JP7790976B2 - Signal processing device, signal processing method, and signal processing program - Google Patents

Description

The present invention relates to signal processing technology that eliminates noise, interference signals, echoes, etc. that are mixed in with signals.

Speech signals input from microphones, handsets, etc. are often superimposed with background noise, posing a significant problem in speech coding and speech recognition. Patent documents 1 and 2 disclose a two-input noise canceller using two adaptive filters as a signal processing device designed to eliminate acoustically superimposed noise. A degraded signal (a signal containing a mixture of a desired signal and noise) and a reference signal (mainly including signals correlated with the noise) are input, and some or all of the noise is canceled, resulting in an enhanced signal (a signal in which the desired signal is enhanced). A step size calculation unit calculates the coefficient update step size of the second adaptive filter using the signal-to-noise ratio of the degraded signal estimated using the first adaptive filter. The first adaptive filter operates in the same way as the second adaptive filter, but the coefficient update step size of the first adaptive filter is set to a larger value than the coefficient update step size of the second adaptive filter. As a result, the output of the first adaptive filter is highly responsive to environmental changes, but its noise estimation accuracy is inferior to that of the second adaptive filter.

The step size calculation unit evaluates the signal-to-noise ratio of the noisy signal estimated using the first adaptive filter, and when the speech signal is louder than the noise, it assumes that interference from the speech signal is large and provides a small coefficient update step size to the second adaptive filter. Conversely, when the speech signal is quieter than the noise, it assumes that interference from the speech signal is small and provides a large coefficient update step size to the second adaptive filter. In this way, by controlling the second adaptive filter with the coefficient update step size provided by the step size calculation unit, sufficient adaptability to environmental changes and low distortion in the signal after noise cancellation are output.

Patent Document 3 discloses a configuration in which the first adaptive filter is omitted from the configurations of Patent Documents 1 and 2. The signal-to-noise ratio is approximated by the ratio between the desired signal (such as speech) estimated using a second adaptive filter and the output of the second adaptive filter, and the second adaptive filter itself is controlled using a step size calculated based on that signal-to-noise ratio. Furthermore, Patent Document 3 discloses a noise cancellation device configuration that expands on the configurations of Patent Documents 1 and 2, and also cancels speech signals mixed into noise when crosstalk due to speech signals is present, which is when the influence of speech signals mixed into noise at the input of a two-noise input device is significant. In addition to the configurations of Patent Documents 1 and 2, Patent Document 3 includes a third adaptive filter that cancels crosstalk from the reference signal. To accurately cancel noise from the speech signal input, a second step size calculation unit calculates a coefficient update step size and controls the third adaptive filter.

In other words, the noise cancellation devices in Patent Documents 1 to 3 control the updating of adaptive filter coefficients using a signal-to-noise ratio estimated using the post-noise cancellation signal and the adaptive filter output. By using a small step size when the signal-to-noise ratio is high and a large step size when the signal-to-noise ratio is low, they achieve both fast convergence and low-distortion output signals.

However, in the noise cancellers of Patent Documents 1 to 3, the adaptive filter coefficients are not updated at all. This is because the initial values of the adaptive filter coefficients are typically set to zero. An adaptive filter with zero coefficients outputs zero. Because this is the denominator of the estimated signal-to-noise ratio, the estimated signal-to-noise ratio becomes an extremely large value, and the corresponding step size is set to zero. A step size of zero means that no coefficient update is performed. To avoid this, the step size must be forcibly set to a non-zero value immediately after the coefficient update begins, but no clear design method is disclosed for what value the step size should actually be set to or for how long it should be set to a non-zero value. In other words, manual control of the step size is necessary to achieve both fast convergence and low-distortion output signals in a two-input noise canceller.

Patent Document 4 discloses the configuration of a two-input noise canceller that does not require manual control of the step size and achieves both fast convergence and low-distortion output signals. A new signal-to-noise ratio, defined as the ratio of the post-noise-cancellation signal to the reference signal, is used immediately after the device is operated to control the update of the adaptive filter's coefficients, solving the problem of coefficient updates not being performed. When the adaptive filter's coefficients grow, they are switched to the signal-to-noise ratio disclosed in Patent Document 3. The coefficient growth is evaluated when the new signal-to-noise ratio becomes sufficiently close to the conventional signal-to-noise ratio. When crosstalk is present, a third adaptive filter is introduced that cancels the speech signal from the noise input signal, and the step size is controlled using the same principle as the second adaptive filter.

Japanese Patent Application Publication No. 10-215193 Japanese Patent Application Laid-Open No. 2000-172299 International Publication No. 2012/046582 International Publication No. 2019/092798

However, in the signal processing device described in Patent Document 4, if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1, the estimated signal-to-noise ratio is not switched. This is because with an acoustic impulse response gain of less than 1, the new signal-to-noise ratio is not sufficiently close to the previous signal-to-noise ratio. As a result, the accuracy of the signal-to-noise ratio estimation is low, and it is not possible to achieve both fast convergence and a low-distortion output signal.

The object of the present invention is to provide technology that solves the above problems.

In order to achieve the above object, the device according to the present invention comprises:
a first input means for inputting a first mixed signal in which the first signal and the second signal are mixed;
a second input means for inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed;
a first adaptive filter for filtering the second mixed signal to generate an estimate of the second signal;
a first subtraction unit that generates an estimate of the first signal from the first mixed signal and an estimate of the second signal;
an estimation unit that estimates a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, and a coefficient of the first adaptive filter;
Equipped with
The first mixing ratio is used to control the first adaptive filter.

In order to achieve the above object, the method according to the present invention comprises:
A first mixed signal in which the first signal and the second signal are mixed is input;
a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed is input;
filtering the second mixed signal to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
estimating an amplitude or power ratio between the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, and a coefficient of the first adaptive filter;
The first mixing ratio is used to control generation of an estimate of the second signal.

In order to achieve the above object, the program according to the present invention comprises:
On the computer,
inputting a first mixed signal in which a first signal and a second signal are mixed;
inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed;
filtering the second mixture signal to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
a step of estimating a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, and a coefficient of the first adaptive filter;
using the first mixing ratio to control generation of an estimate of the second signal;
Execute the following.

The present invention makes it possible to obtain a signal processing device that can achieve both fast convergence and low-distortion output signals without manually controlling the step size, even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

1 is a block diagram showing a configuration of a signal processing device according to a first embodiment of the present invention. FIG. 10 is a block diagram showing the configuration of a signal processing device according to a second embodiment of the present invention. FIG. 10 is a block diagram showing a first configuration of an estimating unit according to a second embodiment of the present invention. FIG. 10 is a diagram showing time transitions of values according to the second embodiment of the present invention. FIG. 10 is a block diagram showing a second configuration of the estimation unit according to the second embodiment of the present invention. FIG. 10 is a block diagram showing a configuration of a signal processing device according to a third embodiment of the present invention. FIG. 11 is a block diagram showing a first configuration of an estimation unit according to a third embodiment of the present invention. FIG. 11 is a block diagram showing a second configuration of the estimation unit according to the third embodiment of the present invention. 1 is a block diagram showing the configuration of a computer according to a first embodiment of the present invention. 10 is a flowchart showing an example of signal processing by a processor of the computer shown in FIG. 9 .

Embodiments of the present invention will be described in detail below with reference to the drawings. However, the components described in the following embodiments are merely examples and are not intended to limit the technical scope of the present invention to these components alone.

1. First Embodiment
A signal processing device 100 according to a first embodiment of the present invention will be described with reference to Fig. 1. The signal processing device 100 in Fig. 1 is a device that calculates an estimate e1(k) of a first signal from a first mixture signal xP(k) in which a first signal and a second signal are mixed.

As shown in FIG. 1, the signal processing device 100 includes a first input unit 101, a second input unit 102, an adaptive filter 103, a subtraction unit 104, an estimation unit 106, and a coefficient update control unit 107.

Of these, the first input unit 101 inputs a first mixed signal xP(k) that is a mixture of the first signal and the second signal. The second input unit 102 inputs a second mixed signal xR(k) that is a mixture of the third signal and the fourth signal. The first signal and the third signal originate from the same signal source A and are correlated with each other. The second signal and the fourth signal originate from the same signal source B and are correlated with each other.

The subtraction unit 104 receives the first mixture signal xP(k) and an estimate n1(k) of the second signal mixed into the first mixture signal xP(k), and outputs an estimate e1(k) of the first signal. Then, to obtain the estimate n1(k) of the second signal, the adaptive filter 103 performs filtering on the second mixture signal xR(k) using coefficients 141 that are updated based on the estimate e1(k) of the first signal.

The estimation unit 106 estimates the amplitude or power ratio of the first signal to the second signal as a first mixture ratio R1(k) using the estimated value e1(k) of the first signal, the estimated value n1(k) of the second signal, the first mixture signal xP(k), the second mixture signal xR(k), and the coefficient vector w1(k) of the adaptive filter 103. When the value of the first mixture ratio R1(k) obtained by the estimation unit 106 is large, the coefficient update control unit 107 outputs a control signal μ1(k) to the adaptive filter 103 to reduce the amount of update of the coefficients 141 of the adaptive filter 103.

With this configuration, this embodiment can remove the second signal from a mixed signal containing a first signal and a second signal with a low amount of calculation and without delay.As a result, even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1, it is possible to obtain an estimate of the first signal with little residual second signal and little distortion.

2. Second Embodiment
As a signal processing device according to a second embodiment of the present invention, a noise canceller will be described which receives a noisy signal (a signal obtained by mixing a desired signal and noise) and a reference signal (mainly including a signal correlated with the noise), cancels part or all of the noise, and outputs an emphasized signal (a signal obtained by emphasizing the desired signal). Here, the noisy signal corresponds to a first mixed signal obtained by mixing a first signal and a second signal, the reference signal corresponds to a second mixed signal obtained by mixing a third signal and a fourth signal, and the emphasized signal (an estimated value of the first signal) corresponds to the desired signal.

[2.1. Explanation of basic noise cancellation technology]
The following is a brief explanation of the basic technology of noise cancellation, which uses an adaptive filter to cancel noise, interference signals, echoes, etc. mixed in with a desired signal input from a microphone, handset, communication channel, etc., or to emphasize the desired signal.

As disclosed in Patent Documents 1 to 3, a two-input noise cancellation device uses an adaptive filter that approximates the impulse response of the acoustic path from the noise source to the audio input terminal to generate, from a reference signal, pseudo-noise (an estimated value of the second signal) corresponding to the noise components mixed into the audio at the audio input terminal. The device then operates to suppress the noise components by subtracting this pseudo-noise from the signal input to the audio input terminal (first mixed signal). Here, the mixed signal refers to a signal that contains a mixture of a desired (audio) signal and noise, and is generally supplied to the audio input terminal from a microphone or handset. The reference signal is a signal that is correlated with the noise components at the noise source and is captured near the noise source. By capturing the reference signal near the noise source in this way, the reference signal can be considered to be approximately equal to the noise components at the noise source. The adaptive filter receives the reference signal supplied to the reference input terminal.

The adaptive filter's coefficients are corrected by correlating the error, obtained by subtracting pseudo-noise from the degraded signal, with the reference signal input to the reference input terminal. Patent documents 1 to 3 disclose algorithms for correcting the adaptive filter's coefficients, such as the "Least Mean-Square Algorithm (LMS)" and "Learning Identification Method (LIM)." LIM is also known as the normalized LMS (NLMS) algorithm.

Coefficient update using the normalized LMS algorithm is expressed by equation (1), where μ(k) is the step size at time k. The coefficient vector w(k) is defined by equation (2) using its elements. The reference signal vector x(k) is expressed by equation (3) using its elements. Note that T represents the transpose of the vector, and L is the total number of coefficients.

The LMS algorithm and LIM are types of algorithms known as gradient methods, and the speed and accuracy of coefficient updates depend on a constant known as the coefficient update step size. Filter coefficients are updated by multiplying the coefficient update step size by the error. To reduce interference with the coefficient update caused by the desired signal (the estimated value of the first signal) contained in the error, the coefficient update step size must be set to an extremely small value or zero. Constantly setting the coefficient update step size to a small value reduces the adaptive filter coefficient's ability to track environmental changes. Patent documents 1 to 3 above disclose a method for solving the problem of increased output error or distortion of the desired signal. Furthermore, because the desired signal is generally speech, the term "speech" will be used hereafter. However, the scope of this invention is not limited to speech and can refer to any type of signal, including acoustic (audio) signals. Furthermore, the coefficient update algorithm is not limited to LMS or LIM.

2.2. Configuration of the Noise Canceller
2 is a block diagram showing the overall configuration of a noise cancellation device 200 according to this embodiment. The noise cancellation device 200 may function as part of a device such as a digital camera, a laptop computer, a mobile phone, a hearing aid, a television, a smart speaker, or a robot, but the present invention is not limited to these and may be applied to any signal processing device that is required to cancel noise from an input signal.

2, noise cancellation device 200 receives a noisy signal (first mixture signal) xP(k) containing a mixture of speech (first signal) and noise (second signal) from input terminal 201. It also receives a reference signal (second mixture signal) xR(k) containing a mixture of speech and noise from input terminal 202, and outputs an estimated speech value e1(k) (emphasis signal) from output terminal 205. Furthermore, noise cancellation device 200 includes adaptive filter 203, subtraction unit 204, and estimation unit 206. Adaptive filter 203 includes adaptive filter 103 and coefficient update control unit 107 in FIG. 1, and receives first mixture ratio R1(k) to calculate step size μ1(k) and updates coefficient vector w1(k) using the calculated step size μ1(k). The noise canceller 200 generates pseudo-noise n1(k) by modifying a reference signal xR(k) that is correlated with the noise to be cancelled using an adaptive filter 203, and then subtracts this from a noisy signal xP(k), which is an audio signal with noise superimposed on it, thereby canceling the noise.

The degraded signal xP(k) is supplied to the input terminal 201 as a sequence of sample values. The degraded signal xP(k) is transmitted to the subtraction unit 204. The reference signal xR(k) is supplied to the input terminal 202 as a sequence of sample values. The reference signal xR(k) is transmitted to the adaptive filter 203 and the estimation unit 206.

The adaptive filter 203 performs a convolution operation on the reference signal xR(k) and the filter coefficients, and transmits the result as pseudo-noise n1(k) to the subtraction unit 204 and the estimation unit 206. The adaptive filter 203 also supplies the coefficient vector w1(k) to the estimation unit 206.

The subtraction unit 204 receives the noisy signal xP(k) from the input terminal 201 and the pseudo-noise n1(k) from the adaptive filter 203. The subtraction unit 204 subtracts the pseudo-noise n1(k) from the noisy signal xP(k) and transmits the result as an estimated value e1(k) of the audio signal (estimated value of the first signal) to the output terminal 205, while also feeding it back to the adaptive filter 203.

The estimation unit 206 receives the estimated speech value e1(k), the pseudo-noise n1(k) (output of the adaptive filter 203), the reference signal xR(k), the noisy signal xP(k), and the coefficient vector w1(k) of the adaptive filter 203, estimates the amplitude or power ratio of speech to noise at the input terminal 201 as the first mixture ratio R1(k), and transmits this to the adaptive filter 203. The adaptive filter 203 updates the coefficient vector w1(k) using a small step size μ1(k) when the first mixture ratio R1(k) is large and a large step size μ1(k) when the first mixture ratio R1(k) is small. Methods for controlling the step size using the estimated first mixture ratio R1(k), i.e., the signal-to-noise ratio, are disclosed in detail in Patent Documents 1 to 3. Furthermore, as disclosed in Patent Documents 1 to 3, the first mixture ratio R1(k) may be averaged before being used to calculate the step size μ1(k). Improved estimation accuracy for the ratio of speech to noise amplitude or power.

2.3. First Configuration of Estimation Unit 206
FIG. 3 is a block diagram showing a first internal configuration of estimation unit 206. Estimation unit 206 includes signal ratio estimation unit 301, signal ratio estimation unit 302, mixer 305, and correction unit 310. Signal ratio estimation unit 301 receives speech estimate e1(k) and pseudo-noise n1(k) and estimates the amplitude or power ratio of speech to noise at input terminal 201 as second mixture ratio R2(k). Second mixture ratio R2(k) may be the ratio of the amplitude or power of speech estimate e1(k) and pseudo-noise n1(k), or may be calculated by adding a small constant to the amplitude or power of the speech estimate e1(k) and pseudo-noise n1(k). Adding a small constant has the effect of preventing divergence of the quotient due to division. Alternatively, one or both of speech estimate e1(k) and pseudo-noise n1(k) may be averaged before use. Averaging can improve the accuracy of the ratio calculation.

The correction unit 310 receives the noisy signal xP(k) and the reference signal xR(k), corrects the reference signal xR(k), and obtains the corrected reference signal xRC(k). The power of the noisy signal xP(k) and the reference signal xR(k) is averaged for M1 samples immediately after the noise canceller 200 starts operating, and the ratio is obtained as the multiplier g1. The natural number M1 is a predetermined constant. L sampling periods are required until non-zero input signal samples are supplied to all coefficients of the adaptive filter 203 and normal operation begins. Therefore, one way to set M1 is to set M1 = L. The correction unit 310 obtains the corrected reference signal xRC(k) by multiplying the square root of the multiplier g1 by the reference signal xR(k).

In typical audio applications, the amplitude of the first signal can be assumed to be zero immediately after the device is turned on. This is because, for example, conversation does not begin during the short period of time immediately after startup, such as 100 ms or 200 ms, and no voice (first signal) is present. However, background noise is present immediately after the device is turned on. In this case, the degradation signal xP(k) is equal to the noise (second signal), so the scaling factor g1 can be estimated as the ratio of the power of the degradation signal (= noise) xP(k) to the reference signal (mixed noise at input terminal 201) xR(k). The scaling factor g1 estimated in this way corresponds to the ratio of input to output when white noise is input to adaptive filter 203, i.e., the average gain of the impulse response of the acoustic system approximated by adaptive filter 203. The product of the power of reference signal xR(k) and scaling factor g1 is an approximation of the power of pseudo-noise n1(k) (estimated value of the second signal) that reflects only the gain of adaptive filter 203, and the accuracy of the power of pseudo-noise n1(k) is higher than that of reference signal xR(k). Therefore, by using the product of the power of reference signal xR(k) and scaling factor g1 instead of the power of pseudo-noise n1(k), the ratio of the amplitude or power of the first signal to the second signal, i.e., the speech to noise at input terminal 201, can be determined with higher accuracy.

The determination of the multiplier g1 may be performed at any timing and any number of times as long as the amplitude of the audio (first signal) is zero. By determining the value of the multiplier g1 more frequently, the adaptive filter 203 can more accurately track changes in the impulse response of the acoustic system it approximates. The audio estimate e1(k) can be used to determine that the audio amplitude is zero other than immediately after startup. Since the audio estimate e1(k) always contains a certain amount of error even when the adaptive filter 203 has sufficiently converged, this error is taken into account when comparing the audio estimate e1(k) with a predetermined threshold β1. Setting the threshold β1 to a large value increases the frequency at which the amplitude is determined to be zero, but also increases the number of determination errors. Setting the threshold β1 to a small value decreases the frequency at which the amplitude is determined to be zero, resulting in poor tracking of changes in the acoustic system impulse response.

The signal ratio estimation unit 302 receives the speech estimate e1(k) and the corrected reference signal xRC(k), and estimates the ratio of the amplitude or power of speech to noise at the input terminal 201 as the third mixture ratio R3(k). The third mixture ratio R3(k) may be the ratio of the amplitude or power of the speech estimate e1(k) and the corrected reference signal xRC(k), or may be calculated by adding a small constant to these amplitudes or powers. Adding a small constant has the effect of preventing the quotient from diverging due to division. Alternatively, either or both of the speech estimate e1(k) and the corrected reference signal xRC(k) may be averaged before use.

The mixer 305 mixes the second mixture ratio R2(k) and the third mixture ratio R3(k) and outputs the mixture result as the first mixture ratio R1(k). The second mixture ratio R2(k) and the third mixture ratio R3(k) may be mixed using weighted addition, or may be mixed using a more complex higher-order polynomial. Prior to mixing, one or both of the second mixture ratio R2(k) and the third mixture ratio R3(k) may be averaged. Averaging can improve the calculation accuracy of the first mixture ratio R1(k), i.e., the approximation accuracy of the amplitude or power of the speech and noise at the input terminal 201.

For simplicity's sake, consider the case where the first mixture ratio R1(k) is calculated by mixing the second mixture ratio R2(k) and the third mixture ratio R3(k) using weighted addition. The sum of the weights for both is set to 1. The coefficients of the adaptive filter 203 are generally initialized to zero. Therefore, when coefficient updating begins, the pseudo-noise n1(k) is zero, and the second mixture ratio R2(k) has a denominator of zero and is infinite. Therefore, when the step size μ1(k) of the adaptive filter 203 is calculated using the second mixture ratio R2(k), the result is an extremely small value or zero, and the coefficient does not grow. If the coefficient does not grow, the pseudo-noise n1(k) does not increase, and the same problem persists.

On the other hand, the denominator of the third mixture ratio R3(k) is the corrected reference signal xR(k), which is not necessarily zero at the start of coefficient updating. This is because the microphone input contains minute signals such as environmental noise. Even if the corrected reference signal xR(k) is zero, it will not remain at zero. Therefore, the third mixture ratio R3(k) will never become infinite, and the corresponding step size μ1(k) will not become a minimum value. Therefore, the coefficients of the adaptive filter 203 grow as the coefficients are updated, converging to a value representing the acoustic characteristics of the path from the noise signal source to the input terminal 201. When the corrected reference signal xR(k) is zero, the reference signal xR(k) is also zero, and the coefficients of the adaptive filter 203 are not updated. Therefore, even if the third mixture ratio R3(k) becomes extremely large, it does not pose a problem. However, when the coefficients of the adaptive filter 203 grow to a certain extent and the pseudo-noise n1(k) grows, the third mixture ratio R3(k) has lower approximation accuracy for the ratio of the amplitude or power of speech to noise at the input terminal 201 than the second mixture ratio R2(k).

Therefore, the mixer 305 sets the weight of the third mixture ratio R3(k) to a large value when the coefficient update of the adaptive filter 203 begins, and decreases it as the coefficients grow. The weight of the second mixture ratio R2(k) is set to a small value when the coefficient update of the adaptive filter 203 begins, and increases it over time. This means that the proportion of the third mixture ratio R3(k) in the first mixture ratio R1(k) decreases in accordance with the number of coefficient updates.

For example, if the weight of the third mixture ratio R3(k) is set to 1 when the adaptive filter 203 begins updating its coefficients, the weight of the second mixture ratio R2(k) will be 0, since the sum of the weights must be 1. Coefficient growth corresponds to the number of coefficient updates. Therefore, the weight of the third mixture ratio R3(k) is set to 1 when the adaptive filter 203 begins updating its coefficients, and the weight decreases toward 0 in accordance with the number of coefficient updates for the coefficient vector w1(k). Correspondingly, the weight of the second mixture ratio R2(k) increases from 0 to 1. If the initial value of the weight of the third mixture ratio R3(k) is set to 1, and the weight reaches 1 at some point or is set back to 1, the third mixture ratio R3(k) will be switched to the second mixture ratio R2(k). Similar effects can be achieved by setting the sum of the weights to a value other than 1 or by setting the initial weight for the third mixture ratio R3(k) to a value other than 1. The weights can be determined using the amount of change in the coefficient vector w1(k) of the adaptive filter 203.

Figure 4 is a schematic diagram showing the time progression of a signal according to the second embodiment, illustrating the corrected reference signal xRC(k), the pseudo-noise n1(k) output from the adaptive filter 203, and the coefficient vector w1(k). The vertical axis in Figure 4 represents the signal value in terms of power, and the coefficient is the norm of the coefficient vector. The horizontal axis represents the number of coefficient updates for the adaptive filter 203, expressed as the number of samples k. Comparing xRC(k) and n1(k) is equivalent to comparing the third mixture ratio R3(k) and the second mixture ratio R2(k), which have the same numerator and denominators xRC(k) and n1(k), respectively. Because xRC(k) is unrelated to the coefficient updates, n1(k) determines the relationship between the third mixture ratio R3(k) and the second mixture ratio R2(k). n1(k) is determined by xRC(k) and the coefficient vector w1(k), and its amplitude or power increases with coefficient updates, excluding changes due to increases or decreases in xRC(k). In other words, xRC(k) is constant, and n1(k) increases as the coefficient vector w1(k) increases due to coefficient updates.

The relationship between the corrected reference signal xRC(k), the pseudo-noise n1(k) output by the adaptive filter 203, and the coefficient vector w1(k) described above is clear from Figure 4. In Figure 4, xRC(k), which is unrelated to coefficient updates, is represented as a constant for simplicity. n1(k), which depends on xRC(k), increases smoothly, and when the coefficient vector w1(k) converges, n1(k) also saturates. Because the amount of change in the coefficient vector w1(k) of the adaptive filter 203 decreases with coefficient updates, the amount of change in the coefficient vector w1(k) can be used as an indicator of how much n1(k) has grown from zero. In other words, the mixer 305 determines the weight of the third mixing ratio R3(k) based on the change in the coefficient vector w1(k) over time.

By correcting xR(k) by the multiplier g1, the difference between the power of n1(k) and the power of xR(k) in Figure 4 can be reduced. This means that the difference between the power of n1(k) and the power of xRC(k) is smaller than the difference between the power of n1(k) and the power of xR(k). In other words, xRC(k) can approximate n1(k) with higher accuracy than xR(k). This leads to more accurate step size control through the high-accuracy third mixing ratio R3(k), which is desirable for controlling the adaptive filter 203. Therefore, it is possible to obtain an estimated speech value with less residual noise and less distortion.

The change over time of the coefficient vector w1(k) may be the change over time of the sum of squares or the sum of absolute values of the elements of the coefficient vector w1(k), or the change over time of the partial sum of squares or the partial sum of absolute values. When using a partial sum, elements of the coefficient vector may be thinned out, or a portion of the coefficient vector may be extracted. By using a partial sum, the amount of calculation required for the change over time of the coefficients can be reduced while minimizing the deterioration in the accuracy of the evaluation of the change over time. Furthermore, since the coefficients change independently of the input signal, they are smooth. Therefore, they have minimal fluctuation compared to other indices that depend on the input signal, allowing for accurate detection of saturation.

Equation (4) shows an example of mixing the second mixing ratio R2(k) and the third mixing ratio R3(k) by weighted addition. ζ1(k) expressed in equation (5) satisfies ζ1(0) = 0 and increases as the coefficients are updated. δ1(k), which corresponds to the time change in the coefficient vector w1(k) of the adaptive filter 203, can be calculated using equation (6). M is a predetermined natural number that determines the frequency at which the time change in the coefficient vector w1(k) is calculated. The larger M is, the smaller the fluctuation in δ1(k) and the more accurate the detection of coefficient saturation becomes; however, the larger the delay in the change in δ1(k), the longer the detection of coefficient saturation becomes. Equation (5) represents an example in which, when the value of the time change δ1(k) becomes less than the threshold value ε1, the content ratio of the third mixture ratio R3(k) in the first mixture ratio R1(k) is set to 0, and the first mixture ratio R1(k) is switched from the third mixture ratio R3(k) to the second mixture ratio R2(k). This is an example of a method for determining whether the value of the time change δ1(k) has become sufficiently small. If ζ1(k) = δ1(k)/|δ1(M)| is used instead of equation (5), the content ratio of the third mixture ratio R3(k) is determined by the time change of the coefficient vector w1(k).

In this way, the mixer 305 can set the content ratio of the third mixture ratio R3(k) in the first mixture ratio R1(k) to either 100% or 0% based on the comparison result between the time change δ1(k) and the threshold ε1. Alternatively, the mixer 305 can repeatedly compare the time change δ1(k) of the coefficient vector w1(k) with the threshold ε1, and if the time change δ1(k) of the coefficient vector w1(k) is equal to or greater than the threshold ε1, set the content ratio of the third mixture ratio R3(k) in the first mixture ratio R1(k) to 100%. Otherwise, set the content ratio of the third mixture ratio R3(k) in the first mixture ratio R1(k) to 0%. This allows for a first mixture ratio R1(k) that is highly responsive to environmental changes, enabling more accurate coefficient updating, even if, for example, the amount of updates to the coefficient vector w1(k) of the adaptive filter 203 suddenly increases.

The mixing unit 305 can set the content ratio of the third mixing ratio R3(k) in the first mixing ratio R1(k) to 0% if the time change δ1(k) of the coefficient vector w1(k) is less than the threshold ε1 for a predetermined first constant L1 or more consecutive times. Otherwise, the mixing unit 305 can set the content ratio of the third mixing ratio R3(k) in the first mixing ratio R1(k) to 100%. After the content ratio of the third mixing ratio R3(k) is set to 0% once, the comparison of δ1(k) and ε1 can be stopped by setting the threshold to a sufficiently large value. Here, L1 is an integer greater than or equal to 2.

The mixing unit 305 may set the content ratio of the third mixture ratio R3(k) in the first mixture ratio R1(k) to 0% when δ1(k) is less than the threshold ε1 for a predetermined third constant L3 or more consecutive times in a section where the time change δ1(k) of the coefficient vector w1(k) has a number of samples equal to the predetermined second constant L2. Otherwise, the content ratio of the third mixture ratio R3(k) in the first mixture ratio R1(k) may be set to 100%. The second constant L2 and the third constant L3 are both integers greater than or equal to 2, satisfying L2 > L3. Furthermore, the continuous condition may be removed from the above evaluation, and the number of times δ1(k) is less than the threshold ε1 for a predetermined third constant L3 or more may be evaluated. After the content ratio of the third mixture ratio R3(k) is set to 0% once, the evaluation of δ1(k) may be stopped by setting L3 to a value greater than L2.

The mixing unit 305 may also set the minimum weighting values of the second mixing ratio R2(k) and the third mixing ratio R3(k) to a value greater than 0 rather than 0.

2.4. Second Configuration of Estimation Unit 206
FIG. 5 is a block diagram showing a second internal configuration of the estimation unit 206. The estimation unit 206 includes a correction unit 310, a mixer 506, and a signal ratio estimation unit 503. The correction unit 310 receives the noisy signal xP(k) and the reference signal xR(k) and corrects the reference signal xR(k) to obtain a corrected reference signal xRC(k). The operation of the correction unit 310 is similar to that of the correction unit 310 already described. The mixer 506 mixes the corrected reference signal xRC(k) (corrected second mixed signal) with the pseudo-noise n1(k) (estimated value of the second signal) based on the coefficient vector w1(k) to generate a first mixed signal n3(k). The signal ratio estimation unit 503 receives the speech estimate e1(k) and the first mixed signal n3(k) and estimates the amplitude or power ratio of speech to noise at the input terminal 201 as a first mixed ratio R1(k). The first mixture ratio R1(k) may be the ratio of the amplitude or power of the speech estimate e1(k) to the first mixed signal n3(k), or may be calculated by adding a small constant to the amplitude or power. Adding a small constant has the effect of preventing the quotient from diverging due to division. Alternatively, either or both of the estimate e1(k) and the first mixed signal n3(k) may be averaged before use. Averaging can improve the accuracy of the ratio calculation.

The second internal configuration of the estimation unit 206 shown in FIG. 5 is equivalent to the first internal configuration of the estimation unit 206 shown in FIG. 3. That is, in the first internal configuration shown in FIG. 3, two estimates of the amplitude or power ratio of speech to noise at the input terminal 201 are generated by the signal ratio estimation units 301 and 302, and these estimates are mixed to calculate the first mixture ratio R1(k). The second internal configuration of the estimation unit 206 shown in FIG. 5 mixes two types of noise estimates, i.e., the corrected reference signal xRC(k) and the pseudo-noise n1(k), to generate the first mixed signal n3(k), determine the denominator, and then calculate the first mixture ratio R1(k) by combining this with the numerator, the speech estimate e1(k). These two types of configurations are possible because the first internal configuration shown in FIG. 3 and the second internal configuration shown in FIG. 5 use the same numerator, i.e., the speech estimate e1(k), when estimating the amplitude or power ratio of speech to noise at the input terminal 201. The second internal configuration of the estimation unit 206 shown in Figure 5 has fewer components and is simpler than the first internal configuration shown in Figure 3.

The operation of the mixer 506 is similar to that of the mixer 305, except that the second and third mixing ratios, which are input signals, are replaced with pseudo-noise and a reference signal (second mixing signal), respectively. Therefore, the description of the operation of the mixer 305 can also be applied to the mixer 506.

With the above configuration, this embodiment can smoothly update the coefficients without forcibly setting a special value for the step size, even if the gain of the acoustic impulse response approximated by the adaptive filter 203 is less than 1. As a result, it is possible to obtain an output signal with little residual noise and little signal distortion.

3. Third embodiment
In the explanation so far, it has been assumed that the reference signal is noise itself, by capturing the reference signal near the noise source. However, in reality, there are cases where this condition cannot be met. In such cases, the reference signal is composed of noise and an audio signal mixed in with it. Such an audio signal mixed into the reference signal is called crosstalk. The configuration of a noise cancellation device when crosstalk exists is disclosed in Patent Document 3.

In this embodiment, a second adaptive filter is introduced to cancel crosstalk, similar to the noise cancellation. The second adaptive filter, which approximates the impulse response of the acoustic path (crosstalk path) from the audio signal source to the reference input terminal, is used to generate pseudo-crosstalk corresponding to the audio signal components mixed in at the reference input terminal. Then, by subtracting this pseudo-crosstalk from the signal (reference signal) input to the reference input terminal, the audio signal components (crosstalk) mixed in the reference input are canceled.

A noise cancellation device according to a third embodiment of the present invention will be described with reference to Figure 6. Compared to the second embodiment, the noise cancellation device according to this embodiment includes a subtraction unit 804 and an adaptive filter 803 in addition to the subtraction unit 204 and adaptive filter 203, and the estimation unit 206 has been replaced with an estimation unit 806. Other configurations and operations are similar to those of the second embodiment, and therefore the same components are designated by the same reference numerals and detailed descriptions will be omitted.

Noise cancellation device 800 uses an adaptive filter to modify a signal (output at output terminal 205 = estimated speech signal or emphasis signal) that is correlated with the crosstalk (third signal) to be canceled, to generate pseudo crosstalk n2(k) (estimated value of the third signal). This is then subtracted from a reference signal xR(k) that contains a mixture of speech and noise, thereby canceling the crosstalk. When updating the coefficients of adaptive filter 803, the step size is controlled using a fourth mixture ratio R4(k), which approximates the ratio of the amplitude or power of the fourth signal to the third signal. This allows for smooth coefficient updating, resulting in an output signal with minimal residual noise and minimal signal distortion.

The degradation signal xP(k) is supplied to the input terminal 201 as a sample value sequence and transmitted to the subtraction unit 204. The reference signal xR(k) is supplied to the input terminal 202 as a sample value sequence and transmitted to the subtraction unit 804.

The subtraction unit 804 receives the reference signal xR(k) from the input terminal 202 and the pseudo crosstalk n2(k) from the adaptive filter 803. The subtraction unit 804 subtracts the pseudo crosstalk n2(k) from the reference signal xR(k) and transmits the result as a noise estimate e2(k) (estimate of the fourth signal) to the output terminal 805, while simultaneously feeding it back to the adaptive filter 803. The subtraction unit 804 also supplies the noise estimate e2(k) to the estimation unit 806.

The adaptive filter 803 performs a convolution operation on the speech estimate e1(k) (emphasis signal) and the filter coefficients, and transmits the result as pseudo crosstalk n2(k) (estimate of the third signal) to the subtraction unit 804 and the estimation unit 806. The adaptive filter 803 also supplies the coefficient vector w2(k) to the estimation unit 806.

The estimation unit 806 receives the speech estimate e1(k), the pseudo-noise n1(k) (output of the adaptive filter 203), the noise estimate e2(k) (the estimate of the fourth signal), the coefficient vector w1(k) of the adaptive filter 203, and the noisy signal xP(k), estimates the amplitude or power ratio of speech to noise at the input terminal 201 as the first mixture ratio R1(k), and transmits this to the adaptive filter 203. The adaptive filter 203 updates the coefficients using a small step size μ1(k) when the first mixture ratio R1(k) is large and a large step size μ1(k) when the first mixture ratio R1(k) is small. Note that the adaptive filter 203 shown in FIG. 6 differs from the adaptive filter 203 shown in FIG. 2 in that the noise estimate e2(k) is input instead of the reference signal xR(k). Methods of controlling the step size using the first mixture ratio R1(k), i.e., an estimated value of the signal-to-noise ratio, are disclosed in detail in Patent Documents 1 to 3. Also, as disclosed in Patent Documents 1 to 3, the first mixture ratio R1(k) may be averaged and then used to calculate the step size μ1(k). This improves the estimation accuracy of the ratio of the amplitude or power of speech to noise at input terminal 201.

The estimation unit 806 further receives the pseudo crosstalk n2(k) (output of the adaptive filter 803), the coefficient vector w2(k) of the adaptive filter 803, and the reference signal xR(k), estimates the ratio of the amplitude or power of the noise and the crosstalk (the fourth signal and the third signal) at the input terminal 202 as a fourth mixing ratio R4(k), and transmits this to the adaptive filter 803. The size of the coefficient vector w2(k) shown in equation (7) may be equal to or different from the size L of the coefficient vector w1(k).

The adaptive filter 803 updates the coefficients using a small step size μ2(k) when the fourth mixing ratio R4(k) is large and a large step size μ2(k) when the fourth mixing ratio R4(k) is small. Methods for controlling the step size μ2(k) using the fourth mixing ratio R4(k), i.e., an estimate of the noise-to-crosstalk ratio at the input terminal 202, are disclosed in detail in Patent Documents 1 to 3. Furthermore, as disclosed in detail in Patent Documents 1 to 3, the fourth mixing ratio R4(k) may be averaged before being used to calculate the step size μ2(k). This improves the accuracy of estimating the ratio of the amplitude or power of noise to crosstalk at the input terminal 202.

3.1. First Configuration of Estimation Unit 806
FIG. 7 is a block diagram showing a first internal configuration of the estimation unit 806. In addition to the configuration of the estimation unit 206, the estimation unit 806 includes a signal ratio estimation unit 901, a signal ratio estimation unit 902, a mixer 905, and a correction unit 710. The correction unit 310 shown in FIG. 7 differs from the correction unit 310 shown in FIG. 3 in the following respects. First, instead of the reference signal xR(k), a noise estimate (fourth signal estimate) e2(k) is input. Second, instead of the corrected reference signal xRC(k), a corrected noise estimate e2C(k) is output. The remaining operations are the same. The signal ratio estimation unit 302 shown in FIG. 7 is the same as the signal ratio estimation unit 302 shown in FIG. 3 except that the corrected noise estimate e2C(k) is used instead of the corrected reference signal xRC(k).

The signal ratio estimator 901 receives the noise estimate e2(k) and the pseudo crosstalk n2(k) and estimates the ratio of the amplitude or power of the noise to the crosstalk as the fifth mixture ratio R5(k). The fifth mixture ratio R5(k) may be the ratio of the amplitude or power of the noise estimate e2(k) to the pseudo crosstalk n2(k), or may be calculated by adding a small constant to these amplitudes or powers. Adding a small constant has the effect of preventing the quotient from diverging due to division. Alternatively, either or both of the noise estimate e2(k) and the pseudo crosstalk n2(k) may be averaged before use. Averaging can improve the accuracy of the ratio calculation.

The correction unit 710 receives the reference signal xR(k) and the speech estimate e1(k), corrects the speech estimate e1(k), and obtains the corrected speech (first signal) estimate e1C(k). The operation of the correction unit 710 is identical to that of the correction unit 310, replacing the noise estimate e2(k) with the speech estimate e1(k) and the noisy signal xP(k) with the reference signal xR(k). However, there is a difference in how the scaling factor g2 used for correction is calculated.

Similar to the correction unit 310, the correction unit 710 averages the power of the reference signal xR(k) and the speech estimate e1(k) for M2 samples immediately after the noise cancellation device 800 starts operating, and calculates the ratio as the multiplier g2. The natural number M2 is a predetermined constant, and similar to the correction unit 310, M2 can be set to a number of samples equal to the number of taps of the adaptive filter 803. When the noise is continuous and does not have silent intervals, an appropriate multiplier g2 cannot be calculated using this method, so g2 = g1 can be used. This is because the multipliers g1 and g2 correspond to the same acoustic space, and g1 and g2, which are dominated by the reflectivity of the acoustic space, can be considered approximately equal in determining their values. By using the product of the multiplier g2 calculated in this way and the power of the speech estimate e1(k) instead of the power of the pseudo crosstalk n2(k), the ratio of the amplitude or power of the fourth signal to the third signal, i.e., the noise and crosstalk at the input terminal 202, can be calculated with higher accuracy.

The multiplier g2 can be determined at any timing and any number of times as long as the amplitude of the noise (fourth signal) is 0. By determining the value of the multiplier g2 more frequently, adaptive filter 803 can more accurately track changes in the impulse response of the acoustic system it approximates. The noise estimate e2(k) can be used to determine that the noise amplitude is 0 other than immediately after startup. Since the noise estimate e2(k) always contains a certain amount of error even when adaptive filter 803 has sufficiently converged, this error is taken into account when comparing the noise estimate e2(k) with a predetermined threshold β2. Setting threshold β2 to a large value increases the frequency at which the amplitude is determined to be 0, but also increases the likelihood of incorrect determination. Setting threshold β2 to a small value decreases the frequency at which the amplitude is determined to be 0, resulting in poor tracking of changes in the acoustic system impulse response.

The signal ratio estimation unit 902 receives the noise estimate e2(k) and the corrected speech estimate e1C(k) (the corrected estimate of the first signal) and estimates the ratio of the amplitude or power of the noise to the crosstalk at the input terminal 202 as the sixth mixture ratio R6(k). The sixth mixture ratio R6(k) may be the ratio of the amplitude or power of the noise estimate e2(k) and the corrected speech estimate e1C(k), or may be calculated by adding a small constant to these amplitudes or powers. Adding a small constant has the effect of preventing the quotient from diverging due to division. Alternatively, either or both of the noise estimate e2(k) and the corrected speech estimate e1C(k) may be averaged before use.

The mixer 905 mixes the fifth mixture ratio R5(k) and the sixth mixture ratio R6(k) using the coefficient vector w2(k) of the adaptive filter 803 and outputs the mixed result as the fourth mixture ratio R4(k). The fifth mixture ratio R5(k) and the sixth mixture ratio R6(k) may be mixed by weighted addition using the coefficient vector w2(k) of the adaptive filter 803, or may be mixed using a more complex higher-order polynomial. Prior to mixing, one or both of the fifth mixture ratio R5(k) and the sixth mixture ratio R6(k) may be averaged. Averaging can improve the calculation accuracy of the fourth mixture ratio R4(k), i.e., the approximation accuracy of the amplitude or power of noise and crosstalk.

The operation of the mixer 905 is the same as that of the mixer 305 by changing the input and output signals of the mixer 305 to those of the mixer 905 as shown in Figure 7, so a detailed explanation will be omitted.

3.3. Second Configuration of Estimation Unit 806
FIG. 8 is a block diagram showing a second internal configuration of the estimation unit 806. In addition to the mixing unit 506, the signal ratio estimation unit 503, and the correction unit 310, the estimation unit 806 further includes a mixing unit 1106, a signal ratio estimation unit 1103, and a correction unit 710. The correction units 310 and 710 in FIG. 8 are identical to the correction units 310 and 710 in FIG. 7 in terms of input/output signals and operation, and therefore detailed description thereof will be omitted. The mixing unit 506 in FIG. 8 is configured to input a noise correction estimate value e2C(k) instead of the correction reference signal xRC(k) (second mixed signal) in the mixing unit 506 in FIG. 5. The operation of the mixing unit 506 in FIG. 8 is identical to the operation of the mixing unit 506 in FIG. 5, and therefore detailed description thereof will be omitted.

The mixer 1106 mixes the corrected speech estimate e1C(k) (the corrected emphasis signal, i.e., the corrected estimate of the first signal) and the pseudo crosstalk n2(k) (the estimate of the third signal) using the coefficient vector w2(k) of the adaptive filter 803 to generate a second mixed signal n4(k). The operation of the mixer 1106 becomes identical to that of the mixer 506 by changing the input and output signals of the mixer 506 to those of the mixer 1106 as shown in FIG. 8, so a detailed description will be omitted.

The signal ratio estimator 1103 receives the noise estimate e2(k) and the second mixed signal n4(k) and estimates the ratio of the amplitude or power of the noise to the crosstalk as the fourth mixed ratio R4(k). The fourth mixed ratio R4(k) may be the ratio of the amplitude or power of the noise estimate e2(k) to the second mixed signal n4(k), or may be calculated by adding a small constant to these amplitudes or powers. Adding a small constant has the effect of preventing the quotient from diverging due to division. Alternatively, either or both of the noise estimate e2(k) and the second mixed signal n4(k) may be averaged before use. Averaging can improve the accuracy of the ratio calculation.

The second internal configuration of the estimation unit 806 shown in Figure 8 is equivalent to the first internal configuration of the estimation unit 806 shown in Figure 7. That is, the first internal configuration shown in Figure 7 generates two estimates for the amplitude or power ratio of noise and crosstalk using signal ratio estimation units 901 and 902, and mixes them to calculate the fourth mixture ratio R4(k). The second internal configuration shown in Figure 8 mixes two types of estimates, namely, the corrected speech estimate e1C(k) and the pseudo crosstalk n2(k), using the coefficient vector w2(k) of the adaptive filter 803 to generate a second mixed signal n4(k), determine the denominator, and then combines this with the noise estimate e2(k), which is the numerator, to calculate the fourth mixture ratio R4(k). These two types of configurations are possible because the first internal configuration shown in FIG. 7 and the second internal configuration shown in FIG. 8 use the same numerator, i.e., the noise estimate e2(k), when estimating the ratio of the amplitude or power of noise to crosstalk at the input terminal 202. The second internal configuration of the estimator 806 shown in FIG. 8 has fewer components and is simpler than the first internal configuration shown in FIG. 7.

With the above configuration, this embodiment can smoothly update the coefficients without forcibly setting a special value for the step size, even when the gain of the acoustic impulse response approximated by the adaptive filters 203 and 803 is less than 1 and crosstalk is present. As a result, it is possible to obtain an output signal with little residual noise and little signal distortion.

4. Other Embodiments
Although multiple embodiments of the present invention have been described in detail above, any system or device that combines the separate features included in each embodiment also falls within the scope of the present invention.

The present invention may also be applied to a system consisting of multiple devices, or to a stand-alone device. Furthermore, the present invention can also be applied when an information processing program (signal processing program) that realizes the functions of the above-described embodiments is supplied directly or remotely to a system or device. Such a program is executed by a processor such as a DSP (Digital Signal Processor) that constitutes a signal processing device or noise cancellation device. Furthermore, the scope of the present invention also includes programs installed on a computer to realize the functions of the present invention on the computer, media on which such programs are stored, and WWW (World Wide Web) servers from which such programs can be downloaded.

Figure 9 is a configuration diagram of a computer 1200 that executes a signal processing program when the first to third embodiments are configured using the signal processing program. The computer 1200 includes an input unit 1201, a processor 1203, an output unit 1202, and a memory 1204.

Processor 1203 controls the operation of computer 1200 by reading signal processing programs stored in memory 1204. Processor 1203 is, for example, a processor such as a DSP, a CPU (Central Processing Unit), or an MPU (Micro-Processing Unit). Memory 1204 includes one or more of RAM (Random Access Memory), ROM (Read-Only Memory), flash memory, EPROM (Erasable Programmable Read-Only Memory), and EEPROM (Electrically Erasable Programmable Read-Only Memory).

Figure 10 is a flowchart showing an example of signal processing by the processor of the computer shown in Figure 9. The example shown in Figure 10 is a flowchart when the computer 1200 functions as the signal processing device 100 according to the first embodiment.

As shown in FIG. 10 , in step S10, the processor 1203 that has executed the signal processing program first inputs, from the input unit 1201, a first mixture signal xP(k) in which a first signal and a second signal are mixed, and then inputs a second mixture signal xR(k) in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed.
In step S11, processor 1203 processes second mixture signal xR(k) with a first adaptive filter (adaptive filter 103) to generate an estimate n1(k) of the second signal, and in step S12, generates an estimate e1(k) of the first signal from first mixture signal xP(k) and the estimate n1(k) of the second signal.
In step S13, processor 1203 estimates the ratio of the amplitudes or powers of the first signal and the second signal as a first mixture ratio R1(k) using the estimated value e1(k) of the first signal, the estimated value n1(k) of the second signal, the second mixture signal xR(k), the first mixture signal xP(k), and the coefficient vector w1(k) of the first adaptive filter (adaptive filter 103).
In step S14, the processor 1203 controls the generation of the estimated value n1(k) of the second signal using the first mixture ratio R1(k), thereby achieving the same effects as in the first embodiment.

Furthermore, when the computer 1200 functions as the signal processing device according to the second embodiment, the processor 1203 receives the first mixture signal xP(k) and the second mixture signal xR(k) in step S10, and processes the second mixture signal xR(k) using a first adaptive filter (adaptive filter 203) in step S11 to generate an estimated value n1(k) of the second signal. In step S12, the processor 1203 generates an estimated value e1(k) of the first signal from the first mixture signal xP(k) and the estimated value n1(k) of the second signal. In step S13, the processor 1203 generates a first mixture ratio R1(k) using the estimated value e1(k) of the first signal, the estimated value n1(k) of the second signal, the second mixture signal xR(k), the first mixture signal xP(k), and coefficient 141 of the first adaptive filter (adaptive filter 203). In step S14, the processor 1203 controls the generation of the estimated value n1(k) of the second signal using the first mixture ratio R1(k).

Furthermore, when the computer 1200 functions as the signal processing device according to the third embodiment, in step S13, the processor 1203 generates a first mixture ratio R1(k) using the estimated value e1(k) of the first signal, the estimated value n1(k) of the second signal, the estimated value e2(k) of the fourth signal, the first mixture signal xP(k), and the coefficient 141 of the first adaptive filter (adaptive filter 203). Furthermore, the processor 1203 processes the estimated value e1(k) of the first signal with the second adaptive filter (adaptive filter 803) to generate an estimated value n2(k) of the third signal, and subtracts the estimated value n2(k) of the third signal from the second mixture signal xR(k) to generate an estimated value e2(k) of the fourth signal. The processor 1203 also processes the fourth signal estimate e2(k) in place of the third signal estimate xR(k) in the first adaptive filter (adaptive filter 203), and further estimates the amplitude or power ratio of noise to crosstalk as a fourth mixture ratio R4(k) using the fourth signal estimate e2(k), the third signal estimate n2(k), the first signal estimate e1(k), the second mixture signal xR(k), and the coefficients of the second adaptive filter (adaptive filter 803).

[5. Other]
Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using known methods. In addition, the information including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

Furthermore, the components of each device shown in the figure are functional concepts and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

6. Effects
As described above, the signal processing device 100 according to the first embodiment includes a first input unit 101 (corresponding to an example of a first input means), a second input unit 102 (corresponding to an example of a second input means), an adaptive filter 103 (corresponding to an example of a first adaptive filter), a subtraction unit 104 (corresponding to an example of a first subtraction unit), and an estimation unit 106. The first input unit 101 receives a first mixture signal xP(k) in which a first signal and a second signal are mixed. The second input unit 102 receives a second mixture signal xR(k) in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed. The adaptive filter 103 filters the second mixture signal xR(k) to generate an estimate n1(k) of the second signal. The subtractor 104 generates an estimate e1(k) of the first signal from the first mixture signal xP(k) and the estimate n1(k) of the second signal. The estimator 106 estimates a first mixture ratio R1(k), which is the ratio of the amplitudes or powers of the first signal and the second signal at the input terminal 201, using the estimate e1(k) of the first signal, the estimate n1(k) of the second signal, the second mixture signal xR(k), the first mixture signal xP(k), and the coefficients 141 of the adaptive filter 103. When the value of the first mixture ratio R1(k) obtained by the estimator 106 is large, the coefficient update controller 107 outputs a control signal μ1(k) to the adaptive filter 103 to reduce the amount of update of the coefficients 141 of the adaptive filter 103. The signal processing device 100 controls the adaptive filter 103 using the control signal μ1(k). As a result, the signal processing device 100 can achieve both fast convergence and a low-distortion output signal without manually controlling the step size μ1(k) even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The signal processing device according to the second embodiment also includes an input terminal 201 (corresponding to an example of a first input means), an input terminal 202 (corresponding to an example of a second input means), an adaptive filter 203 (corresponding to an example of a first adaptive filter), a subtraction unit 204 (corresponding to an example of a first subtraction unit), and an estimation unit 206. The input terminal 201 receives a noisy signal xP(k) (corresponding to an example of a first mixed signal) that is a mixture of an audio signal (corresponding to an example of a first signal) and noise (corresponding to an example of an second signal). The input terminal 202 receives a reference signal xR(k) (corresponding to an example of a second mixed signal) that is a mixture of a signal correlated with the audio signal (corresponding to an example of a third signal) and a signal correlated with the noise (corresponding to an example of an fourth signal). The adaptive filter 203 filters the reference signal xR(k) to generate pseudo-noise n1(k) (corresponding to an example of an estimated value of the second signal). The subtractor 104 generates a speech estimate e1(k) (corresponding to an example of the first signal estimate e1(k)) from the noisy signal xP(k) and the pseudo-noise n1(k). The estimator 206 estimates the amplitude or power ratio of the speech signal to noise at the input terminal 201 as a first mixture ratio R1(k) using the speech estimate e1(k), the pseudo-noise n1(k), the reference signal xR(k), the noisy signal xP(k), and the coefficient vector w1(k) of the adaptive filter 203. The noise canceller 200 controls the adaptive filter 203 using the first mixture ratio R1(k). As a result, the signal processing device according to the second embodiment can achieve both fast convergence and a low-distortion output signal without manually controlling the step size μ1(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The estimation unit 206 also includes a signal ratio estimation unit 301 (corresponding to an example of a first signal ratio estimation unit), a signal ratio estimation unit 302 (corresponding to an example of a second signal ratio estimation unit), a correction unit 310, and a mixing unit 305 (corresponding to an example of a first mixing unit). The signal ratio estimation unit 301 uses the speech estimate e1(k) and the pseudo-noise n1(k) to estimate the amplitude or power ratio of the speech signal to noise at the input terminal 201 as a second mixing ratio R2(k). The correction unit 310 receives the reference signal xR(k) and the noisy signal xP(k) as inputs and calculates a corrected reference signal xRC(k). The signal ratio estimation unit 302 uses the speech estimate e1(k) and the corrected reference signal xRC(k) to estimate the amplitude or power ratio of the speech signal to noise at the input terminal 201 as a third mixing ratio R3(k). The mixer 305 generates the first mixture ratio R1(k) by mixing the second mixture ratio R2(k) and the third mixture ratio R3(k) based on the time change of the coefficient vector w1(k) of the adaptive filter 203. As a result, the signal processing device according to the second embodiment can achieve both fast convergence and a low-distortion output signal without manually controlling the step size μ1(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

Furthermore, the mixer 305 sets the content ratio of the third mixing ratio R3(k) in the first mixing ratio R1(k) to 100% when the coefficient update of the adaptive filter 203 begins, and sets the content ratio of the third mixing ratio R3(k) in the first mixing ratio R1(k) to 0% when the change over time in the coefficient vector w1(k) of the adaptive filter 203 becomes sufficiently small. As a result, the signal processing device according to the second embodiment can achieve both fast convergence and a low-distortion output signal without manually controlling the step size μ1(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The estimation unit 206 also includes a correction unit 310, a mixing unit 506 (corresponding to an example of a second mixing unit), and a signal ratio estimation unit 503 (corresponding to an example of a third signal ratio estimation unit). The correction unit 310 receives the reference signal xR(k) and the degradation signal xP(k) as input and calculates a corrected reference signal xRC(k). The mixing unit 506 mixes the corrected reference signal xRC(k) and the pseudo-noise n1(k) based on the time variation of the coefficient vector w1(k) of the adaptive filter 203 to generate a first mixed signal n3(k). The signal ratio estimation unit 503 uses the first mixed signal n3(k) and the estimated value e1(k) of the speech to estimate the amplitude or power ratio of the speech signal to noise at the input terminal 201 as a first mixing ratio R1(k). As a result, the signal processing device according to the second embodiment can achieve both fast convergence and low-distortion output signals without manually controlling the step size μ1(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

Furthermore, the mixer 506 sets the content ratio of the correction reference signal xRC(k) in the first mixed signal n3(k) to 100% when the coefficient update of the adaptive filter 203 begins, and sets the content ratio of the correction reference signal xRC(k) in the first mixed signal n3(k) to 0% when the time change of the coefficient vector w1(k) of the adaptive filter 203 becomes sufficiently small. As a result, the signal processing device according to the second embodiment can achieve both fast convergence and a low-distortion output signal without manually controlling the step size μ1(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The signal processing device according to the third embodiment also includes an input terminal 201, an input terminal 202, an adaptive filter 203, a subtraction unit 204, a subtraction unit 804 (corresponding to an example of a second subtraction unit), an adaptive filter 803 (corresponding to an example of a second adaptive filter), and an estimation unit 806. The adaptive filter 803 filters the estimated value e1(k) of the first signal to generate pseudo crosstalk n2(k) (corresponding to an example of an estimated value of the third signal). The subtraction unit 804 subtracts the pseudo crosstalk n2(k) from the reference signal xR(k) to generate a noise estimated value e2(k) (corresponding to an example of an estimated value of the fourth signal). The adaptive filter 203 receives the noise estimated value e2(k) as input instead of the reference signal xR(k). In addition to the functions of the estimation unit 206, the estimation unit 806 further uses the noise estimate e2(k), the pseudo crosstalk n2(k), the first signal estimate e1(k), the reference signal xR(k), and the coefficients of the adaptive filter 803 to estimate the ratio of the amplitude or power of the noise to the crosstalk at the input terminal 202 as a fourth mixing ratio R4(k). The signal processing device according to the third embodiment controls the adaptive filter 803 using the fourth mixing ratio R4(k). As a result, the signal processing device according to the third embodiment can achieve both fast convergence and a low-distortion output signal without manually controlling the step size μ2(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The estimation unit 806 also includes a signal ratio estimation unit 301, a correction unit 310, a signal ratio estimation unit 302, a mixing unit 305, a signal ratio estimation unit 901 (corresponding to an example of a fourth signal ratio estimation unit), a correction unit 710, a signal ratio estimation unit 902 (corresponding to an example of a fifth signal ratio estimation unit), and a mixing unit 905 (corresponding to an example of a third mixing unit). The signal ratio estimation unit 301 uses the speech estimate e1(k) and the pseudo-noise n1(k) to estimate the amplitude or power ratio of the speech signal to noise at the input terminal 201 as a second mixing ratio R2(k). The correction unit 310 receives the noise estimate e2(k) and the noisy signal xP(k) as input and calculates a corrected noise estimate e2C(k). The signal ratio estimator 302 uses the speech estimate e1(k) and the corrected noise estimate e2C(k) to estimate the amplitude or power ratio of speech to noise at the input terminal 201 as a third mixture ratio R3(k). The mixer 305 mixes the second mixture ratio R2(k) and the third mixture ratio R3(k) based on time changes in the coefficient vector w1(k) of the adaptive filter 203 to generate the first mixture ratio R1(k). The signal ratio estimator 901 uses the noise estimate e2(k) and the pseudo crosstalk n2(k) to estimate the amplitude or power ratio of noise to crosstalk at the input terminal 202 as a fifth mixture ratio R5(k). The corrector 710 receives the speech estimate e1(k) and the reference signal xR(k) as input and calculates the corrected speech estimate e1C(k). The signal ratio estimator 902 uses the noise estimate e2(k) and the corrected speech estimate e1C(k) to estimate the amplitude or power ratio of noise to crosstalk at the input terminal 202 as a sixth mixture ratio R6(k). The mixer 905 mixes the fifth mixture ratio R5(k) and the sixth mixture ratio R6(k) based on time changes in the coefficient vector w2(k) of the adaptive filter 803 to generate a fourth mixture ratio R4(k). As a result, the signal processing device according to the third embodiment can achieve both fast convergence and low-distortion output signals without manually controlling the step sizes μ1(k) and μ2(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The estimation unit 806 also includes a correction unit 310, a mixing unit 506, a signal ratio estimation unit 503, a correction unit 710, a mixing unit 1106 (corresponding to an example of a fourth mixing unit), and a signal ratio estimation unit 1103 (corresponding to an example of a sixth signal ratio estimation unit). The correction unit 310 receives the noise estimate e2(k) and the noisy signal xP(k) as input and calculates a corrected noise estimate e2C(k). The mixing unit 506 mixes the corrected noise estimate e2C(k) and the pseudo-noise n1(k) based on the time variation of the coefficient vector w1(k) of the adaptive filter 203 to generate a first mixed signal n3(k). The signal ratio estimation unit 503 uses the first mixed signal n3(k) and the speech estimate e1(k) to estimate the amplitude or power ratio of speech to noise at the input terminal 201 as a first mixing ratio R1(k). The correction unit 710 receives the speech estimate e1(k) and the reference signal xR(k) as input and calculates the corrected speech estimate e1C(k). The mixer 1106 mixes the corrected speech estimate e1C(k) and the pseudo crosstalk n2(k) based on time changes in the coefficient vector w2(k) of the adaptive filter 803 to generate a second mixed signal n4(k). The signal ratio estimator 1103 uses the second mixed signal n4(k) and the noise estimate e2(k) to estimate the ratio of the amplitude or power of the noise to the crosstalk at the input terminal 202 as a fourth mixing ratio R4(k). As a result, the signal processing device according to the third embodiment can achieve both fast convergence and low-distortion output signals without manually controlling the step sizes μ1(k) and μ2(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

Furthermore, the mixer 506 sets the content ratio of the noise correction estimate value e2C(k) in the first mixed signal n3(k) to 100% when the coefficient update of the adaptive filter 203 begins, and sets the content ratio of the noise correction estimate value e2C(k) in the first mixed signal n3(k) to 0% when the time change of the coefficient vector w1(k) of the adaptive filter 203 becomes sufficiently small. Furthermore, the mixer 1106 sets the content ratio of the speech correction estimate value e1C(k) in the second mixed signal n4(k) to 100% when the coefficient update of the adaptive filter 803 begins, and sets the content ratio of the speech correction estimate value e1C(k) in the second mixed signal n4(k) to 0% when the time change of the coefficient vector w2(k) of the adaptive filter 803 becomes sufficiently small. As a result, the signal processing device according to the third embodiment can achieve both fast convergence and low-distortion output signals without manually controlling the step sizes μ1(k) and μ2(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

Furthermore, the time change in coefficient vector w1(k) of adaptive filter 203 is the time change in the sum of squares or sum of absolute values of coefficient vector w1(k), and the time change in coefficient vector w2(k) of adaptive filter 803 is the time change in the sum of squares or sum of absolute values of coefficient vector w2(k). As a result, the signal processing devices according to the first to third embodiments can achieve both fast convergence and low-distortion output signals without manually controlling the step sizes μ1(k) and μ2(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

Furthermore, the time change in coefficient vector w1(k) of adaptive filter 203 is the time change in the partial sum of squares or absolute values of coefficient vector w1(k), and the time change in coefficient vector w2(k) of adaptive filter 803 is the time change in the partial sum of squares or absolute values of coefficient vector w2(k). As a result, the signal processing devices according to the first to third embodiments can achieve both fast convergence and low-distortion output signals without manually controlling the step sizes μ1(k) and μ2(k), even if the gain of the acoustic impulse response approximated by the adaptive filter is less than 1.

The above describes in detail the embodiments of the present application based on the drawings, but this is merely an example, and the present invention can be implemented in other forms that incorporate various modifications and improvements based on the knowledge of those skilled in the art, including the aspects described in the Disclosure of the Invention section.

Furthermore, the above-mentioned "section, module, unit" can be read as "means" or "circuit", etc. For example, a subtraction section can be read as subtraction means or subtraction circuit.

(Appendix 1)
a first input means for inputting a first mixed signal in which the first signal and the second signal are mixed;
a second input means for inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed;
a first adaptive filter (adaptive filter 103, 203) for filtering the second mixture signal to generate an estimate of the second signal;
a first subtraction unit (subtraction unit 104, 204) that generates an estimate of the first signal from the first mixed signal and an estimate of the second signal;
an estimation unit (estimation unit 106, 206, 806) that estimates a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, the first mixture signal, and a coefficient of the first adaptive filter;
Equipped with
The signal processing device controls the first adaptive filter using the first mixture ratio.
(Appendix 2)
The estimation unit (estimation unit 106, 206)
a first signal ratio estimator (signal ratio estimator 301) that estimates a ratio of amplitudes or powers of the first signal and the second signal as a second mixture ratio using the estimated value of the first signal and the estimated value of the second signal;
a first correction unit (correction unit 310) that corrects the second mixture signal using the second mixture signal and the first mixture signal to generate a corrected second mixture signal;
a second signal ratio estimator (signal ratio estimator 302) that estimates a ratio of amplitudes or powers of the first signal and the second signal as a third mixture ratio using the estimated value of the first signal and the corrected second mixture signal;
a first mixer (mixer 305) that mixes the second mixture ratio and the third mixture ratio based on a time change in a coefficient of the first adaptive filter to generate the first mixture ratio;
2. The signal processing device according to claim 1, comprising:
(Appendix 3)
The first mixing section (mixing section 305)
the signal processing device according to claim 2, wherein a content ratio of the third mixture ratio in the first mixture ratio is set to 100% when a coefficient update of the first adaptive filter starts, and when a change over time in the coefficient of the first adaptive filter becomes sufficiently small, a content ratio of the third mixture ratio in the first mixture ratio is set to 0%.
(Appendix 4)
The estimation unit (estimation unit 106, 206)
a first correction unit (correction unit 310) that corrects the second mixture signal using the second mixture signal and the first mixture signal to generate a corrected second mixture signal;
a second mixer (mixer 506) that mixes the corrected second mixed signal and the estimated value of the second signal based on time changes in the coefficients of the first adaptive filter to generate a first mixed signal;
a third signal ratio estimator (signal ratio estimator 503) that estimates a ratio of amplitudes or powers of the first signal and the second signal as the first mixture ratio using the first mixed signal and an estimated value of the first signal;
2. The signal processing device according to claim 1, comprising:
(Appendix 5)
The second mixing unit (mixing unit 506)
The signal processing device according to claim 4, wherein the content ratio of the corrected second mixture signal in the first mixture signal is set to 100% when coefficient update of the first adaptive filter starts, and when a change over time in the coefficient of the first adaptive filter becomes sufficiently small, the content ratio of the corrected second mixture signal in the first mixture signal is set to 0%.
(Appendix 6)
a second adaptive filter (adaptive filter 803) that filters the estimate of the first signal to produce an estimate of the third signal;
a second subtraction unit (subtraction unit 804) that subtracts the estimated value of the third signal from the second mixed signal to generate the estimated value of the fourth signal,
The first adaptive filter is
an estimate of the fourth signal is input instead of the second mixed signal;
The estimation unit
further estimating, as a fourth mixture ratio, a ratio of amplitudes or powers of the fourth signal and the third signal using the estimated value of the fourth signal, the estimated value of the third signal, the estimated value of the first signal, the second mixture signal, and a coefficient of the second adaptive filter;
2. The signal processing device according to claim 1, wherein the second adaptive filter is controlled using the fourth mixing ratio.
(Appendix 7)
The estimation unit (estimation unit 806)
a first signal ratio estimator (signal ratio estimator 301) that estimates a ratio of amplitudes or powers of the first signal and the second signal as a second mixture ratio using the estimated value of the first signal and the estimated value of the second signal;
a first correction unit (correction unit 310) that corrects the estimate of the fourth signal using the estimate of the fourth signal and the first mixed signal to generate a corrected estimate of the fourth signal;
a second signal ratio estimator (signal ratio estimator 302) that estimates a ratio of amplitudes or powers of the first signal and the second signal as a third mixture ratio using the estimated value of the first signal and the corrected estimated value of the fourth signal;
a first mixer (mixer 305) that mixes the second mixture ratio and the third mixture ratio based on a time change in a coefficient of the first adaptive filter to generate the first mixture ratio;
a fourth signal ratio estimator (signal ratio estimator 901) that estimates a ratio of amplitudes or powers of the fourth signal and the third signal as a fifth mixture ratio using the estimated value of the fourth signal and the estimated value of the third signal;
a second correction unit (correction unit 710) that corrects the estimate of the first signal using the estimate of the first signal and the second mixed signal to generate a corrected estimate of the first signal;
a fifth signal ratio estimator (signal ratio estimator 902) that estimates a ratio of amplitudes or powers of the fourth signal and the third signal as a sixth mixture ratio using the estimated value of the fourth signal and the corrected estimated value of the first signal;
a third mixer (mixer 905) that mixes the fifth mixture ratio and the sixth mixture ratio based on a time change in a coefficient of the second adaptive filter to generate the fourth mixture ratio;
7. The signal processing device according to claim 6, comprising:
(Appendix 8)
The first mixing section (mixing section 305)
setting a content ratio of the third mixture ratio in the first mixture ratio to 100% when a coefficient update of the first adaptive filter starts, and setting a content ratio of the third mixture ratio in the first mixture ratio to 0% when a time change in the coefficient of the first adaptive filter becomes sufficiently small;
The third mixing section (mixing section 905)
the signal processing device according to claim 7 or 8, wherein a content ratio of the sixth mixing ratio in the fourth mixing ratio is set to 100% when a coefficient update of the second adaptive filter starts, and a content ratio of the sixth mixing ratio in the fourth mixing ratio is set to 0% when a change over time in the coefficient of the second adaptive filter becomes sufficiently small.
(Appendix 9)
The estimation unit (estimation unit 806)
a first correction unit (correction unit 310) that corrects the estimate of the fourth signal using the estimate of the fourth signal and the first mixed signal to generate a corrected estimate of the fourth signal;
a second mixer (mixer 506) that mixes the corrected estimate of the fourth signal and the estimate of the second signal based on time variations in the coefficients of the first adaptive filter to generate a first mixed signal;
a third signal ratio estimator (signal ratio estimator 503) that estimates a ratio of amplitudes or powers of the first signal and the second signal as the first mixture ratio using estimated values of the first mixed signal and the first signal;
a second correction unit (correction unit 710) that corrects the estimate of the first signal using the estimate of the first signal and the second mixed signal to generate a corrected estimate of the first signal;
a fourth mixer (mixer 1106) that mixes the corrected estimate of the first signal and the estimate of the third signal based on time variations in the coefficients of the second adaptive filter to generate a second mixed signal;
a sixth signal ratio estimator (signal ratio estimator 1103) that estimates a ratio of amplitudes or powers of the fourth signal and the third signal as the fourth mixture ratio using estimated values of the second mixed signal and the fourth signal;
7. The signal processing device according to claim 6, comprising:
(Appendix 10)
The second mixing unit (mixing unit 506)
setting a content ratio of the corrected estimate value of the fourth signal in the first mixed signal to 100% when coefficient update of the first adaptive filter starts, and setting a content ratio of the corrected estimate value of the fourth signal in the first mixed signal to 0% when a time change in the coefficient of the first adaptive filter becomes sufficiently small;
The fourth mixing unit (mixing unit 1106)
10. The signal processing device according to claim 9, wherein a content ratio of the corrected estimate value of the first signal in the second mixed signal is set to 100% when coefficient update of the second adaptive filter starts, and a content ratio of the corrected estimate value of the first signal in the second mixed signal is set to 0% when a time change in the coefficient of the second adaptive filter becomes sufficiently small.
(Appendix 11)
The time change of the coefficient is
The signal processing device according to claim 3, 5, 8, or 10, wherein the sum of squares or sum of absolute values of the coefficients is a time change.
(Appendix 12)
The time change of the coefficient is
The signal processing device according to claim 3, 5, 8, or 10, wherein the partial sum of squares or the partial sum of absolute values of the coefficients is a time change.
(Appendix 13)
A first mixed signal in which the first signal and the second signal are mixed is input;
a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed is input;
processing the second mixed signal with a first adaptive filter (adaptive filter 103, 203) to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
estimating a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, the first mixture signal, and a coefficient of the first adaptive filter;
and using the first mixing ratio to control generation of an estimate of the second signal.
(Appendix 14)
processing the estimate of the first signal with a second adaptive filter (adaptive filter 803) to produce an estimate of the third signal;
subtracting the estimate of the third signal from the second mixed signal to generate an estimate of the fourth signal;
the first adaptive filter processes an estimate of the fourth signal instead of the second mixed signal;
further using the estimated value of the fourth signal, the estimated value of the third signal, the estimated value of the first signal, the second mixed signal, and a coefficient of the second adaptive filter,
further estimating a ratio of amplitude or power of the fourth signal to the third signal as a fourth mixture ratio;
14. The signal processing method of claim 13, further comprising: controlling generation of an estimate of the third signal using the fourth mixing ratio.
(Appendix 15)
On the computer,
a step of inputting a first mixed signal in which a first signal and a second signal are mixed; and a step of inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed.
processing the second mixed signal with a first adaptive filter (adaptive filter 103, 203) to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
a step of estimating a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, the first mixture signal, and a coefficient of the first adaptive filter;
and controlling generation of an estimate of the second signal using the first mixture ratio.
(Appendix 16)
On the computer,
processing the estimate of the first signal with a second adaptive filter (adaptive filter 803) to produce an estimate of the third signal;
subtracting the estimate of the third signal from the second mixture signal to generate an estimate of the fourth signal;
the first adaptive filter processes an estimate of the fourth signal instead of the second mixed signal;
further using the estimated value of the fourth signal, the estimated value of the third signal, the estimated value of the first signal, the second mixed signal, and a coefficient of the second adaptive filter,
further estimating a ratio of amplitude or power of the fourth signal to the third signal as a fourth mixing ratio;
and controlling generation of an estimate of the third signal using the fourth mixing ratio.

100 Signal processing device 101 First input unit 102 Second input unit 103, 203 Adaptive filter (corresponding to an example of a first adaptive filter)
104, 204 Subtraction unit (corresponding to an example of a first subtraction unit)
106, 206, 806 Estimation unit 107 Coefficient update control unit 141 Coefficient 200, 800 Noise canceller (corresponding to an example of a signal processing device)
201 Input terminal (corresponding to an example of a first input unit)
202 Input terminal (corresponding to an example of a second input unit)
205, 805 Output terminal 301 Signal ratio estimation unit (corresponding to an example of a first signal ratio estimation unit)
302 Signal ratio estimation unit (corresponding to an example of a second signal ratio estimation unit)
305 Mixing section (corresponding to an example of the first mixing section)
310 Correction unit (corresponding to an example of a first correction unit)
506 Mixing section (corresponding to an example of the second mixing section)
503 Signal ratio estimation unit (corresponding to an example of a third signal ratio estimation unit)
710 Correction unit (corresponding to an example of a second correction unit)
803 Adaptive filter (corresponding to an example of a second adaptive filter)
804 Subtraction unit (corresponding to an example of a second subtraction unit)
901 Signal ratio estimation unit (corresponding to an example of a fourth signal ratio estimation unit)
902 signal ratio estimator (corresponding to an example of a fifth signal ratio estimator)
905 Mixing section (corresponding to an example of the third mixing section)
1106 Mixing section (corresponding to an example of the fourth mixing section)
1103: Signal ratio estimation unit (corresponding to an example of a sixth signal ratio estimation unit)
A: Signal source B: Signal source xP(k): First mixture signal xR(k): Second mixture signal xRC(k): Corrected second mixture signal e1(k): Estimated value of the first signal, estimated value of the speech signal e2(k): Estimated value of noise (corresponding to an example of the estimated value of the fourth signal)
e1C(k): Corrected estimate of the first signal, corrected estimate of the speech signal e2C(k): Corrected estimate of the noise (corresponding to an example of the corrected estimate of the fourth signal)
n1(k) is an estimate of the second signal, pseudo-noise (corresponding to an example of an estimate of the second signal)
n2(k) pseudo crosstalk (corresponding to an example of an estimate of the third signal)
n3(k) mixed signal (corresponding to an example of the first mixed signal)
n4(k) mixed signal (corresponding to an example of the second mixed signal)
R1(k) 1st mixing ratio R2(k) 2nd mixing ratio R3(k) 3rd mixing ratio R4(k) 4th mixing ratio R5(k) 5th mixing ratio R6(k) 6th mixing ratio

Claims (17)

a first input means for inputting a first mixed signal in which the first signal and the second signal are mixed;
a second input means for inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed;
a first adaptive filter for filtering the second mixed signal to generate an estimate of the second signal;
a first subtraction unit that generates an estimate of the first signal from the first mixed signal and an estimate of the second signal;
an estimation unit that estimates a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, the first mixture signal, and a coefficient of the first adaptive filter;
Equipped with
controlling the first adaptive filter using the first mixture ratio ;
The estimation unit
a first signal ratio estimator that estimates a ratio of amplitudes or powers of the first signal and the second signal as a second mixture ratio using the estimated value of the first signal and the estimated value of the second signal;
a first correction unit that averages the power of the second mixture signal and the first mixture signal, sets a ratio between the averaged second mixture signal and the first mixture signal as a magnification, and generates a corrected second mixture signal by multiplying the square root of the magnification by the second mixture signal;
a second signal ratio estimator that estimates a ratio of amplitudes or powers of the first signal and the second signal as a third mixture ratio using the estimated value of the first signal and the corrected second mixture signal;
a first mixer that mixes the second mixture ratio and the third mixture ratio by setting a weight of the third mixture ratio to a large value at the start of updating the coefficients and decreasing the weight as the coefficients grow, to generate the first mixture ratio;
A signal processing device comprising : The first mixing section is
2. The signal processing device according to claim 1, wherein a content ratio of the third mixture ratio in the first mixture ratio is set to 100% when a coefficient update of the first adaptive filter starts, and when a change over time in the coefficient of the first adaptive filter becomes sufficiently small, a content ratio of the third mixture ratio in the first mixture ratio is set to 0%. a first input means for inputting a first mixed signal in which the first signal and the second signal are mixed;
a second input means for inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed;
a first adaptive filter for filtering the second mixed signal to generate an estimate of the second signal;
a first subtraction unit that generates an estimate of the first signal from the first mixed signal and an estimate of the second signal;
an estimation unit that estimates a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the estimated value of the first signal, the estimated value of the second signal, the second mixture signal, the first mixture signal, and a coefficient of the first adaptive filter;
Equipped with
controlling the first adaptive filter using the first mixture ratio;
The estimation unit
a first correction unit that averages the power of the second mixture signal and the first mixture signal, sets a ratio between the averaged second mixture signal and the first mixture signal as a magnification, and generates a corrected second mixture signal by multiplying the square root of the magnification by the second mixture signal;
a second mixer that mixes the corrected second mixture signal and the estimated value of the second signal by setting a weight of the corrected second mixture signal to a large value when the coefficient update starts and decreasing the weight as the coefficient grows, to generate a first mixture signal;
a third signal ratio estimator that estimates, as the first mixture ratio, a ratio of amplitudes or powers of the first signal and the second signal using the first mixed signal and an estimated value of the first signal;
A signal processing device comprising : The second mixing section is
4. The signal processing device according to claim 3, wherein a content ratio of the corrected second mixture signal in the first mixture signal is set to 100% when coefficient update of the first adaptive filter starts, and when a change over time in the coefficient of the first adaptive filter becomes sufficiently small, a content ratio of the corrected second mixture signal in the first mixture signal is set to 0%. a second adaptive filter that filters the estimate of the first signal to generate an estimate of the third signal;
a second subtraction unit that subtracts the estimated value of the third signal from the second mixed signal to generate the estimated value of the fourth signal,
The first adaptive filter is
an estimate of the fourth signal is input instead of the second mixed signal;
The estimation unit
further estimating, as a fourth mixture ratio, a ratio of amplitudes or powers of the fourth signal and the third signal using the estimated value of the fourth signal, the estimated value of the third signal, the estimated value of the first signal, the second mixture signal, and a coefficient of the second adaptive filter;
The signal processing device according to claim 1 or 3 , wherein the second adaptive filter is controlled using the fourth mixing ratio. The estimation unit
a first signal ratio estimator that estimates a ratio of amplitudes or powers of the first signal and the second signal as a second mixture ratio using the estimated value of the first signal and the estimated value of the second signal;
a first correction unit that corrects the estimate of the fourth signal using the estimate of the fourth signal and the first mixed signal to generate a corrected estimate of the fourth signal;
a second signal ratio estimator that estimates a ratio of amplitudes or powers of the first signal and the second signal as a third mixture ratio using the estimated value of the first signal and the corrected estimated value of the fourth signal;
a first mixer that mixes the second mixture ratio and the third mixture ratio based on a time change in a coefficient of the first adaptive filter to generate the first mixture ratio;
a fourth signal ratio estimator that estimates a ratio of amplitudes or powers of the fourth signal and the third signal as a fifth mixture ratio using the estimated value of the fourth signal and the estimated value of the third signal;
a second correction unit that corrects the estimate of the first signal using the estimate of the first signal and the second mixed signal to generate a corrected estimate of the first signal;
a fifth signal ratio estimator that estimates a ratio of amplitudes or powers of the fourth signal and the third signal as a sixth mixture ratio using the estimated value of the fourth signal and the corrected estimated value of the first signal;
a third mixer that mixes the fifth mixture ratio and the sixth mixture ratio based on a time change in a coefficient of the second adaptive filter to generate the fourth mixture ratio;
The signal processing device according to claim 5 , comprising: The first mixing section
setting a content ratio of the third mixture ratio in the first mixture ratio to 100% when a coefficient update of the first adaptive filter starts, and setting a content ratio of the third mixture ratio in the first mixture ratio to 0% when a time change in the coefficient of the first adaptive filter becomes sufficiently small;
The third mixing section is
7. The signal processing device according to claim 6, wherein a content ratio of the sixth mixing ratio in the fourth mixing ratio is set to 100% when a coefficient update of the second adaptive filter starts, and a content ratio of the sixth mixing ratio in the fourth mixing ratio is set to 0% when a change over time in the coefficient of the second adaptive filter becomes sufficiently small. The estimation unit
a first correction unit that corrects the estimate of the fourth signal using the estimate of the fourth signal and the first mixed signal to generate a corrected estimate of the fourth signal;
a second mixer that mixes the corrected estimate of the fourth signal and the estimate of the second signal based on time changes in coefficients of the first adaptive filter to generate a first mixed signal;
a third signal ratio estimator that estimates a ratio of amplitudes or powers of the first signal and the second signal as the first mixture ratio using estimated values of the first mixed signal and the first signal;
a second correction unit that corrects the estimate of the first signal using the estimate of the first signal and the second mixed signal to generate a corrected estimate of the first signal;
a fourth mixer that mixes the corrected estimate of the first signal and the estimate of the third signal based on time changes in coefficients of the second adaptive filter to generate a second mixed signal;
a sixth signal ratio estimator that estimates a ratio of amplitudes or powers of the fourth signal and the third signal as the fourth mixture ratio using estimated values of the second mixed signal and the fourth signal;
The signal processing device according to claim 5 , comprising: The second mixing section is
setting a content ratio of the corrected estimate value of the fourth signal in the first mixed signal to 100% when coefficient update of the first adaptive filter starts, and setting a content ratio of the corrected estimate value of the fourth signal in the first mixed signal to 0% when a time change in the coefficient of the first adaptive filter becomes sufficiently small;
The fourth mixing section is
When coefficient updating of the second adaptive filter starts, the content ratio of the corrected estimated value of the first signal in the second mixed signal is set to 100%, and when time change in the coefficient of the second adaptive filter becomes sufficiently small, the content ratio of the corrected estimated value of the first signal in the second mixed signal is set to 0%.
The signal processing device according to claim 8 . The time change of the coefficient is
The signal processing device according to claim 2 , 4 , 7 or 9 , wherein the sum of squares or sum of absolute values of the coefficients is a time variation. The time change of the coefficient is
The signal processing device according to claim 2 , 4 , 7 or 9 , wherein the partial sum of squares or the partial sum of absolute values of the coefficients is a time variation. A first mixed signal in which the first signal and the second signal are mixed is input;
a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed is input;
processing the second mixed signal with a first adaptive filter to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
estimating a ratio of amplitudes or powers of the first signal and the second signal as a second mixture ratio using the estimated value of the first signal and the estimated value of the second signal;
averaging the power of the second mixture signal and the first mixture signal, determining a ratio between the averaged second mixture signal and the first mixture signal as a magnification, and generating a corrected second mixture signal by multiplying the square root of the magnification by the second mixture signal;
estimating a ratio of amplitudes or powers of the first signal and the second signal as a third mixture ratio using the estimated value of the first signal and the corrected second mixture signal;
generating a first mixing ratio by mixing the second mixing ratio and the third mixing ratio by setting a weight of the third mixing ratio to a large value at the start of updating the coefficients of the first adaptive filter and decreasing the weight as the coefficients grow;
and using the first mixing ratio to control generation of an estimate of the second signal. A first mixed signal in which the first signal and the second signal are mixed is input;
a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed is input;
processing the second mixed signal with a first adaptive filter to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
averaging the power of the second mixture signal and the first mixture signal, determining a ratio between the averaged second mixture signal and the first mixture signal as a magnification, and generating a corrected second mixture signal by multiplying the square root of the magnification by the second mixture signal;
generating a first mixed signal by mixing the corrected second mixed signal and the estimated value of the second signal, by setting a weight of the corrected second mixed signal to a large value when updating the coefficients of the first adaptive filter starts and decreasing the weight as the coefficients grow;
using the first mixed signal and the estimated value of the first signal, estimating an amplitude or power ratio between the first signal and the second signal as a first mixture ratio;
and controlling generation of an estimate of the second signal using the first mixing ratio.
Signal processing methods. processing the estimate of the first signal with a second adaptive filter to generate an estimate of the third signal;
subtracting the estimate of the third signal from the second mixed signal to generate an estimate of the fourth signal;
the first adaptive filter processes an estimate of the fourth signal instead of the second mixed signal;
further using the estimated value of the fourth signal, the estimated value of the third signal, the estimated value of the first signal, the second mixed signal, and a coefficient of the second adaptive filter,
further estimating a ratio of amplitude or power of the fourth signal to the third signal as a fourth mixture ratio;
14. The signal processing method according to claim 12 or 13 , wherein the fourth mixing ratio is used to control generation of the estimate of the third signal. On the computer,
a step of inputting a first mixed signal in which a first signal and a second signal are mixed; and a step of inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed.
processing the second mixed signal with a first adaptive filter to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
estimating a ratio of amplitudes or powers of the first signal and the second signal as a second mixture ratio using the estimated value of the first signal and the estimated value of the second signal;
a step of averaging the power of the second mixture signal and the first mixture signal, determining a ratio of the averaged second mixture signal to the first mixture signal as a scaling factor, and generating a corrected second mixture signal by multiplying the square root of the scaling factor by the second mixture signal;
estimating a ratio of amplitudes or powers of the first signal and the second signal as a third mixture ratio using the estimated value of the first signal and the corrected second mixture signal;
generating a first mixture ratio by mixing the second mixture ratio and the third mixture ratio by setting a weight of the third mixture ratio to a large value at the start of updating the coefficients of the first adaptive filter and decreasing the weight as the coefficients grow;
and controlling generation of an estimate of the second signal using the first mixture ratio. On the computer,
inputting a first mixed signal in which the first signal and the second signal are mixed;
inputting a second mixed signal in which a third signal correlated with the first signal and a fourth signal correlated with the second signal are mixed;
processing the second mixed signal with a first adaptive filter to generate an estimate of the second signal;
generating an estimate of the first signal from the first mixed signal and an estimate of the second signal;
a step of averaging the power of the second mixture signal and the first mixture signal, determining a ratio of the averaged second mixture signal to the first mixture signal as a scaling factor, and generating a corrected second mixture signal by multiplying the square root of the scaling factor by the second mixture signal;
a step of mixing the corrected second mixed signal and the estimated value of the second signal to generate a first mixed signal by setting a weight of the corrected second mixed signal to a large value at the start of updating the coefficients of the first adaptive filter and decreasing the weight as the coefficients grow;
a step of estimating a ratio of amplitudes or powers of the first signal and the second signal as a first mixture ratio using the first mixed signal and the estimated value of the first signal;
using the first mixing ratio to control generation of an estimate of the second signal;
A signal processing program that executes the above. On the computer,
processing the estimate of the first signal with a second adaptive filter to generate an estimate of the third signal;
subtracting the estimate of the third signal from the second mixture signal to generate an estimate of the fourth signal;
the first adaptive filter processes an estimate of the fourth signal instead of the second mixed signal;
further using the estimated value of the fourth signal, the estimated value of the third signal, the estimated value of the first signal, the second mixed signal, and a coefficient of the second adaptive filter,
further estimating a ratio of amplitude or power of the fourth signal to the third signal as a fourth mixing ratio;
and controlling generation of an estimate of the third signal using the fourth mixture ratio.