JP4821635B2 - Signal state detection device, echo canceller, and signal state detection program - Google Patents

Description

  The present invention relates to a signal state detection device and a signal state detection program, and can be applied to, for example, an echo canceller and a noise canceller. Another invention relates to an echo canceller.

  One of the factors that deteriorates the quality of calls during a call at a telephone terminal is echo. A method often used for preventing echo is a method of providing an echo canceller in a telephone terminal. The echo canceller generates a pseudo echo signal by applying adaptive filter processing to the signal transmitted from the far end side, and removes the echo by subtracting the pseudo echo signal from the wraparound echo. . FIG. 4 is a diagram showing a configuration in which the line echo by the hybrid circuit is canceled by the echo canceller. FIG. 5 is a diagram showing a configuration in which an acoustic echo due to the acoustic space is erased by an echo canceller.

  However, the adaptive filter of the echo canceller has a problem that if the sound is present in both the near-end side and the far-end side, the filter coefficient diverges and the echo cannot be canceled. is there. Therefore, it is determined whether or not the device is in the double talk state. If the state is the double talk state, the update operation of the filter coefficient of the adaptive filter is stopped, and if it is not the double talk state (for example, in the single talk state). The method of continuing the filter coefficient updating operation is taken. Thus, as a specific method for detecting the double talk state, a double talk state detection method as disclosed in Patent Document 1 has been proposed.

FIG. 6 is a block diagram showing a functional configuration of a conventional echo canceller 600 that realizes the method for detecting the double talk state described in Patent Document 1. In FIG. In the double-talk detection unit 601, the correlation matrix calculation unit 601a calculates a correlation matrix (each element) such as the following equation (1), and the determinant calculation unit 601b calculates the matrix of the obtained correlation matrix. Calculate the formula. Based on the obtained determinant and a predetermined threshold value stored in the determination threshold value storage unit 601c, the state determination unit 601d determines whether or not it is in a double talk state. Since the value of the determinant of the correlation matrix shown in the equation (1) has a very large peak value in the double talk state as compared to the non-double talk state, and the state continues, It is possible to detect a double talk state.

  In addition, the meaning of the symbol in said (1) Formula is as showing in FIG. The echo source signal (far end input signal) is S_0, the echo is S_1, the double talk signal (near end talker signal) is D, the transmission signal (sum of echo and double talk signal) is P (= S_1 + D), adaptive filter Is represented by w, and in the equation (1), each is a horizontal vector. The pseudo echo signal is shown by the above equation (2) (convolution of w and S_0). In this specification, “near end” and “far end” are expressed according to the distance from the echo canceller.

Further, in the 2 × 2 correlation matrix in the equation (1), the element in the first row and the first column is the autocorrelation of the transmission signal P, and the elements in the first row and the second column and the second row and the first column are the pseudo echo signal and the transmission signal. The element in 2 rows and 2 columns represents the autocorrelation of the pseudo echo signal.
JP-T-2004-517579

  However, the double talk detection unit 601 of the conventional echo canceller 600 uses the echo determination because the state determination unit 601d determines the double talk state based on the fixed threshold stored in the determination threshold storage unit 601c. When the values of various input signals to the canceller 600 change, it is not determined that the double talk state is set in spite of the double talk state, or the non-double talk state is set in spite of the non-double talk state. There has been a problem that misjudgment such as not being judged may occur.

  The state determination unit 601d detects the double talk state when the value of the determinant input from the determinant calculation unit 601b exceeds a predetermined threshold value. There is a problem that detection cannot be performed until the value of the expression exceeds a threshold value, and the time from detection to detection after the double talk state may be prolonged.

  Therefore, a signal state detection device, an echo canceller, and a signal state detection program that can detect a signal state without depending on fluctuations in the input signal level or that can quickly detect the change in signal state are desired.

The signal state detection apparatus according to the first aspect of the present invention includes (1) an unnecessary signal erasure target signal including an unnecessary signal which is an unnecessary signal other than the main signal and the main signal, and the unnecessary signal is erased from the unnecessary signal erasure target signal. In order to do so, a correlation calculation unit that calculates a value indicating the degree of correlation with the pseudo unnecessary signal formed using the adaptive filter, and (2) a change trend that calculates a change trend of the value calculated by the correlation calculation unit State detection for determining whether the state related to the unnecessary signal erasure target signal is a non-double talk state or a double talk state based on the calculation means, and (3) the change tendency calculated by the change tendency calculation means and the threshold value and means, (4) calculating the correlation calculating means calculates a correlation matrix of the unnecessary signal erased signal and the pseudo unwanted signal, the value of the determinant of the calculated correlation matrix, as a value representing a correlation degree And ( 5) The state detection means determines that the state is a double talk state when the gradient value of the determinant value calculated by the correlation calculation means exceeds a threshold value .

Echo canceller of the second present invention has a signal state detection device of the first present invention as a double-talk detecting apparatus, in the double-talk detecting apparatus, and inputs a pseudo echo signal as the pseudo unwanted signals, A transmission signal is input as the unnecessary signal erasure target signal.

The signal state detection program according to the third aspect of the present invention provides a computer that (1) the unnecessary signal including the unnecessary signal including the main signal and the unnecessary signal other than the main signal, and the unnecessary signal from the unnecessary signal erasing target signal. A correlation calculating means for calculating a value indicating the degree of correlation with a pseudo unnecessary signal formed by using an adaptive filter in order to eliminate the signal; and (2) calculating a change tendency of the value calculated by the correlation calculating means. a change trend calculating unit that, (3) and changing trend calculated by the change trend calculating unit, based on the threshold value, the state according to the unnecessary signal erased signal whether non double-talk state or a double talk state as a determining condition detecting means is caused to function, and (4) the correlation calculating means calculates a correlation matrix of the unnecessary signal erased signal and the pseudo unwanted signal, the value of the determinant of the calculated correlation matrix, the correlation Calculated as a value representing a slip, (5) the state detecting means, when the gradient value of the values of the determinant of the correlation calculating means has calculated exceeds the threshold value, and wherein the determining a double-talk state .

  According to the present invention, the signal state can be detected without depending on the fluctuation of the input signal level, or it can be quickly detected when the signal state changes.

(A) First Embodiment Hereinafter, a first embodiment of a signal state detection device, an echo canceller, and a signal state detection program according to the present invention will be described in detail with reference to the drawings. Note that the signal state detection device and the signal state detection program of the first embodiment are a double talk detection device and a double talk detection program.

(A-1) Configuration of the First Embodiment FIG. 2 is a block diagram showing the configuration of the double talk detection device according to the first embodiment, together with the configuration of the echo canceller according to the first embodiment.

  The echo canceller 100 includes a double talk detection unit 101, an adaptive filter control unit 102, and an adaptive filter unit 103.

  The adaptive filter unit 103 performs adaptive filter processing or the like on the signal transmitted from the far end side to generate a pseudo echo signal, and inputs the pseudo echo signal to the subtractor 104 and the double talk detection unit 101.

  The subtracter 104 subtracts the pseudo echo signal generated by the adaptive filter unit 103 from the transmission signal (near-end input signal: signal from the near-end side to the far-end side) to cancel the echo.

  The adaptive filter control unit 102 controls the adaptive operation of the adaptive filter unit 103, and determines the control content using the signal input from the double talk detection unit 101. For example, when a signal indicating that a double talk state has been detected is input from the double talk detection unit 101, the adaptive filter control unit 102 controls the adaptive filter unit 103 to stop the update of the filter coefficient and to perform non-double When a signal indicating that the talk state has been detected is input, updating of the filter coefficient is continued or resumed.

  The double talk detection unit 101 detects a double talk state or a non-double talk state, and includes a correlation matrix calculation unit 101a, a determinant calculation unit 101b, a determination threshold storage unit 101c, a state determination unit 101d, and a determination threshold normalization unit. 101e. The double talk detection unit 101 can be constructed by installing a double talk detection program in an information processing apparatus such as a personal computer or an embedded system, for example, but can be functionally shown in FIG. Note that the double talk detecting unit 101 may be configured by hardware logic or the like.

  The correlation matrix calculation unit 101a calculates, for example, a correlation matrix (each element) as in the above equation (1) based on the transmission signal and the pseudo echo signal input from the adaptive filter unit 103, and the matrix This is input to the formula calculation unit 101b and the determination threshold value normalization unit 101e.

  The determinant calculation unit 101b calculates the determinant of the correlation matrix input from the correlation matrix calculation unit 101a, and inputs the obtained determinant (value) to the state determination unit 101d.

  In the case of the first embodiment, the correlation matrix calculation unit 101a and the determinant calculation unit 101b serve as correlation calculation means for capturing the degree of correlation between the transmitted signal and the pseudo echo signal.

  The determination threshold value normalization unit 101e normalizes the threshold value stored in the determination threshold value storage unit 101c based on the information of the correlation matrix (elements) input from the correlation matrix calculation unit 101a, and normalizes the threshold value to the state determination unit 101d. The converted threshold value is input. As a method for the determination threshold value normalization unit 101e to normalize the threshold value, for example, the value stored in the determination threshold value storage unit 101c is used as the 1-by-1 element of the correlation matrix obtained by the correlation matrix calculation unit 101a ( A method of dividing by the above formula (1)) can be mentioned.

  Although not shown, as a method for normalizing the threshold, an echo source signal (far-end input signal: signal from the far-end side to the near-end side) is acquired and a value based on the signal level is obtained. And a method of dividing by a value based on the sum of the signal levels of the transmission signal and the echo source signal.

  The determination threshold storage unit 101c stores a target threshold to be normalized by the determination threshold normalization unit 101e.

  The state determination unit 101d determines whether or not the state is a double talk state based on values input from the determinant calculation unit 101b and the determination threshold value normalization unit 101e. When the value of the determinant input from the determinant calculation unit 101b is larger than the normalized threshold value input from the determination threshold value normalization unit 101e, the state determination unit 101d judge.

(A-2) Operation of the First Embodiment Next, the detection operation of the double talk state in the echo canceller 100 of the first embodiment having the above configuration will be described.

  First, the correlation matrix calculation unit 101a calculates each element of the correlation matrix as in the above equation (1) based on the transmission signal and the pseudo echo signal, and the correlation matrix (each element) is calculated as the determinant calculation unit. 101b and the determination threshold value normalization unit 101e.

  When the correlation matrix is input, the determinant calculation unit 101b calculates the determinant of the input correlation matrix and inputs the calculation result to the state determination unit 101d.

  When the correlation matrix (element) is input from the correlation matrix calculation unit 101a, the determination threshold normalization unit 101e uses the threshold stored in the determination threshold storage unit 101c based on the input correlation matrix (element). The threshold value thus normalized is input to the state determination unit 101d.

  When the normalized threshold value is input from the determination threshold value normalization unit 101e and the determinant value is input from the determinant calculation unit 101b, the state determination unit 101d compares these values to determine whether or not the state is the double talk state. And the result is input to the adaptive filter control unit 102.

(A-3) Effects of First Embodiment According to the first embodiment, the double talk detection unit 101 includes the determination threshold value normalization unit 101e, so that the threshold value stored in the determination threshold value storage unit 101c is set. Then, the value is normalized to a value corresponding to the level fluctuation of the transmission signal and the like, and based on the value, the state determination unit 101d can determine whether or not it is in the double talk state. As a result, it is possible to correctly detect the double talk state regardless of the fluctuation state of the input signal or the like input to the echo canceller 100.

(B) Second Embodiment Hereinafter, a second embodiment of the signal state detection device, the echo canceller, and the signal state detection program according to the present invention will be described in detail with reference to the drawings. Note that the signal state detection device and the signal state detection program of the second embodiment are a double talk detection device and a double talk detection program.

(B-1) Configuration of Second Embodiment FIG. 3 is a block diagram showing the configuration of the double talk detection device according to the second embodiment, together with the configuration of the echo canceller according to the second embodiment.

  The echo canceller 200 includes a double talk detection unit 201, an adaptive filter control unit 202, and an adaptive filter unit 203.

  The adaptive filter control unit 202, the adaptive filter unit 203, and the subtractor 204 are the same as the adaptive filter control unit 102, the adaptive filter unit 103, and the subtractor 104 of the first embodiment, and thus description thereof is omitted.

  The double talk detection unit 201 detects whether the state is a double talk state or a non-double talk state. The correlation matrix calculation unit 201a, the determinant calculation unit 201b, the determination threshold storage unit 201c, the state determination unit 201d, and the determinant gradient calculation Part 201e.

  The double talk detection unit 201 can be constructed by installing a double talk detection program in an information processing apparatus such as a personal computer or an embedded system, and can be functionally shown in FIG. Note that the double talk detecting unit 101 may be configured by hardware logic or the like.

  Since the correlation matrix calculation unit 201a has substantially the same configuration as the correlation matrix calculation unit 101a of the first embodiment, a description thereof will be omitted. In the first embodiment, the correlation matrix calculation unit 101a inputs the correlation matrix obtained by the determinant calculation unit 101b and the determination threshold value normalization unit 101e. In the second embodiment, the determinant calculation The only difference is that it is input only to the part 201b.

  The determinant calculation unit 201b has substantially the same configuration as the determinant calculation unit 101b of the first embodiment, and thus the description thereof is omitted. The determinant calculating unit 201b is different from the determinant calculating unit 101b of the first embodiment in that the value of the obtained determinant is input to the determinant gradient calculating unit 201e.

The determinant gradient calculation unit 201e calculates the gradient (slope) of the change in the determinant (value) given from the determinant calculation unit 201b, and inputs the calculation result to the state determination unit 201d. For example, the rate of change (gradient) is calculated from the values of two determinants at different times. Here, the values of the two determinants do not matter when they are acquired, but the latest value may be applied as one of the values. The value calculated by the determinant gradient calculating unit 201e can be expressed by the following equation (3). Note that R represents the value of the determinant of the correlation matrix shown in the above equation (1) output from the determinant calculating unit 201b, and Δ | R | represents the value of the gradient output from the determinant gradient calculating unit 201e. Is. Also, t2> t1, t1 represents the time when the value was acquired from the determinant calculation unit 201b in the past, and t2 represents the time when the latest value was acquired.

  The determination threshold storage unit 201c stores a threshold for the state determination unit 201d to determine the double talk state.

  The state determination unit 201d determines whether or not the state is a double talk state based on the determinant gradient obtained by the determinant gradient calculation unit 201e and the value stored in the determination threshold storage unit 201c. The state determination unit 201d determines that the state is a double talk state when the gradient value input from the determinant gradient calculation unit 201e is larger than the threshold value stored in the determination threshold value storage unit 201c.

(B-2) Operation of Second Embodiment Next, the detection operation of the double talk state in the echo canceller 200 of the second embodiment having the above configuration will be described.

  First, in the correlation matrix calculation unit 201a, based on the transmission signal and the pseudo echo signal, each element of the correlation matrix like the above equation (1) is calculated, and the obtained correlation matrix (each element) is It inputs into the determinant calculation part 201b.

  When the correlation matrix is input, the determinant calculating unit 201b calculates the determinant and inputs the calculation result to the determinant gradient calculating unit 201e.

  When the determinant calculation result is input, the determinant gradient calculation unit 201e calculates the gradient value based on the past input value and the latest input value, and the result is the state determination unit. Input to 201d.

  When the gradient value is input, the state determination unit 201d compares the threshold value stored in the determination threshold value storage unit 201c with the gradient value input from the determinant gradient calculation unit 201e to determine whether the state is the double talk state. The result is input to the adaptive filter control unit 202.

(B-3) Effect of Second Embodiment According to the second embodiment, in the double talk detection unit 201, the determinant gradient calculation unit 201e is based on the value of the determinant calculated by the determinant calculation unit 201b. The change ratio is calculated as a gradient, and the state determination unit 201d determines whether or not the state is a double talk state by comparing with the value stored in the determination threshold storage unit 201c. Thereby, since the double talk state can be detected before the peak appears with respect to the change in the value of the determinant calculated by the determinant calculating unit 201b, the double talk state can be detected in a shorter time. It becomes possible.

(C) Third Embodiment Hereinafter, a third embodiment of the signal state detection device, the echo canceller, and the signal state detection program according to the present invention will be described in detail with reference to the drawings. Note that the signal state detection device and the signal state detection program of the first embodiment are a double talk detection device and a double talk detection program.

(C-1) Configuration of Third Embodiment FIG. 1 is a block diagram showing a configuration of a double talk detection device according to the third embodiment, together with a configuration of an echo canceller according to the third embodiment.

  The echo canceller 300 includes a double talk detection unit 301, an adaptive filter control unit 302, and an adaptive filter unit 303.

  The adaptive filter control unit 302, the adaptive filter unit 303, and the subtractor 304 are the same as the adaptive filter control unit 102, the adaptive filter unit 103, and the subtracter 104 of the first embodiment, and thus description thereof is omitted.

  The double talk detection unit 301 detects whether the state is a double talk state or a non-double talk state. The correlation matrix calculation unit 301a, the determinant calculation unit 301b, the determination threshold storage unit 301c, the state determination unit 301d, and the determinant gradient calculation are performed. Part 301e and determination threshold value normalization part 301f. The double talk detection unit 301 can be constructed by installing a double talk detection program in an information processing apparatus such as a personal computer or an embedded system, and can be functionally shown in FIG. Note that the double talk detecting unit 101 may be configured by hardware logic or the like.

  Since the correlation matrix calculation unit 301a has substantially the same configuration as the correlation matrix calculation unit 201a of the second embodiment, a description thereof will be omitted. The correlation matrix calculation unit 301a differs from the correlation matrix calculation unit 201a of the second embodiment in that the obtained correlation matrix is input not only to the determinant calculation unit 301b but also to the determination threshold value normalization unit 301f. Yes.

  The determinant calculating unit 301b and the determinant gradient calculating unit 301e have the same configuration as the determinant calculating unit 201b and the determinant gradient calculating unit 201e of the second embodiment, and thus description thereof is omitted.

  The determination threshold value storage unit 301c has substantially the same configuration as the determination threshold value storage unit 201c of the second embodiment. It differs from the determination threshold value storage unit 201c of the second embodiment in that it is connected to the determination threshold value normalization unit 301f to refer to the stored information.

  The determination threshold normalization unit 301f normalizes the threshold stored in the determination threshold storage unit 301c based on the correlation matrix input from the correlation matrix calculation unit 301a and normalizes the threshold to the state determination unit 301d Is input. The method described in the first embodiment can be applied as a method by which the determination threshold value normalization unit 301f normalizes.

  The determination threshold value storage unit 301c stores a threshold value to be normalized by the determination threshold value normalization unit 301f.

  The state determination unit 301d determines whether or not the state is a double talk state based on values input from the determinant calculation unit 301b and the determination threshold value normalization unit 301f. When the value of the determinant input from the determinant calculation unit 301b is larger than the normalized threshold value input from the determination threshold value normalization unit 301f, the double talk state is determined.

(C-2) Operation of the Third Embodiment Next, the detection operation of the double talk state in the echo canceller 300 of the third embodiment having the above configuration will be described.

  The correlation matrix calculation unit 301a calculates a correlation matrix (each element) like the above equation (1) based on the transmission signal and the pseudo echo signal, and sends the obtained correlation matrix to the determinant calculation unit 301b. input.

  When the correlation matrix is input, the determinant calculation unit 301b calculates the determinant of the input correlation matrix, and inputs the calculation result to the determinant gradient calculation unit 301e.

  When the value of the determinant of the correlation matrix is input, the determinant gradient calculation unit 301e calculates the value of the gradient based on the previously input value and the latest input value, and the result is the state. It inputs into the determination part 301d.

  When the correlation matrix element is input from the correlation matrix calculation unit 301a, the determination threshold normalization unit 301f normalizes the threshold stored in the determination threshold storage unit 301c based on the correlation matrix element, and normalizes the threshold. Is input to the state determination unit 301d.

  When the normalized threshold value is input from the determination threshold value normalization unit 301f and the gradient value is input from the determinant gradient calculation unit 301e, the state determination unit 301d compares these values to determine whether the state is a double talk state. The result is input to the adaptive filter control unit 302.

(C-3) Effect of Third Embodiment An echo canceller 300 according to the third embodiment includes a determination threshold value normalization unit 101e according to the first embodiment and a determinant gradient calculation unit 201e according to the second embodiment. A determinant gradient calculation unit 301e and a determination threshold value normalization unit 301f, which are the same means. As a result, the effects of the first embodiment and the second embodiment can be exhibited together. That is, it is possible to determine the double talk state regardless of the fluctuation state of the input signal or the like input to the echo canceller 300, and to shorten the detection time of the double talk state.

(D) Other Embodiments In addition to the above-described embodiments, the following embodiments may be further exemplified.

(D-1) In each of the above embodiments, the correlation matrix calculation units 101a, 201a, and 301a may input the correlation matrix to the determinant calculation units 101b, 201b, and 301b for each sample, or for each of a plurality of samples. May be entered.

In the case of performing input for each of a plurality of samples, for example, as an operation of the correlation matrix calculation units 101a, 201a, and 301a, a method of using an average value of a plurality of samples for each element of the correlation matrix of the above equation (1) can be given. . For example, in the case of determining every 10 samples, the result of the determinant calculation units 101b, 201b, and 301b is obtained as the value obtained by the following equation (4) for the element in the first row and the first column of the above equation (1). Enter. The following equation (4) is an equation for calculating an average value by taking the arithmetic average of 10 samples of the above equation (1) (“i” represents time). A similar calculation is performed on all elements of the correlation matrix to obtain an average value of the correlation matrix.

  In addition, as a method of using an average value of a plurality of samples, a weighted average value (weighted average) may be used. For example, there is a method of increasing the weight of new samples.

(D-2) In the first embodiment, the value of the determinant is used in the determination by the state determining unit 101d. However, the value is limited to the value of the determinant as long as it is the same amount obtained from the correlation matrix. For example, eigenvalues obtained from a correlation matrix may be used for state determination. For example, when the eigenvalue related to the correlation matrix of the above equation (1) is λ, the value of the eigenvalue λ of the correlation matrix R can be obtained by solving the equation according to the following equation (5). E represents a unit matrix.

  In the second and third embodiments, the gradient values calculated by the determinant gradient calculation units 201e and 301f are used for the determination by the state determination units 201d and 301d. The gradient is not limited to the gradient calculated based on the determinant such as 1), but may be an amount obtained from a similar correlation matrix. For example, it is possible to use the gradient of the eigenvalue obtained by the above equation (5).

(D-3) In each of the above embodiments, whether or not a double talk state is determined is determined based on the determinant of the correlation matrix shown in equation (1). If it has similar characteristics such as having a very large peak value, the determinant of the correlation matrix (or the gradient of the determinant) expressed by an expression other than the expression (1) should be used. May be.

  In each of the above embodiments, the double talk state is determined based on the value based on the correlation matrix as in the above equation (1). However, the correlation between the pseudo echo signal by the adaptive filter and the transmission signal is determined. As long as the value is based on the degree, the value may not be based on the correlation matrix.

(D-4) In each of the embodiments described above, the double talk detectors 101, 201, and 301 are provided only for the detection of the double talk state, but the same configuration is used for detecting the fluctuation of the echo path. Also good. Since the double-talk detection unit 601 included in the conventional echo canceller 600 described in Patent Document 1 can detect fluctuations in the echo path by means similar to the double-talk, each of the above-described implementations. Similarly, the double-talk detectors 101, 201, and 301 of the embodiment can detect the echo path variation. Therefore, the double-talk detection units 101, 201, and 301 of the present invention have the same effects as those of the above-described embodiments with respect to detection of echo path fluctuations.

(D-5) In each of the above embodiments, the change from the non-double talk state to the double talk state and the change from the double talk state to the non-double talk state are detected by the same method. The method for detecting the change from the non-double talk state to the double talk state may be different from the method for detecting the change from the double talk state to the non-double talk state. For example, as a method for detecting a change from the non-double talk state to the double talk state, the method described in the second embodiment using the gradient of the determinant for state detection is applied, and the double talk state is changed to the non-double talk state. As a method for detecting the change in the above, the method described in the first embodiment in which a determinant is used for state detection may be applied.

(D-6) In each of the embodiments described above, each unit performs a calculation based on the time domain, but may be replaced with a calculation performed in the frequency domain.

(D-7) The same configuration as that of the double talk detection units 101, 201, and 301 in each of the above embodiments is not only an echo canceller, but also other devices that erase unnecessary signals other than the main signal using an adaptive filter. It can also be applied to. For example, the present invention can be applied to a noise canceller (background noise elimination device).

  In a noise canceller, a reference signal such as white noise is passed through a pseudo noise generator using an adaptive filter that performs adaptive operation to generate pseudo noise, and the background is obtained by subtracting the pseudo noise from the main signal including background noise. Eliminate noise. Therefore, by treating the pseudo noise in the same manner as the pseudo echo signal and treating the main signal including the background noise in the same manner as the transmission signal, the same configuration as that of the double talk detection units 101, 201, and 301 in each of the above embodiments, It can be detected whether the adaptive operation of the adaptive filter can be continued or stopped, and at that time, the same effects as those of the above embodiments can be obtained.

It is a block diagram which shows the whole structure of 3rd Embodiment. It is a block diagram which shows the whole structure of 1st Embodiment. It is a block diagram which shows the whole structure of 2nd Embodiment. It is explanatory drawing shown about the structure which erase | eliminates a line echo by an echo canceller. It is explanatory drawing shown about the structure which erases an acoustic echo with an echo canceller. It is a block diagram which shows the functional structure of the conventional echo canceller. It is explanatory drawing of the symbol used for description of the conventional echo canceller.

Explanation of symbols

  300 ... Echo canceller, 301 ... Double talk detector, 301a ... Correlation matrix calculator, 301b ... Determinant calculator, 301c ... Determination threshold storage unit, 301d ... State determiner, 301e ... Determinant gradient calculator, 301f ... Determination Threshold normalization unit.

Claims (4)

An unnecessary signal erasure target signal including an unnecessary signal that is an unnecessary signal other than the main signal and the main signal, and a pseudo unnecessary signal formed by using an adaptive filter for erasing the unnecessary signal from the unnecessary signal erasure target signal Correlation calculating means for calculating a value representing the degree of correlation with
A change tendency calculating means for calculating a change tendency of the value calculated by the correlation calculating means;
State detecting means for determining whether the state related to the unnecessary signal erasure target signal is a non-double talk state or a double talk state based on the change tendency calculated by the change tendency calculating means and a threshold ;
The correlation calculating means calculates a correlation matrix between the unnecessary signal erasure target signal and the pseudo unnecessary signal, calculates a value of a determinant of the calculated correlation matrix as a value indicating a degree of correlation,
The signal state detecting device according to claim 1, wherein the state detecting unit determines that the state is a double talk state when a gradient value of a determinant value calculated by the correlation calculating unit exceeds a threshold value . 2. The signal state detection device according to claim 1, further comprising a threshold variable unit that changes a threshold applied by the state detection unit according to the level of one or both input signals to the correlation calculation unit. . 3. The signal state detection device according to claim 1 or 2 as a double talk detection device, wherein a pseudo echo signal is input to the double talk detection device as the pseudo unnecessary signal, and transmission is performed as the unnecessary signal erasure target signal. An echo canceller characterized by inputting a signal. Computer
An unnecessary signal erasure target signal including an unnecessary signal that is an unnecessary signal other than the main signal and the main signal, and a pseudo unnecessary signal formed by using an adaptive filter for erasing the unnecessary signal from the unnecessary signal erasure target signal Correlation calculating means for calculating a value representing the degree of correlation with
A change tendency calculating means for calculating a change tendency of the value calculated by the correlation calculating means;
A change trend that is calculated by the change trend calculating unit, based on the threshold value, to function in the state of the undesired signal erased signals and determines the state detecting means whether non double-talk state or a double talk state ,
The correlation calculating means calculates a correlation matrix between the unnecessary signal erasure target signal and the pseudo unnecessary signal, calculates a value of a determinant of the calculated correlation matrix as a value indicating a degree of correlation,
The signal detection program according to claim 1, wherein the state detection means determines a double talk state when the gradient value of the determinant value calculated by the correlation calculation means exceeds a threshold value .

Images