (ω) | ² , and the short-time spectrum Y ^t of the transmission signal y (k) at the current time t.
The echo suppression gain G (ω) is calculated based on (ω).

The echo suppression processing section 45 multiplies the echo suppression gain G (ω) by the short time spectrum Y ^t (ω) of the transmission signal y (k) at the current time t to obtain the short time spectrum Y ^t. The processed signal S ^t e (ω) in which the echo signal component included in (ω) is suppressed and the transmission signal component is emphasized is generated. Then, this processed signal S ^t e
(Ω) is converted into a time signal se (k) by the fast inverse Fourier transform unit 33 and sent to the transmission line 9.

As described above, in this embodiment, the echo path coupling amount | He for determining the echo suppression gain G (ω)
(Ω) | ² is calculated based on the transmission signal y (k) at the time T when the transmission signal s (k) does not exist, that is, the echo signal b (k). That is, the conventional adaptive filk estimates both the amplitude and the phase of the impulse response of the echo path, whereas in the present embodiment, the echo path coupling amount | H.
e (ω) | ² That is, only the square value of the amplitude is used. Therefore, the fluctuation of the echo path, that is, the echo path transfer characteristic h
With respect to the variation of (k), the echo can be robustly suppressed because the variation of the phase value is ignored as compared with the processing by the adaptive filter.

Further, by applying G ^EM (ω) obtained using the intermediate variable υ (ω) to the echo suppression gain G (ω), the echo signal b for the transmission signal y (k) is obtained.
The echo suppression gain G (ω) is set based on the ratio of (k). This gives the transmitted signal y (k)
It is possible to suppress only the echo signal b (k) included in. Therefore, it is possible to suppress only the echo while minimizing the distortion of the transmitted voice and the interruption of the sound, which are problems in the conventional nonlinear echo suppression processing.

Further, the power of the echo signal b (k) is converted into the echo path coupling amount | He (ω) | ² and the reception signal power | X.
It is determined from ^t (ω) | ^2, but this alone does not take into account the influence of the reverberation component of the echo. Therefore, the echo signal power | E estimated one unit time before (t-1)
^t-1 e (ω) | ² is compared with the echo signal power | E ^t e (ω) | ² estimated at the current time t, and a larger value of these is determined as the echo signal power. Therefore, the echo signal power can be determined in consideration of the reverberation component of the echo generated one unit time before (t-1).
Therefore, it is possible to suppress an echo more fully.

The echo returned to the speaker (in the room on the reception signal side) is masked by the voice emitted by the speaker, and therefore the frequency characteristic is not flat in terms of hearing. Therefore, the echo signal power is multiplied on the frequency axis by the auditory sense correction gain based on the required echo suppression amount proposed in Japanese Patent Application No. 9-59549 to weight the auditory sense. For this reason, it is possible to further suppress the echo component in the frequency band that is easily audible. On the contrary, by not unnecessarily suppressing the frequency band including the echo component, which is not so noticeable to the listener, it is possible to minimize the distortion of the transmitted voice.

Therefore, for the above four reasons, it is possible to sufficiently suppress the echo in terms of hearing without distorting the transmitted voice or breaking the sound even when the echo path changes during double talk.

FIG. 2 shows the configuration of a subjective evaluation experiment system for confirming the effects of this embodiment. In this subjective evaluation experiment, the call environment condition is set to simulate double talk in a loud voice call system, and the rater 12 evaluates the quality of the other party's voice at that time. Specifically, on the evaluation side 51, the speaker 10 amplifies a short sentence of a female voice, and from the speaker 2, a different short sentence of a male voice uttered by a speaker 11 on the opposite side 52 is amplified.

At this time, the echo of the female voice on the evaluation side sneaked into the microphone 3 in the other party's room 52 is superimposed on the male voice. This echo is suppressed. The frequency band of voice is 100 Hz to 7 kHz, and the volume of the received voice is ITU Recommendation P. 34. The reverberation time in the other party room 52 is about 150 ms.

FIG. 3 shows the opinion mean value (MOS) of the experimental results for the evaluation speech processed by the conventional and the present embodiment in the above subjective evaluation experiment system. Here, the FF when performing the processing of the present embodiment
The number of T points is 512, and the transmission delay as a communication condition is 0 ms.
(Processing delay only), acoustic coupling amount was set to -3 dB.
As is clear from this figure, in the present embodiment, the processing quality of echo suppression is improved compared to the conventional method even in the subjective evaluation.

In this embodiment, the echo suppressing device is configured by software means based on the echo suppressing program, but the present invention is not limited to this. For example, the functional configuration shown in FIG. 1 can be configured by circuit elements in hardware.

In the above embodiment, the received signal x
The fast Fourier transform (FFT) is adopted as a means for converting (k) and the transmission signal y (k) into a short-time spectrum. However, the present invention is not limited to this, and by dividing the received signal x (k) and the transmitted signal y (k) into each fractional band using a discrete cosine transform, a wavelet transform, or a frequency division circuit. It is possible to convert to a short-time spectrum.

[Second Embodiment] Next, a second embodiment of the present invention will be described with reference to FIGS. In the second embodiment, a residual echo is further suppressed by combining the first embodiment with an echo canceller which is substantially the same as the echo canceller 21 shown in FIG. In addition, in the following description, description of the same components as those shown in FIGS. 1 and 9 will be omitted.

FIG. 4 is a block diagram showing the functional arrangement of the echo suppressor according to this embodiment. In this figure, reference numeral 60 is an echo canceller, 61 is an echo suppressor, 62 is an echo path variation detector, 63 is an echo path coupling amount calculator (second echo path transfer function calculator), 6
Reference numeral 4 is a one-sided speech state detection unit.

As shown in FIG. 5, the echo canceller 60 has almost the same structure as the conventional echo canceller 21 shown in FIG.
The configuration of 3'is different from the conventional echo path estimation circuit 23. That is, the echo path estimation circuit 2 of the present embodiment
3'is a kind of adaptive filter, which estimates the pseudo echo path he (k) based on the control signal output from the echo path fluctuation detection unit 62 and also
It is configured not only to output (k) to the pseudo echo path convolution operation circuit 22, but also to the echo path coupling amount calculation unit 63.

Such an echo canceller 60 simulates from the transmission signal y (k) (= b (k) + s (k)) given as the sum of the echo signal b (k) and the transmission signal s (k). By subtracting the echo signal be (k), the echo reduction signal e (k) (=
y (k) -be (k) = bd (k) + s (k)) is output to the echo suppressing section 61, the echo path variation detecting section 62, and the one-sided speech state detecting section 64. Where the signal bd (k)
Is the pseudo echo signal be (k) from the echo signal b (k).
Is a residual echo signal obtained by subtracting.

The echo suppressor 61 includes the fast Fourier transform units 31, 32, the fast inverse Fourier transform unit 33, the echo signal power calculator 36, and the delay operator 3 in the first embodiment.
7, 43, echo signal power determination unit 38, hearing correction gain storage unit 39, hearing correction unit 40, smoothing factor storage unit 41, transmission signal power calculation unit 42, echo suppression gain determination unit 44, echo suppression processing unit 45, This is a collection of an echo path coupling amount calculation reception signal selection unit 46 and an echo path coupling amount calculation transmission signal selection unit 47, and is a new echo path coupling amount calculation unit different from the echo path coupling amount calculation unit 35. 35 '.

Echo path coupling amount calculation unit 3 of this embodiment
5 ', as shown in FIG.
Remaining echo path coupling in a state other than the state where the echo path h (k) fluctuates in the double talk state (that is, the receiving state, the transmitting state, or the state where the echo path does not fluctuate in the double talk state) based on the control signal input from Quantity | H
d (ω) | ² (residual amplitude frequency function) is received signal x (k)
Short-time spectrum X ^{t = T} (ω) and echo canceller 6
Short-time spectrum E ^{t = T} (ω) of the output signal e (k) of 0
(That is, the short-time spectrum Bd (ω) of the residual echo signal bd (k)) is used for the calculation.

Further, the echo path coupling amount calculation unit 35 'is
Each of the above short-time spectra ^{Xt = T} (ω), E
During the period when ^{t = T} (ω) is input, the residual echo path coupling amount | Hd (ω) | ² from the amplitude characteristics | X ^{t = T} (ω) | and | E ^{t = T} (ω) | Are sequentially calculated and output to the echo signal power calculator 36. Also other than this
During the period of, the echo path coupling amount calculation unit 35 ′ calculates the most recently calculated residual echo path coupling amount | Hd unless a control signal is input from the echo path variation detection unit 62.
(Ω) | ² is output to the echo signal power calculation unit 36.

Further, the echo path coupling amount calculation unit 35 'is
When the control signal indicating the double talk state and the state in which the echo path changes is input from the echo path change detection unit 62, all processing is stopped and the echo signal power calculation unit 3
The output of the residual echo path coupling amount | Hd (ω) | ² to 6 is stopped. Therefore, the residual echo path coupling amount | Hd
(Ω) | ² is in the double talk state and the echo path h
It is supplied to the echo signal power calculation unit 36 only when the echo path h (k) does not fluctuate in a state other than the state in which (k) fluctuates, that is, in the receiving state, the transmitting state, or the double-talk state.

Such an echo path coupling amount calculator 35 '
The echo suppressing unit 61 including the received signal x (k) and the echo reduction signal e output from the echo canceller 60.
The time signal se (k) is generated based on (k) and output to the transmission path. That is, the echo suppressing unit 61 regards the echo reduction signal e (k) as the transmission signal y (k) of the first embodiment, and also the residual echo signal bd (k) superimposed on the echo reduction signal e (k). Is the echo signal b of the first embodiment
(K), and the path g (k) in which the echo canceller 60 is included in the echo path h (k) is regarded as the echo path h (k) of the first embodiment, and the residual echo is the same as in the first embodiment. A time signal se (k) is generated by suppressing the signal bd (k).

The one-sided utterance state detecting section 64 determines the transmission signal y.
An echo reduction amount, which is the ratio of (k) to the echo reduction signal e (k), is calculated. Then, when it is determined that the echo reduction amount monotonically increases with time in the presence of the reception signal x (k) based on the echo reduction amount, the transmission signal s is included in the transmission signal y (k). It is determined that (k) is not included (one-sided utterance state), and in other states, it is determined that the transmission signal s (k) is included (double talk state). The determination result is output as a control signal to the echo path coupling amount calculation reception signal selection unit 46 and the echo path coupling amount calculation transmission signal selection unit 47 in the echo suppression unit 61.

In this embodiment, since the one-sided utterance state detecting section 64 is provided, the transmission signal y (k) is more accurately converted into the transmission signal s as compared with the first embodiment.
A state in which (k) is not included can be detected.

The echo path variation detector 62 receives the received signal x
(K), transmission signal y (k) and echo attenuation signal e (k)
On the basis of the received signal x (k) and the transmitted signal s (k) at the same time, the fluctuation of the echo path h (k) is detected, and the detection result is used as a control signal for the echo path coupling amount. Calculation unit 35 'and echo canceller 60
And the echo path estimation circuit 23 'in the inside.

When the echo path estimation circuit 23 'receives a control signal indicating that the echo path h (k) has changed in the double talk state from the echo path change detector 62, the echo path estimation circuit 23' estimates its own pseudo echo path he. All values of (k) are set to "0". As a result, the pseudo echo path convolution operation circuit 22 stops the generation of the pseudo echo signal bd (k), so that the subtractor 24 uses the transmission signal y (k) input from the transmission microphone 3 as it is in the echo suppressing unit. 61
And output to.

Further, when the echo path estimation circuit 23 'receives a control signal indicating that the echo path h (k) has changed in the double talk state from the echo path change detector 62, the detection result is input. The pseudo echo path he (k) estimated immediately before is output to the echo path coupling amount calculation unit 63.

The echo path coupling amount calculator 63 is a pseudo echo path which is the square value of the amplitude characteristic of the pseudo echo path frequency characteristic He (ω) obtained by performing the fast Fourier transform on the pseudo echo path he (k). The coupling amount | He (ω) | ² (pseudo amplitude frequency function) is output to the echo signal power calculation unit 36 in the echo suppression unit 61. That is, when the echo path h (k) fluctuates in the double talk state , the echo path coupling amount calculator 63 calculates the pseudo echo path coupling amount | He (from the pseudo echo path he (k) estimated immediately before the fluctuation. ω) | ² is calculated and output to the echo signal power calculation unit 36.

As a result, the echo signal power calculation unit 36 outputs the residual echo path coupling amount | Hd (ω) | ² when the echo path h (k) does not change in the receiving state, the transmitting state or the double talk state. When the echo path h (k) is input and fluctuates in the double talk state , the pseudo echo based on the pseudo echo path he (k) estimated by the echo path estimation circuit 23 ′ immediately before the fluctuation. The path coupling amount | He (ω) | ² is alternatively input. The echo signal power calculation unit 36 predicts an echo based on the residual echo path coupling amount | Hd (ω) | ² or the pseudo echo path coupling amount | He (ω) | ² that is alternatively input according to the call situation. The signal power | E ^t e (ω) | ² is calculated.

In the double talk state, the echo path h
When (k) fluctuates, the operation of the pseudo echo path convolution operation circuit 22 is stopped and the transmission signal y (k) from the microphone 3 for transmission is input to the echo suppressing unit 61 as it is. That is, the echo suppressor 61 operates in a state of passing the echo canceller 60.

FIG. 7 is a simulation result showing the effect of this embodiment. In this simulation, the echo path estimation circuit 23 '(adaptive filter) uses the NLMS algorithm, and the number of taps is 512. Further, F in each fast Fourier transform of the echo suppressor 61
The FT score was also set to 512.

In FIG. 7, (a) shows the received signal x
(K) and (b) are the transmission signal s (k), and (c) is a signal obtained by subtracting the transmission signal s (k) from the echo reduction signal e (k) (= bd (k) + s (k)). (That is, the residual echo signal bd
(K) and (d) are output signals se of the echo suppressor 61.
A signal obtained by subtracting the transmission signal s (k) from (k) is shown. In addition, these (a) to (d) are each in the receiving state (Reseive) → the transmitting state (Send) with the passage of time.
→ Double-talk state (Double-talk) → Echo path change (Path Change) → Receiving state (Reseive).

As shown in this figure, it can be seen that the echo suppressor of this embodiment effectively suppresses the echo signal in each call state. In particular, comparing (c) and (d), the echo suppressor 61 shows that the residual echo signal bd (k) included in the echo reduction signal e (k) in the double-talk state without fluctuation of the echo path.
Of the residual echo signal bd (k) even when the echo path changes in the double talk state.
It can be seen that the increase in is suppressed.

[0137]

As described above, according to the echo suppression method and apparatus and the computer-readable storage medium storing the echo suppression program according to the present invention, the following effects can be obtained. (1) The received signal x (k) propagates along the echo path of the echo path transfer function h (k) that changes with time, and the echo signal b
The transmission signal y superimposed on the transmission signal s (k) as (k)
In the method of suppressing the echo signal b (k) superposed on the transmission signal y (k) with respect to (k), the reception signal x
(K) and the transmission signal y (k) are each converted into a short-time spectrum, and the short-time spectrum of the transmission signal y (k) and the reception signal x (k) in a state where the transmission signal s (k) is not included. Of the echo path transfer function h (k) is calculated by using the short-time spectrum of the echo path transfer function h (k), the short-time spectrum of the transmission signal y (k), and the short-time spectrum of the reception signal x (k). Is used to estimate the amplitude characteristic value of the short-time spectrum of the echo signal b (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) at the following times, and the echo characteristic b (k) of the echo signal b (k) is estimated. The echo suppression gain G (is calculated using an arithmetic expression based on the amplitude characteristic value of the short-time spectrum, the amplitude characteristic value of the short-time spectrum of the transmission signal s (k), and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k). ω Is calculated, the echo suppression gain G
Since the echo signal b (k) is suppressed by multiplying the short-time spectrum of the transmission signal y (k) by (ω), the transmission voice is distorted with respect to the fluctuation of the echo path during double talk, It is possible to suppress the echo sufficiently in the sense of hearing and keep the communication quality in good condition without causing sound interruption. (2) The received signal x (k) propagates in the echo path of the echo path transfer function h (k) that changes with time, and the echo signal b
The transmission signal y superimposed on the transmission signal s (k) as (k)
In the method of suppressing the echo signal b (k) superposed on the transmission signal y (k) with respect to (k), the reception signal x
(K) and the transmission signal y (k) are converted into short-time spectra, and the echo path transfer function h during double talk is converted.
The fluctuation of (k) is detected, and the echo path transfer function h (k) is calculated based on the echo reduction signal e (k) and the received signal x (k).
The pseudo echo path transfer function he (k), which is the estimated value of, is calculated, and the echo signal b (k) included in the transmission signal y (k) is suppressed by using the pseudo echo path transfer function he (k). When the echo reduction signal e (k) in which the residual echo signal bd (k) is included in the transmission signal s (k) is generated and the fluctuation of the echo path transfer function h (k) during double talk is detected. , The pseudo echo path transfer function he before the fluctuation detection
When the pseudo amplitude frequency function | He (ω) | ² is calculated from (k) and the fluctuation of the echo path transfer function h (k) during double talk is not detected, the transmission signal s (k) is not included. The residual amplitude frequency function | Hd (ω) | ² is calculated by using the short-time spectrum of the transmission signal y (k) and the short-time spectrum of the reception signal x (k) in the state, and the residual amplitude frequency function | Hd (ω) | ² is calculated alternatively. Estimate the amplitude characteristic value of the short-time spectrum of the echo signal b (k) or the residual echo signal bd (k) based on the pseudo amplitude frequency function | He (ω) | ² or the residual amplitude frequency function | Hd (ω) | ^2. The echo signal b
(K) or the amplitude characteristic value of the short-time spectrum of the residual echo signal bd (k), the amplitude characteristic value of the short-time spectrum of the transmission signal s (k), and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k). The echo suppression gain G (ω) is calculated by using an arithmetic expression based on Eq. 1, and when the fluctuation of the echo path transfer function h (k) during double talk is not detected, the echo suppression gain G (ω) is reduced by the echo. The residual echo signal bd (k) is suppressed by multiplying the short-time spectrum of the signal e (k), and when the fluctuation of the echo path transfer function h (k) during double talk is detected, the echo suppression gain Since the echo signal b (k) is suppressed by multiplying G (ω) by the short-time spectrum of the transmission signal y (k), the transmitted voice may be distorted with respect to the fluctuation of the echo path during double talk. Sound off Without, it is possible to further suppress the echo signal. Therefore, it is possible to further improve the call quality.

[Brief description of drawings]

FIG. 1 is a block diagram showing a functional configuration of an echo suppressor according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a configuration of a subjective evaluation experiment system according to the first embodiment of the present invention.

FIG. 3 is a characteristic diagram showing an experimental result of a subjective evaluation experiment according to the first embodiment of the present invention.

FIG. 4 is a block diagram showing a functional configuration of an echo suppressor according to a second embodiment of the present invention.

FIG. 5 is a block diagram showing a functional configuration of an echo canceller according to a second embodiment of the present invention.

FIG. 6 is a block diagram showing generation of residual echo path coupling amount and pseudo echo path coupling amount according to the second embodiment of the present invention.

FIG. 7 is an experimental result of computer simulation showing the effect of the second embodiment of the present invention.

FIG. 8 is an explanatory diagram showing a configuration of a voice call system.

FIG. 9 is a block diagram showing a configuration example of a conventional full-band echo canceller.

[Explanation of symbols]

1, 3 ... Transmitting microphones 2, 4 ... Receiving speakers 5, 7 ... Transmitting signal amplifiers 6, 8 ... Receiving signal amplifiers 9 ... Transmission line 10 ... Speaker 11 ... Receiver 21 ...... Echo canceller 22 …… Pseudo echo path convolution operation circuit 23 …… Echo path estimation circuit 24 …… Subtractor 25 …… Nonlinear echo suppression processing unit 31 …… Fast Fourier transform unit (received signal conversion means) 32 …… High speed Fourier transform unit (transmission signal conversion unit) 33 ... Fast inverse Fourier transform unit (transmission signal inverse conversion unit) 34 ... One-sided speech state detection unit 35 ... Echo path coupling amount calculation unit (echo path transfer function calculation unit) 36 ... Echo signal power calculation section (echo signal amplitude frequency characteristic estimation means) 37 ... Delay operator (echo signal amplitude frequency characteristic estimation means) 38 ... Echo signal power determination section (echo signal swing) Frequency characteristic estimating means) 39 ... Auditory correction gain storage section (echo signal amplitude frequency characteristic estimating means, echo signal auditory sense correcting means) 40 ... Auditory correction section (echo signal amplitude frequency characteristic estimating means, echo signal auditory sense correcting means) 41 ...... Smoothing factor storage section (transmitting signal amplitude frequency characteristic estimating means) 42 ...... Sending signal power calculating section (sending signal amplitude frequency characteristic estimating means) 43 ...... Delay operator (sending signal amplitude frequency characteristic estimating means) ) 44 ... Echo suppression gain determination section (transmitting signal amplitude frequency characteristic estimating means) 45 ...... Echo suppression processing section 46 ...... Echo path coupling amount calculation receiving signal selecting section (echo path transfer function calculating means) 47 ...... Echo path coupling amount calculation transmission signal selection unit (echo path transfer function calculation means) 23 '... Echo path estimation circuit (adaptive filter) 35' ... -Path coupling amount calculation unit (echo path transfer function calculation unit) 60 ... Echo canceller 61 ... Echo suppression unit 62 ... Echo path variation detection unit 63 ... Echo path coupling amount calculation unit (second echo path transfer function) Calculation means) 64 ... One-sided speech state detection unit b (k) ... Echo signal bd (k) ... Residual echo signal e (k) ... Echo reduction signal | ^{Et = T} e (ω) | ² ...... Echo signal power at time T | E ^t e (ω) | ² , | E ′ ^t e (ω) | ² Predicted echo signal power F (ω) …… Hearing correction gain G (ω) …… Echo suppression gain | He (ω) | ² ... Echo path coupling amount, pseudo-echo path coupling amount | Hd (ω) | ² ... Remaining echo path coupling amount s (k) ... Transmitting signal se (k) ... Time signal S ^t e (ω) ... Processed signal | S ^t e (ω) | ² ... Predicted transmission signal power x (k) ... Reception signal X ^t (ω) ... Short-time spectrum of the reception signal x (k) at time ^t | X ^t (ω) | ² ... Reception signal power y (k) ... Transmission signal Y ^t (ω) ... Time t Spectrum of transmitted signal y (k) at | Y ^t (ω) | ² ... Transmitted signal power β ... Smoothing factor

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References Sumiyu Sakauchi, Yoichi Haneda, “Study on Nonlinear Echo Suppression Processing Based on Short-time Spectral Amplitude Estimation”, The Acoustical Society of Japan, 1998 Spring Research Presentation Proceedings I, March 17, 1998, pp. 551-552, (3-5-11) Sumiuchi Sakauchi, Junko Sasaki, Yoichi Haneda, "Combination of Nonlinear Echo Suppression Based on STSA Estimation and Adaptive Filter", The Acoustical Society of Japan, Spring-Autumn Research 1998 Presentation Meeting Proceedings I, September 24, 1998, pp. 607-608, (1-P-6) (58) Fields investigated (Int.Cl. ⁷ , DB name) H04B 3/00 H04B 7/00 H04M 3/00 H04R 3/00

Claims (34)

(57) [Claims] 1. A transmission signal in which a reception signal x (k) propagates in an echo path of a time-varying echo path transfer function h (k) and is superimposed on a transmission signal s (k) as an echo signal b (k). A method of suppressing an echo signal b (k) superimposed on the transmission signal y (k) with respect to y (k), in which a reception signal x (k) and a transmission signal y (k) are converted into a short-time spectrum. The transmission signal y in the state where the transmission signal s (k) is not converted is converted.
The echo path transfer function h (k) is calculated using the short-time spectrum of (k) and the short-time spectrum of the received signal x (k).
Of the echo signal b (k) at a subsequent time using the amplitude frequency function of the received signal x (k) and the short-time spectrum of the transmission signal y (k). The amplitude characteristic value of the short-time spectrum and the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) are estimated, and the amplitude characteristic value of the short-time spectrum of the echo signal b (k) and the transmission signal s (k) are estimated. Of the short-time spectrum and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k) are used to calculate the echo suppression gain G (ω), and the echo suppression gain G (ω) is calculated. The echo signal b (k) is obtained by multiplying the short-time spectrum of the transmission signal y (k).
An echo suppression method characterized by suppressing the noise. 2. A transmission signal in which a reception signal x (k) propagates in an echo path of a time-varying echo path transfer function h (k) and is superimposed on a transmission signal s (k) as an echo signal b (k). A method of suppressing an echo signal b (k) superimposed on the transmission signal y (k) with respect to y (k), in which a reception signal x (k) and a transmission signal y (k) are converted into a short-time spectrum. Echo path transfer function h (k) estimation based on echo reduction signal e (k) and received signal x (k) by detecting the fluctuation of echo path transfer function h (k) during double talk. The pseudo echo path transfer function he (k), which is a value, is calculated, and the echo signal b (k) included in the transmission signal y (k) is suppressed by using the pseudo echo path transfer function he (k). The echo reduction signal in which the residual echo signal bd (k) is included in the signal s (k). When the fluctuation of the echo path transfer function h (k) during double talk is detected by generating e (k), the pseudo amplitude path function | He from the pseudo echo path transfer function he (k) before the fluctuation detection is generated. (Ω) ｜ ²
And the fluctuation of the echo path transfer function h (k) during double talk is not detected, the short-time spectrum of the transmission signal y (k) and the reception of the reception signal s (k) are not included. The residual amplitude frequency function | Hd (ω) | ² is calculated using the short-time spectrum of the signal x (k), and the pseudo amplitude frequency function | He calculated as an alternative is calculated.
(Ω) | ² or residual amplitude frequency function | Hd (ω) | ²
The amplitude characteristic value of the short-time spectrum of the echo signal b (k) or the residual echo signal bd (k) is estimated based on the above, and the echo signal b (k) or the residual echo signal bd is estimated.
(K) Amplitude characteristic value of short-time spectrum and transmission signal s
(K) Amplitude characteristic value of short-time spectrum and transmission signal y
When the echo suppression gain G (ω) is calculated by using the arithmetic expression based on the amplitude characteristic value of the short-time spectrum of (k), and the fluctuation of the echo path transfer function h (k) during double talk is not detected, The echo suppression gain G (ω)
To suppress the residual echo signal bd (k) by multiplying by the short-time spectrum of the echo reduction signal e (k), and when the fluctuation of the echo path transfer function h (k) during double talk is detected, An echo suppression method characterized by suppressing an echo signal b (k) by multiplying an echo suppression gain G (ω) by a short-time spectrum of a transmission signal y (k). 3. A transmission signal y (k) and an echo reduction signal e.
The reduction amount of the echo signal b (k) is obtained from (k) and the transmission signal s (k) is included when the reception signal x (k) is present and the reduction amount is monotonically increasing. The echo suppression method according to claim 2, wherein it is determined that there is no state. 4. The larger one of the amplitude characteristic value of the short-time spectrum of the echo signal b (k) and the amplitude characteristic value of the short-time spectrum of the echo signal b (k) one unit time before, whichever is larger. The echo suppression method according to claim 1, wherein the estimated value is an amplitude characteristic value of a short-time spectrum of the signal b (k). 5. The echo signal b is a value obtained by subjecting the amplitude characteristic value of the short-time spectrum of the echo signal b (k) to audibility correction.
The echo suppression method according to claim 1, wherein the estimated value of the amplitude characteristic value of the short-time spectrum of (k) is used. 6. The audibility correction gain is set based on the masking level of the echo signal by the received voice, the human audible level and the transmission delay amount, and the audibility correction gain is set to the echo signal b.
The echo suppression method according to claim 5, wherein the amplitude characteristic value of the short-time spectrum of (k) is multiplied to perform auditory sense correction. 7. The amplitude characteristic value of the short-time spectrum of the echo signal b (k), the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) one unit time before, and the short-time of the transmission signal y (k). The following equation (1) consisting of a spectrum and a smoothing factor β and a half-wave rectification function P for extracting only the positive component of the signal
The estimated value of the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) is calculated based on
6. The echo suppression method according to any one of 6 above. [Equation 1] 8. A smoothing factor β of 0.9 to 0.
The echo suppression method according to claim 7, wherein the value is set in a range of 999. 9. An arithmetic expression (2) consisting of a transmission signal power | S ^t e (ω) | ² and an echo signal power | E ^t e (ω) | ^2.
9. The echo suppression method according to claim 1, wherein the value G ^PE (ω) of is applied to the echo suppression gain G (ω). [Equation 2] 10. The value G ^W (ω) of the arithmetic expression (3) consisting of the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e (ω) | ² is the echo. Suppression gain G
The echo suppression method according to claim 1, wherein the echo suppression method is applied to (ω). [Equation 3] 11. An arithmetic expression (4) composed of a transmission signal power | S ^t e (ω) | ² and an echo signal power | E ^t e (ω) | ^2.
9. The echo suppression method according to claim 1, wherein the value G ^ML (ω) is applied to the echo suppression gain G (ω). [Equation 4] 12. A transmission signal power | S ^t e (ω) | ² and an echo signal power | E ^t e (ω) | ² and a transmission signal power | Y ^t
Intermediate variable υ based on equation (5) consisting of (ω) | ²
(Ω) is calculated, and the intermediate variable υ (ω), the echo signal power │E ^t e (ω) │ ² and the transmission signal power │Y ^t (ω) │ ² and the zero-order and the intermediate variable υ (ω) The value G ^EM (ω) of the arithmetic expression (6) including the modified Bessel functions I ₀ and I _{1 of the} first order
Is applied to the echo suppression gain G (ω). 9. The echo suppression method according to claim 1, wherein [Equation 5] [Equation 6] 13. A transmission signal in which a reception signal x (k) propagates in an echo path of a time-varying echo path transfer function h (k) and is superimposed on a transmission signal s (k) as an echo signal b (k). An echo suppressor for suppressing the echo signal b (k) superimposed on the transmission signal y (k) with respect to y (k), the received signal conversion converting the received signal x (k) into a short-time spectrum. Means, transmission signal conversion means for converting the transmission signal y (k) into a short-time spectrum, and a transmission signal s (k) based on the reception signal x (k) and the transmission signal y (k). One-sided utterance state detection unit for detecting a non-existence state, and a transmission signal y in a state where the transmission signal s (k) is not included
The echo path transfer function h (k) is calculated using the short-time spectrum of (k) and the short-time spectrum of the received signal x (k).
Using the echo path transfer function calculating means for calculating the amplitude frequency function of, and the amplitude frequency function, the short-time spectrum of the transmission signal y (k) and the short-time spectrum of the reception signal x (k), Echo signal amplitude frequency characteristic estimating means for estimating the amplitude characteristic value of the short time spectrum of the echo signal b (k), the amplitude frequency function, the short time spectrum of the transmission signal y (k) and the reception signal x (k). A transmission signal amplitude frequency characteristic estimating means for estimating the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) at a subsequent time using the short-time spectrum, and the short-time spectrum of the echo signal b (k). By using an arithmetic expression based on the amplitude characteristic value of the short-term spectrum of the transmission signal s (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k). An echo suppression gain determination unit that calculates the Cau suppression gain G (ω), an echo suppression processing unit that multiplies the echo suppression gain G (ω) by the short-time spectrum of the transmission signal y (k), and the echo suppression processing unit. And a transmission signal inverse conversion means for inversely converting the output of the above and outputting it to the transmission path. 14. A transmission signal in which a reception signal x (k) propagates along an echo path of a time-varying echo path transfer function h (k) and is superimposed on a transmission signal s (k) as an echo signal b (k). An echo suppressor for suppressing the echo signal b (k) superimposed on the transmission signal y (k) with respect to y (k), the received signal conversion converting the received signal x (k) into a short-time spectrum. Means, transmission signal conversion means for converting the transmission signal y (k) into a short-time spectrum, and a transmission signal s (k) based on the reception signal x (k) and the transmission signal y (k). A one-sided utterance state detection unit that detects a non-existing state, an echo path fluctuation detection unit that detects fluctuations in the echo path transfer function h (k) during double talk, and an echo filter transfer function h (k) Pseudo echo path transfer function he (k ) Is generated, the echo signal b (k) included in the transmission signal y (k) is suppressed by using the pseudo echo path transfer function he (k), and the residual echo signal bd (is included in the transmission signal s (k). When the echo reduction signal e (k) containing k) is output and the echo path variation detecting section detects a variation in the echo path transfer function h (k) during double talk, The echo canceller that outputs the pseudo echo path transfer function he (k) and the transmission signal y (k) as it is, and the fluctuation of the echo path transfer function h (k) during the double talk by the echo path fluctuation detection unit. If not detected, the residual amplitude frequency function | based on the short-time spectrum of the echo reduction signal e (k) and the short-time spectrum of the reception signal x (k) when the transmission signal s (k) is not included. Hd (ω ) | ² for calculating an echo path transfer function calculating means, and the echo path change detecting section detects a change in the echo path transfer function h (k) during double talk,
Second echo path transfer function calculating means for calculating a pseudo amplitude frequency function | He (ω) | ² from the pseudo echo path transfer function he (k) output from the echo canceller; and echo path transfer function calculating means or second Echo signal b (k) or residual signal based on the pseudo-amplitude frequency function | He (ω) | ² or residual amplitude frequency function | Hd (ω) | ² which is alternatively input from the echo path transfer function calculating means Echo signal amplitude frequency characteristic estimation means for estimating the amplitude frequency characteristic of the short time spectrum of the echo signal bd (k), and the amplitude frequency characteristic and the short time spectrum of the echo reduction signal e (k) or the transmission signal y (k) Using the short-time spectrum of the received signal x (k), the residual echo signal bd (k) or the echo signal b at the subsequent time
(K) Echo signal amplitude frequency characteristic estimating means for estimating the amplitude frequency characteristic of the short time spectrum, the amplitude frequency characteristic and the short time spectrum of the echo reduction signal e (k) or the transmission signal y (k) and the reception signal A transmission signal amplitude frequency characteristic estimating means for estimating the amplitude frequency characteristic of the short time spectrum of the transmission signal s (k) at a subsequent time using the short time spectrum of x (k), and the residual echo signal bd ( k) or echo signal b
(K) Amplitude frequency characteristic of short-time spectrum, Amplitude frequency characteristic of short-term spectrum of transmission signal s (k), and Amplitude-frequency characteristic of short-term spectrum of echo reduction signal e (k) or transmission signal y (k) And an echo suppression gain G (ω) for calculating the echo suppression gain G (ω) by using an arithmetic expression based on Eq.
Alternatively, an echo suppression processing unit that multiplies the short-time spectrum of the transmission signal y (k), and a transmission signal inverse conversion unit that inversely converts the output of the echo suppression processing unit and outputs to the transmission path are provided. Characteristic echo suppressor. 15. The one-sided utterance state detector is configured to transmit a signal y.
(K) and the echo reduction signal e (k), the echo signal b
The reduction amount of (k) is calculated, and if the reception signal x (k) is present and the reduction amount is monotonically increasing, the transmission signal s
15. The echo suppressor according to claim 14, wherein it is determined that (k) is not included. 16. The echo signal amplitude frequency characteristic estimating means uses the amplitude frequency function, the short time spectrum of the transmission signal y (k) and the short time spectrum of the reception signal x (k), and uses the short time spectrum of the echo signal at a subsequent time. An echo signal power calculation unit that estimates the amplitude characteristic value of the short-time spectrum of b (k), a first delay operator that delays and outputs this amplitude characteristic value by one unit time, an echo signal power calculation unit, and a 14. The echo signal power determination unit that uses the larger one of the outputs of the delay operator of 1 as the estimated value of the amplitude characteristic value of the short-time spectrum of the echo signal b (k). 15. The echo suppressor according to any one of 15. 17. The echo signal amplitude / frequency characteristic estimating means comprises echo signal audibility correcting means for audibly correcting the amplitude characteristic value of the short-time spectrum of the echo signal b (k). The echo suppressor according to any one of claims. 18. The audibility correction gain storage unit, which stores audibility correction gains set based on a masking level of an echo signal by a received voice, a human audible level, and a transmission delay amount, and the audibility correction unit. 18. The echo suppressor according to claim 17, further comprising: an auditory sense correction unit that multiplies a correction gain by an amplitude characteristic value of a short-time spectrum of the echo signal b (k). 19. The transmission signal amplitude / frequency characteristic estimating means includes a smoothing factor storage unit for storing a smoothing factor β, and an amplitude characteristic value of a short-time spectrum of the transmission signal s (k) is set to 1
A second delay operator for delaying by a unit time, an output of the second delay operator, an amplitude characteristic value of a short-time spectrum of the echo signal b (k), a short-time spectrum of the transmission signal y (k), and a smoothing factor. Transmission signal power calculation for calculating an estimated value of the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) based on the following equation (1) consisting of β and a half-wave rectification function P for extracting only the positive component of the signal 19. The echo suppressor according to claim 13, wherein the echo suppressor comprises: [Equation 7] 20. The smoothing factor β is 0.9 to
20. The echo suppressor according to claim 19, wherein the value is within a range of 0.999. 21. The echo suppression gain determination unit sets the transmission signal power | S ^t e (ω) as the echo suppression gain G (ω).
20. The value G ^PE (ω) is calculated based on an arithmetic expression (2) consisting of | ² and echo signal power | E ^t e (ω) | ² . Echo suppressor. [Equation 8] 22. The echo suppression gain determination unit sets the transmission signal power | S ^t e (ω) as the echo suppression gain G (ω).
20. The value G ^W (ω) is calculated based on an arithmetic expression (3) consisting of | ² and echo signal power | E ^t e (ω) | ² . Echo suppressor. [Equation 9] 23. The echo suppression gain determination unit sets the transmission signal power | S ^t e as the echo suppression gain G (ω).
20. The value G ^ML (ω) is calculated based on the arithmetic expression (4) consisting of (ω) | ² and the echo signal power | E ^t e (ω) | ² . The echo suppressor according to. [Equation 10] 24. The echo suppression gain determination unit is the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e
(Ω) | ² and transmission signal power | Y ^t (ω) | ^{2 is used} to calculate an intermediate variable υ (ω), and the intermediate variable υ (ω) and echo signal power | E ^t Based on e (ω) | ² and the transmission signal power | Y ^t (ω) | ² and the modified Bessel functions I ₀ and I _{1 of the 0th} and 1st order with respect to the intermediate variable υ (ω), The value G ^EM (ω) is calculated as an echo suppression gain G (ω).
20. The echo suppressor according to any one of 19 to 20. [Equation 11] [Equation 12] 25. A transmission signal in which a reception signal x (k) propagates in an echo path of a time-varying echo path transfer function h (k) and is superimposed on a transmission signal s (k) as an echo signal b (k). It is a computer-readable storage medium that stores an echo suppression program for suppressing the echo signal b (k) superimposed on the transmission signal y (k) for y (k). ) And a process of converting the transmission signal y (k) into a short-time spectrum, and a transmission signal y in a state where the transmission signal s (k) is not included.
The echo path transfer function h (k) is calculated using the short-time spectrum of (k) and the short-time spectrum of the received signal x (k).
Of the echo frequency signal b (k) at a subsequent time using the process of calculating the amplitude frequency function of the received signal, the short time spectrum of the transmission signal y (k) and the short time spectrum of the received signal x (k). ) Of estimating the amplitude characteristic value of the short time spectrum and the amplitude characteristic value of the short time spectrum of the transmission signal s (k), and the amplitude characteristic value of the short time spectrum of the echo signal b (k) and the transmission signal a process of calculating an echo suppression gain G (ω) using an arithmetic expression based on the amplitude characteristic value of the short-time spectrum of s (k) and the amplitude characteristic value of the short-time spectrum of the transmission signal y (k); The echo signal b (k) is obtained by multiplying the short-time spectrum of the transmission signal y (k) by the gain G (ω).
A computer-readable storage medium in which an echo suppression program is stored, the process comprising: 26. An amplitude characteristic value of a short time spectrum of the echo signal b (k) and the echo signal b (k) one unit time before
26. The echo according to claim 25, further comprising: a process of using the larger one of the amplitude characteristic values of the short-time spectrum as the estimated value of the amplitude characteristic value of the short-time spectrum of the echo signal b (k). A computer-readable storage medium storing a suppression program. 27. The echo signal b is a value obtained by subjecting the amplitude characteristic value of the short-time spectrum of the echo signal b (k) to auditory correction.
27. The computer-readable storage medium storing the echo suppression program according to claim 25 or 26, further comprising: (k) a process of using an estimated value of the amplitude characteristic value of the short-time spectrum. 28. An audibility correction gain is set based on a masking level of an echo signal by a received voice, an audible level of human beings and a transmission delay amount, and the audibility correction gain is set to an amplitude characteristic of a short-time spectrum of an echo signal b (k). 28. The computer-readable storage medium storing the echo suppression program according to claim 27, comprising a process of multiplying a value to correct the auditory sense. 29. The amplitude characteristic value of the short-time spectrum of the echo signal b (k), the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) one unit time before, and the short-time of the transmission signal y (k). The estimated value of the amplitude characteristic value of the short-time spectrum of the transmission signal s (k) is calculated based on the following equation (1) consisting of the spectrum and the smoothing factor β and the half-wave rectification function P for extracting only the positive component of the signal. A computer-readable storage medium storing the echo suppression program according to any one of claims 25 to 28, which further comprises processing. [Equation 13] 30. The smoothing factor β is 0.9 to
30. The computer-readable storage medium storing the echo suppression program according to claim 29, further comprising processing for setting a value within a range of 0.999. 31. The value G ^PE (ω) of the arithmetic expression (2) consisting of the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e (ω) | ² Suppression gain G
3. A process for applying to (ω) is included.
A computer-readable storage medium storing the echo suppression program according to any one of 5 to 30. [Equation 14] 32. The value G ^W (ω) of the arithmetic expression (3) consisting of the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e (ω) | ² is set to the echo. Suppression gain G
3. A process for applying to (ω) is included.
A computer-readable storage medium storing the echo suppression program according to any one of 5 to 30. [Equation 15] 33. The value G ^ML (ω) of the arithmetic expression (4) consisting of the transmission signal power | S ^t e (ω) | ² and the echo signal power | E ^t e (ω) | ² is set to the echo. Suppression gain G
3. A process for applying to (ω) is included.
A computer-readable storage medium storing the echo suppression program according to any one of 5 to 30. [Equation 16] 34. Transmission signal power | S ^t e (ω) | ² and echo signal power | E ^t e (ω) | ² and transmission signal power | Y ^t
Intermediate variable υ based on equation (5) consisting of (ω) | ²
(Ω) is calculated, and the intermediate variable υ (ω), the echo signal power │E ^t e (ω) │ ² and the transmission signal power │Y ^t (ω) │ ² and the zero-order and the intermediate variable υ (ω) The value G ^EM (ω) of the arithmetic expression (6) including the modified Bessel functions I ₀ and I _{1 of the} first order
31. A computer-readable storage medium in which the echo suppression program according to any one of claims 25 to 30 is stored, the processing including applying to the echo suppression gain G (ω). [Equation 17] [Equation 18]