JP2018031820A - Signal processor, signal processing method, and signal processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately estimate a second signal if the magnitude of the second signal has suddenly increased, in a mixed signal of a first signal and the second signal.SOLUTION: A signal processor 800 includes: an estimation unit 803 for estimating a second signal of a mixed signal, by using the mixed signal of a first signal and the second signal and providing the second signal as an estimated second signal; and a determination unit 802 for determining whether the estimated second signal is too small, by using the mixed signal and the estimated second signal, the estimation unit 803 initializing the estimation processing on the basis of the result of determination by the determination unit 802.SELECTED DRAWING: Figure 14

Description

  The present invention relates to a signal processing technique for estimating a second signal in a mixed signal including a first signal and a second signal.

  In various audio terminals such as mobile phones, a second signal is generated from a mixed signal including a first signal (for example, a notable desired signal such as voice) and a second signal (for example, an undesired signal such as noise). A second signal estimation technique is used to estimate. The second signal estimation technique is used for a voice detection technique for determining whether or not the mixed signal contains the first signal based on the relationship between the mixed signal and the estimated second signal. Further, it is used in a noise suppression technique for generating an output signal in which the first signal is emphasized by suppressing the second signal.

  As a technique related to such a technique, in Patent Document 1, the volume of the second signal is estimated, and the threshold corresponding to the volume of the estimated second signal is compared with the volume of the mixed signal to detect the voice. Techniques for performing are disclosed.

  Patent Document 2 discloses a signal processing device that suppresses the second signal by processing a mixed signal (input signal) in which the first signal and the second signal are mixed. This apparatus estimates a background sound signal included in the mixed signal. And this apparatus suppresses suppression of a 2nd signal so that a 2nd signal may not become smaller than a background sound.

  Patent Document 3 discloses a signal processing device that suppresses the second signal by processing a mixed signal in which the first signal and the second signal are mixed. This apparatus analyzes the importance of the first signal included in the mixed signal for each of a plurality of frequency components constituting the first signal. As a result of the analysis, this apparatus suppresses the suppression of the second signal with respect to the frequency component with high importance as compared with the frequency component with low importance.

  Further, in Non-Patent Document 1, a suppression coefficient is obtained by using an initial average value when a mixed signal in which a first signal and a second signal are mixed is supplied as an estimated value of the second signal, and is included in the mixed signal. A method for suppressing two signals is disclosed.

  Non-Patent Document 2 discloses a method of enhancing the first signal by suppressing the second signal by multiplying the mixed signal by the suppression coefficient obtained from the estimated value of the second signal and the mixed signal. ing. In this method, the second signal is continuously estimated by weighting the mixed signal according to the estimated power ratio between the first signal and the second signal.

JP 2013-005418 International Publication No.2012-070670 International Publication No.2012-070668

Y. Ephraim and D. Malah, "Speech enhancement using a minimum mean-square error short-time spectral amplitude estimator," IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-32, no. 6, pp. 1109 -1121, Dec. 1984. M. Kato, A. Sugiyama, and M. Serizawa, "Noise suppression with high speech quality based on weighted noise estimation and MMSE STSA," IEICE Trans. Fundamentals (Japanese Edition), vol.J87-A, no.7, pp .851-860, July 2004. S. Boll, "Suppression of acoustic noise in speech using spectral subtraction," IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-27, no. 2, pp. 113-120, Apr. 1979. R. Martin, "Spectral subtraction based on minimum statistics," Proc. Of EUSPICO-94, pp.1182-1185, Sept. 1994 R. McAulay and M. Malpass, "Speech enhancement using a soft-decision noise suppression filter," IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-28, no. 2, pp. 137-145, Apr. 1980. N. Virag, "Single channel speech enhancement based on masking properties of the human auditory system," IEEE Trans.Speech and Audio Signal Processing, vol. 7, no. 2, pp. 126-137, March 1999. Y. Ephraim and D. Malah, "Speech enhancement using a minimum mean-square error log-spectral amplitude estimator," IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-33, no. 2, pp. 443- 445, Dec. 1985. M. Kato and A. Sugiyama, "A Wind-Noise Suppressor Based on Wind-Onset Detection and Spectral Gain Modification," Proc. Of IWAENC 2014, pp.145-149, Sept. 2014

  However, the technique described in the above document does not assume that the second signal, which is a noise component, suddenly increases. For this reason, when the second signal included in the mixed signal suddenly increases, the second signal cannot be accurately estimated. As a result, it cannot be correctly determined whether or not the first signal is included in the mixed signal. Further, the second signal cannot be sufficiently suppressed, and the quality of the output signal is lowered. The main object of the present invention is to provide a signal processing apparatus or the like that solves this problem.

  The signal processing apparatus according to an aspect of the present invention uses a mixed signal in which a first signal and a second signal are mixed, estimates a second signal mixed in the mixed signal, and provides an estimated second signal Means and first determination means for determining whether or not the estimated second signal is excessive using the mixed signal and the estimated second signal, wherein the estimating means includes the first The estimation process is initialized based on the determination result of the determination means.

  In another aspect of achieving the above object, a signal processing method according to an aspect of the present invention uses a mixed signal in which a first signal and a second signal are mixed by an information processing device to be mixed in the mixed signal. A second signal is estimated and provided as an estimated second signal. Using the mixed signal and the estimated second signal, it is determined whether or not the estimated second signal is too small. Based on the result of the determination To initialize the estimation process.

  In a further aspect of achieving the above object, a signal processing program according to an aspect of the present invention uses a mixed signal in which a first signal and a second signal are mixed, and a second signal mixed in the mixed signal. And a determination process for determining whether or not the estimated second signal is too small using the mixed signal and the estimated second signal. The estimation process is executed, and the process of estimating the second signal is initialized based on a result of the determination in the determination process.

  Furthermore, the present invention can also be realized by a computer-readable non-volatile recording medium in which such a signal processing program (computer program) is stored.

  The present invention can accurately estimate the second signal even in the mixed signal including the first signal and the second signal even when the magnitude of the second signal suddenly increases.

It is a block diagram which shows the structure of the signal processing apparatus 100 which concerns on the 1st thru | or 6th embodiment of this invention. It is a block diagram which shows the structure of the conversion part 101 which concerns on 1st Embodiment of this invention. It is a block diagram which shows the structure of the inverse transformation part 106 which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of the determination part 102 which concerns on 1st Embodiment of this invention. It is a block diagram which shows the structure of the suppression part 105 which concerns on 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the signal processing apparatus 100 which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of the determination part 202 which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the structure of the determination part 302 which concerns on the 3rd Embodiment of this invention. It is a block diagram which shows the structure of the determination part 402 which concerns on the 4th Embodiment of this invention. It is a block diagram which shows the structure of the determination part 502 which concerns on the 5th Embodiment of this invention. It is a block diagram which shows the structure of the determination part 602 which concerns on the 6th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus 700 which concerns on the 7th Embodiment of this invention. It is a block diagram which shows the structure of the 2nd determination part 707 which concerns on the 7th Embodiment of this invention. It is a block diagram which shows the structure of the signal processing apparatus 800 which concerns on the 8th Embodiment of this invention. It is a block diagram which shows the structure of the information processing apparatus which can execute the signal processing apparatus which concerns on each embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<First Embodiment>
FIG. 1 is a block diagram conceptually showing the structure of the signal processing apparatus 100 according to the first embodiment of the present invention. In the signal processing apparatus according to the second to sixth embodiments described later, only the configuration of the “determination unit” is different. Therefore, for convenience of explanation, reference is made to FIG. 1 again in these embodiments. To do. The signal processing apparatus 100 suppresses the second signal from the mixed signal (input signal) including the first signal (the desired audio signal to be noticed) and the second signal (noise). Is an apparatus that generates an output signal that emphasizes. The signal processing device 100 according to the present embodiment functions as a part of a device such as a digital camera, a notebook computer, or a mobile phone.

  The signal processing apparatus 100 according to the present embodiment includes a conversion unit 101, a determination unit 102, an estimation unit 103, an initialization unit 104, a suppression unit 105, and an inverse conversion unit 106. Hereinafter, each of these components will be described in detail.

(Conversion unit 101)
FIG. 2 is a block diagram conceptually showing the configuration of the conversion unit 101 according to this embodiment. As illustrated in FIG. 2, the conversion unit 101 includes a frame division unit 1011, a window function multiplication unit 1012, and a Fourier transform unit 1013.

  The mixed signal is input to the frame dividing unit 1011 as a sample value series. The frame dividing unit 1011 divides the mixed signal samples (sample value series) into K / 2 samples (K is an arbitrary even number of 2 or more, “/” represents division (the same applies hereinafter)), and thereby divides a plurality of samples. Construct (generate) a frame. The frame division unit 1011 inputs the mixed signal samples divided into frames to the window function multiplication unit 1012.

The window function multiplication unit 1012 multiplies the mixed signal sample input from the frame division unit 1011 by w (t) (t is a variable indicating time) that is a window function. The mixed signal x (t, n) (t = 0, 1, ..., K-1) of the nth frame (n is an arbitrary natural number) is multiplied by w (t) (windowed). The signal x ⁻ (t, n) (“ ⁻ ” is assumed to be described at the top of x) is expressed by a mathematical expression represented by Equation (1).

The window function multiplication unit 1012 may open the window in a state where a part of two consecutive frames is overlapped (overlapped). For example, when the overlap length is 50% of the frame length, the window function multiplication unit 1012 performs the signal x ⁻ (t indicated by Equation 2 for t = 0, 1,..., K / 2-1. , n) is output.

The window function multiplier 1012 uses a symmetric window function for a real signal. Further, the window function is designed so that the mixed signal and the output signal match except for calculation errors when the output from the conversion unit 101 is directly input to the inverse conversion unit 106. This means that “w ² (t) + w ² (t + K / 2) = 1”.

  Hereinafter, a case where two consecutive frames are opened with 50% of the frame length overlapped will be described as an example. As the window function w (t) (t = 0, 1,..., K−1), for example, a Hanning window represented by Equation 3 can be used.

  In addition to these, various functions such as a Hamming window and a triangular window are known, and these functions may be adopted. The window function multiplication unit 1012 inputs the result obtained by multiplying the mixed signal sample input from the frame division unit 1011 by the window function to the Fourier transform unit 1013.

The Fourier transform unit 1013 divides the signal into a plurality of different frequency components by performing Fourier transform on the signal input from the window function multiplication unit 1012. That is, the Fourier transform unit 1013 transforms the signal into a mixed signal spectrum X (k, n). The Fourier transform unit 1013 converts the mixed signal spectrum X (k, n) into a mixed signal phase spectrum argX (k, n) which is a phase component and a mixed signal amplitude spectrum | X (k, n) | which is an amplitude component. To separate. The Fourier transform unit 1013 inputs the mixed signal phase spectrum argX (k, n) to the inverse transform unit 106, converts the mixed signal amplitude spectrum | X (k, n) | into the determination unit 102, the estimation unit 103, and the suppression. Input to the unit 105. The Fourier transform unit 1013 may use a power spectrum | X (k, n) | ² corresponding to the square of the amplitude spectrum | X (k, n) |.

(Inverse conversion unit 106)
FIG. 3 is a block diagram conceptually showing the configuration of the inverse transform unit 106 according to this embodiment. As shown in FIG. 3, the inverse transform unit 106 includes an inverse Fourier transform unit 1061, a window function multiplication unit 1062, and a frame synthesis unit 1063.

  The inverse Fourier transform unit 1061 uses the output signal amplitude spectrum | Y (k, n) | input from the suppression unit 105 and the mixed signal phase spectrum argX (k, n) input from the conversion unit 101, As Equation 4 shows, an output signal spectrum Y (k, n) is obtained. J in Equation 4 is a code representing an imaginary number. In the present application, “||” represents an absolute value.

  The inverse Fourier transform unit 1061 performs inverse Fourier transform on the obtained output signal spectrum Y (k, n) to thereby obtain a time-domain sample value sequence y (t, n) in which one frame includes K samples. (t = 0, 1,..., K−1) is generated, and the generated sample value series y (t, n) is input to the window function multiplier 1062.

The window function multiplication unit 1062 performs multiplication of the sample value series y (t, n) input from the inverse Fourier transform unit 1061 and the window function w (t). The window function multiplier 1062 inputs the signal y ⁻ (t, n) represented by Equation 5 (“ ⁻ ” is described above y) to the frame synthesizer 1063.

The frame synthesis unit 1063 extracts and superimposes two adjacent frames output from the window function multiplication unit 1062 by K / 2 samples, and as shown in Equation 6, t = 0, 1,..., K / 2− The output signal y ^{^} (t, n) at 1 (" ^{^} " is assumed to be described above y) is obtained. The frame synthesizing unit 1063 outputs the obtained output signal y ^{^} (t, n) to the outside.

  Note that the conversion process performed by the conversion unit 101 and the inverse conversion unit 106 is not limited to Fourier transform (inverse Fourier transform). The transform unit 101 and the inverse transform unit 106 may perform other transform processes such as Hadamard transform, Haar transform, and wavelet transform instead of the Fourier transform. Since the signal processing apparatus 100 does not require multiplication by using the Haar transform, the amount of necessary calculation resources can be reduced. Further, since the signal processing apparatus 100 can use the wavelet transform so that the time resolution varies depending on the frequency, the suppression efficiency of the second signal can be improved.

  In addition, the signal processing device 100 illustrated in FIG. 1 includes the conversion unit 101 and the inverse conversion unit 106, but the signal processing device 100 may not include the conversion unit 101 and the inverse conversion unit 106. In this case, a mixed signal is directly input from the outside to the determination unit 102, the estimation unit 103, and the suppression unit 105. And the suppression part 105 outputs an output signal directly outside. In this case, the signal processing apparatus 100 can reduce the amount of necessary calculation resources.

(Estimator 103)
1 is based on the mixed signal amplitude spectrum input from the conversion unit 101 and the information indicating the initialization input from the initialization unit 104, the power of the second signal included in the mixed signal. Estimate the spectrum. The power spectrum of the second signal estimated by the estimation unit 103 is hereinafter referred to as “estimated second signal power spectrum”. In general, the estimation unit 103 estimates the second signal power spectrum based on information indicating the state of the mixed signal input up to now. As a method of estimating the second signal power spectrum by the estimation unit 103, for example, various estimation methods such as the methods described in Non-Patent Documents 1 to 4 can be used. In the methods described in Non-Patent Documents 1 to 4, the second signal power spectrum is estimated using the mixed signal power spectrum obtained by squaring the mixed signal amplitude spectrum.

  By the way, Non-Patent Document 1 shows a method in which the estimated second signal power spectrum is set to an average value of the mixed signal power spectrum when the mixed signal amplitude spectrum starts to be input (estimation processing is started). Has been. In this case, it is necessary to satisfy the condition that the first signal (target sound) is not included immediately after the estimation process is started.

  Non-Patent Document 3 discloses a method in which the estimated second signal power spectrum is an average value of mixed signal power spectra of frames in which no first signal exists. In this method, it is necessary to detect the presence of the first signal. The section where the first signal exists (first signal section) is detected by the power (magnitude) of the output signal, for example.

  When the signal processing apparatus 100 operates ideally, the output signal matches the first signal. In addition, the levels (magnitudes) of the first signal and the second signal usually do not change greatly between adjacent frames. From these things, the estimation part 103 determines whether it is a 1st signal area based on the past output signal level for 1 frame. The estimation unit 103 determines that the current frame is the first signal section when the magnitude of the output signal in the past for one frame is equal to or greater than the threshold. The second signal power spectrum can be estimated by averaging the mixed signal power spectrum in a frame that is not determined to be the first signal interval.

  Non-Patent Document 4 discloses a method for obtaining an estimated second signal power spectrum from the minimum value (minimum statistic) of the mixed signal power spectrum. In this method, the estimation unit 103 holds the minimum value of the mixed signal power spectrum within a fixed time, and estimates the second signal power spectrum based on the minimum value. Since the minimum value of the mixed signal power spectrum is generally similar to the spectrum shape of the second signal, it can be used as an estimated value of the second signal power spectrum. However, in this case, the minimum value is smaller than the original magnitude of the second signal. Therefore, the estimation unit 103 sets a value obtained by appropriately amplifying the minimum value as an estimated second signal power spectrum.

  In addition to the methods described in these non-patent documents, the estimation unit 103 may obtain an estimated second signal power spectrum using a median filter. Further, as described in Non-Patent Document 2, the estimation unit 103 may obtain the estimated second signal power spectrum by utilizing the property that the second signal varies slowly. In this method of following a fluctuating second signal, first, in order to obtain an estimated value of the ratio of the magnitude of the “first signal” to the “second signal”, the estimation unit 103 uses the mixed signal power spectrum as one frame. Divide by the estimated second signal power spectrum in minutes past. Next, the estimation unit 103 calculates a weight by using a function that takes a smaller value as the obtained quotient is larger, and weights the mixed signal power spectrum. And the estimation part 103 averages using the mixed signal power spectrum weighted by the same method in the past using several, and makes the average value an estimated 2nd signal power spectrum. In this case, since a small weight is given to the mixed signal power spectrum containing a large amount of the first signal component, the estimation unit 103 is hardly affected by the first signal, and the second signal power spectrum is not affected. Estimation can be performed continuously.

  The estimation unit 103 can also estimate the second signal after integrating a plurality of frequency components obtained in the conversion unit 101. In this case, the number of frequency components after integration is smaller than the number of frequency components before integration. More specifically, the estimation unit 103 obtains a common estimated second signal for the integrated frequency components obtained by integrating the frequency components, and each of the common estimated second signals belongs to the same integrated frequency component. It may be used in common for the frequency components. Thus, when the second signal is estimated after integrating a plurality of frequency components, the estimation unit 103 can reduce the amount of calculation required for the estimation process because the number of frequency components to be applied is reduced.

(Determination unit 102)
FIG. 4 is a block diagram conceptually showing the configuration of the determination unit 102 according to this embodiment. As illustrated in FIG. 4, the determination unit 102 includes a determination execution unit 1021.

The determination execution unit 1021 uses the mixed signal amplitude spectrum input from the conversion unit 101 and the past estimated second signal power spectrum input from the estimation unit 103 to determine whether the estimated second signal power spectrum satisfies the standard. It is determined whether or not (too small or not). This determination result is used when the estimation unit 103 obtains the estimated second signal power spectrum. Therefore, the estimated second signal power spectrum used for obtaining the determination result uses a result estimated in the past. In the present embodiment, as an example, determination is made using a past estimated second signal power spectrum | N (k, n-1) | ² that is a typical value as a past estimated second signal power spectrum. The case where it does is demonstrated.

The determination execution unit 1021 outputs the estimated second signal power spectrum | N (k, n-1) | ^{2 of the} past for one frame and the mixed signal power spectrum | X (k, n) | ^The threshold value calculated by the linear mapping function with ² as a variable is compared. In this embodiment, the threshold value is calculated by multiplying the mixed signal power spectrum | X (k, n) | ² by a constant multiple (that is, α1 (k, n) times) which is a typical linear mapping function. explain. α1 (k, n) is a real number of 0 or more. Determination execution unit 1021, one frame previous estimated second signal power spectrum | N (k, n-1 ) | 2 is α1 (k, n) | X (k, n) | is smaller than ^2, the estimated It is determined that the second signal power spectrum does not satisfy the criterion (underestimation).

One frame past estimated second signal power spectrum | N (k, n-1 ) | calculation method of the threshold value used for comparison with the ² mixed signal power spectrum | X (k, n) | 2 of the linear mapping function It is not limited to the method using. For example, a linear function such as α1 (k, n) | X (k, n) | ² + C (k, n) may be employed. In that case, if the values of α1 (k, n) and C (k, n) are reduced, the possibility that it is determined that the criterion is not satisfied increases, so the size of the estimated second signal does not satisfy the criterion. It is difficult to overlook that it is in an underestimated state. In addition, a higher-order polynomial function, nonlinear function, or the like can be used for the above-described threshold calculation.

The past estimated second signal power spectrum used by the determination execution unit 1021 is not limited to the past estimated second signal power spectrum | N (k, n-1) | ² for one frame. For example, the determination execution unit 1021 can use an average value for several past frames as the past estimated second signal power spectrum. Since the determination execution unit 1021 can suppress temporal variation of the estimated second signal power spectrum by averaging in this way, the determination accuracy can be improved. The same applies to the mixed signal power spectrum | X (k, n) | ² , and the determination execution unit 1021 can suppress the time variation of the mixed signal power spectrum by using the average value for the past several frames. Accuracy can be improved. As a method for obtaining the average value, in addition to the above-described moving average, the primary leak integration is also an effective method. Further, the determination execution unit 1021 can use a method of selecting the minimum value from the past several frames instead of the average.

In addition to confirming the relationship between the estimated second signal power spectrum and the mixed signal power spectrum described above, the determination execution unit 1021 can also make a determination based on each value in order to improve the determination accuracy. is there. For example, when the estimated second signal power spectrum | N (k, n−1) | ^{2 of} one frame past is smaller than a predetermined threshold, the determination execution unit 1021 determines that the criterion is not satisfied. Thereby, even when the determination execution unit 1021 cannot determine that the values of both the estimated second signal power spectrum and the mixed signal power spectrum are small by chance, the relative relationship alone does not satisfy the criterion. It can be judged correctly.

(Initialization unit 104)
The initialization unit 104 shown in FIG. 1 receives information indicating the state of the mixed signal in the estimation unit 103 when the determination unit 102 determines that the size of the estimated second signal power spectrum does not satisfy the standard. 103 performs processing (hereinafter referred to as “initialization processing”) to return (change) the state immediately after the start of the estimation processing. For example, when the estimation unit 103 performs the estimation process by the method described in Non-Patent Document 4 described above, the initialization unit 104 changes the minimum value described above to a preset value as the initialization process. Further, when the estimation unit 103 performs the estimation process by the method described in Non-Patent Document 2 described above, the initialization unit 104 performs the update using the past estimated value as the initialization process. Instead, control is performed so that the mixed signal power spectrum | X (k, n) | ² , which is a value obtained by squaring the input mixed signal amplitude spectrum, is used as the estimated value.

  As a method for this initialization process, a method that considers an estimated value immediately before the initialization is effective is also effective. For example, as an initialization process, the initialization unit 104 changes the set value for initialization and the estimated value just before the initialization to a mixed value instead of the value set in advance for initialization. In this case, the initialization unit 104 can avoid a sudden change in the estimated value immediately after the initialization, so that it is possible to avoid a deterioration in sound quality caused by a sudden change in the degree of suppression of the second signal. In addition, the estimation unit 103 stores the estimated values before initialization for a plurality of frames, and the initialization unit 104 performs initialization processing such as an average value or minimum value of the frames stored by the estimation unit 103. The statistic may be used as an initial value. Thereby, the initialization unit 104 can avoid a sudden change in the quality of the output signal of the signal processing apparatus 100. Note that the initialization unit 104 may be included in the estimation unit 103.

(Suppression unit 105)
FIG. 5 is a block diagram conceptually showing the configuration of the suppression unit 105 according to this embodiment. As illustrated in FIG. 5, the suppression unit 105 includes a gain calculation unit 1051 and a multiplication unit 1052.

  The gain calculation unit 1051 can obtain a gain G (k, n), which is a value used when suppressing the second signal, using an existing technique. For example, as described in Non-Patent Document 5, the gain calculation unit 1051 may obtain an optimum estimated value that minimizes the mean square error with the first signal as a gain. In this case, the gain calculation unit 1051 calculates the gain G (k, n) as shown in Equation 7.

  The gain calculator 1051 may also obtain the gain G (k, n) using the method described in Non-Patent Document 6. In this case, the gain calculation unit 1051 calculates the gain G (k, n) as shown in Equation 8. In Equation 8, α0, β0, γ1, and γ2 are all real numbers, and α0 and β0 satisfy α0> 1 and 0 <β0 <1, respectively.

  Alternatively, the gain calculation unit 1051 may calculate the gain G (k, n) using the method described in Non-Patent Document 1 or Non-Patent Document 7.

  The multiplier 1052 shown in FIG. 5 multiplies the gain G (k, n) obtained by the gain calculator 1051 by the mixed signal amplitude spectrum | X (k, n) | G (k, n) | X (k , n) | is obtained as an output signal amplitude spectrum. Multiplier 1052 inputs the obtained output signal amplitude spectrum to inverse converter 106.

  Next, the operation (processing) of the signal processing apparatus 100 according to the present embodiment will be described in detail with reference to the flowchart of FIG.

  The conversion unit 101 generates an amplitude spectrum and a phase spectrum of the mixed signal from the mixed signal including the first signal and the second signal (step S101). The estimation unit 103 estimates the second signal from the amplitude spectrum of the mixed signal generated by the conversion unit 101, and generates an estimated second signal as the estimation result (step S102). Based on the amplitude spectrum of the mixed signal and the estimated second signal, the determining unit 102 determines whether or not the magnitude of the estimated second signal satisfies the criterion (whether it is underestimated) (step S103).

  When the magnitude of the estimated second signal satisfies the criterion (Yes in step S104), the suppression unit 105 generates the amplitude spectrum of the output signal by suppressing the second signal using the estimated second signal ( Step S106). If the size of the estimated second signal does not satisfy the standard (No in step S104), the initialization unit 104 estimates the second signal based on information indicating the state of the mixed signal input to date in the estimation unit 103. An initialization process for changing to a state when starting the process to be performed or a state related to the state is performed (step S105), and the process proceeds to step S106.

  The inverse conversion unit 106 generates an output signal using the amplitude spectrum of the output signal generated by the suppression unit 105 and the phase spectrum of the mixed signal generated by the conversion unit 101, and outputs the generated output signal to the outside (Step S107), and the process returns to step S101.

  In the mixed signal including the first signal and the second signal, the signal processing apparatus 100 according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. it can. The reason is that the signal processing apparatus 100 determines whether or not the size of the estimated second signal satisfies the criterion. If the criterion does not satisfy the criterion, the signal processing apparatus 100 indicates the state of the mixed signal used for estimating the second signal. This is because the initialization process is performed to change the state to the state at the time of starting the estimation process or a state related to the state.

  Below, the effect implement | achieved by the signal processing apparatus 100 which concerns on this embodiment is demonstrated in detail.

  The second signal estimation technique used in various voice terminals such as cellular phones generally does not assume that the magnitude of the second signal, which is an undesired signal such as noise, suddenly increases. For this reason, when the magnitude | size of the 2nd signal contained in a mixed signal suddenly becomes large, there exists a problem that a 2nd signal cannot be estimated correctly but the quality of an output signal falls.

  For such a problem, in the signal processing device 100 according to the present embodiment, the estimation unit 103 is a mixed signal in which the second signal is input so far in the mixed signal including the first signal and the second signal. Estimation is performed based on information indicating the signal state, and an estimated second signal is generated as an estimation result. Based on the mixed signal and the estimated second signal, the determining unit 102 determines whether or not the magnitude of the estimated second signal satisfies a criterion (whether it is too small). When the determination unit 102 determines that the size of the estimated second signal does not satisfy the standard, the initialization unit 104 starts processing for estimating the second signal with information indicating the state of the mixed signal. An initialization process is performed to change the current state or a state related to that state. The suppression unit 105 generates an output signal by suppressing the second signal included in the mixed signal using the estimated second signal.

  That is, the signal processing apparatus 100 according to the present embodiment uses the second information indicating the state of the mixed signal used when estimating the second signal so that the second signal whose size suddenly increases can be followed. It returns (changes) to the state at the time of starting the process of estimating the signal or the state related to the state. Thereby, the signal processing apparatus 100 according to the present embodiment accurately estimates the second signal after the noise suppression processing even if the size of the second signal suddenly increases in the mixed signal. The output signal quality can be improved.

<Second Embodiment>
Next, a signal processing device 100A according to a second embodiment of the present invention will be described. As illustrated in FIG. 1, the signal processing device 100A according to the present embodiment includes a conversion unit 101, a determination unit 202, an estimation unit 103, an initialization unit 104, a suppression unit 105, and an inverse conversion unit 106. The signal processing apparatus 100A according to the present embodiment has the same configuration as the first embodiment except for the determination unit 202, and thus the description thereof is omitted.

  FIG. 7 is a block diagram conceptually showing the configuration of the determination unit 202 according to this embodiment. As illustrated in FIG. 7, the determination unit 102 includes a determination execution unit 2021 and a relative relationship calculation unit 2022.

Similar to the determination unit 102 according to the first embodiment, the determination unit 202 according to the present embodiment uses the estimated second signal power spectrum estimated in the past by the estimation unit 103 when performing the determination process. Therefore, in the present embodiment as well, when the determination unit 102 according to the first embodiment is described, the determination unit 202 determines that the estimated second signal power spectrum | N (k, n−1) of one frame in the past. ) | A case where determination is performed using ² will be described.

The relative relationship calculation unit 2022 includes the mixed signal power spectrum | X (k, n) | ² corresponding to the square of the mixed signal amplitude spectrum input from the conversion unit 101 and the estimated second signal input from the estimation unit 103. A value indicating a relative relationship with the power spectrum | N (k, n-1) | ² is calculated. Here, the relative relationship is not only a simple magnitude relationship but also a difference or ratio between the two. For example, when the relative relationship is represented by a magnitude relationship, the relative relationship calculation unit 2022 calculates the relative relationship D (k, n) as shown in Equation 9.

  In Equation 9, a and b are real numbers. When the relative relationship is indicated by a difference, the relative relationship calculation unit 2022 calculates the relative relationship D (k, n) as shown in Equation 10.

  The relative relationship calculation unit 2022 may also calculate the relative relationship D (k, n) as a logarithmic difference as shown in Equation 11. In this case, the value calculated by the relative relationship calculation unit 2022 corresponds to the ratio between the mixed signal power spectrum and the estimated second signal power spectrum.

The relative relationship calculation unit 2022 calculates a relative relationship after smoothing the mixed signal power spectrum and the estimated second signal power spectrum in order to avoid erroneous determination due to an instantaneous sudden change in the mixed signal power spectrum. Also good. Typical examples of the smoothing method include moving average and first-order leak integration. When the moving average in the time direction is adopted as the smoothing method, the relative relationship calculation unit 2022 calculates the smoothed mixed signal power spectrum | X (k, n) | ² _bar as shown in Equation 12. In Formula 12, M1 is an integer of 1 or more.

The relative relationship calculating section 2022, as a method of smoothing, in the case of adopting the primary leaky integration, smoothed mixed signal power spectrum | X (k, n) | a ² _bar, to as calculated shown in Formula 13. In Equation 13, α1 is a real number satisfying 0 <α1 <1.

  The relative relationship calculation unit 2022 may also smooth the mixed signal power spectrum and the estimated second signal power spectrum using a low-pass filter, a median filter, or an ε filter. The relative relationship calculation unit 2022 may calculate the relative relationship instead of calculating the relative relationship after smoothing, and may perform the smoothing after calculating the relative relationship. In this case, the relative relationship calculation unit 2022 can reduce the number of times of smoothing, thereby reducing the amount of calculation.

  The determination execution unit 2021 determines whether or not the estimated second signal satisfies the criterion using a value indicating the relative relationship calculated by the relative relationship calculation unit 2022. For example, as in the determination unit 102 according to the first embodiment, the determination execution unit 2021 determines that the estimated second signal does not satisfy the criterion when the estimated second signal power spectrum is smaller than the mixed signal power spectrum. To do. Therefore, for example, when the relative relationship calculated by the relative relationship calculation unit 2022 is indicated by a simple magnitude relationship as shown in Equation 9, the determination execution unit 2021 determines that the value indicating the relative relationship is “a”. It is determined that the estimated second signal power spectrum does not satisfy the criterion. The determination execution unit 2021 determines that the estimated second signal power spectrum is a reference when the relative relationship D (k, n) exceeds a predetermined threshold when the relative relationship is as shown in Equation 10 or Equation 11. Is determined not to be satisfied.

  In the mixed signal including the first signal and the second signal, the signal processing apparatus 100A according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. can do. The reason is as described for the first embodiment.

  Further, the determination unit 202 according to the present embodiment performs determination using the relative relationship between the mixed signal power spectrum and the estimated second signal power spectrum. That is, since the determination unit 202 according to the present embodiment refers to the state of the estimated second signal power spectrum, the determination accuracy is improved as compared with the first embodiment that performs determination based only on the mixed signal power spectrum. Thereby, the signal processing apparatus 100A according to the present embodiment can further improve the quality of the output signal.

<Third Embodiment>
Next, a signal processing device 100B according to a third embodiment of the present invention will be described. As shown in FIG. 1, the signal processing device 100B according to the present embodiment includes a conversion unit 101, a determination unit 302, an estimation unit 103, an initialization unit 104, a suppression unit 105, and an inverse conversion unit 106. The signal processing apparatus 100B according to the present embodiment is the same as the first embodiment with respect to the configuration excluding the determination unit 302, and thus the description thereof is omitted.

  FIG. 8 is a block diagram conceptually showing the structure of the determination unit 302 according to this embodiment. As illustrated in FIG. 8, the determination unit 302 includes a determination execution unit 3021, a relative relationship calculation unit 2022, and an existence probability calculation unit 3023. Since the operation of the relative relationship calculation unit 2022 is the same as that of the second embodiment, the description thereof is omitted.

  The existence probability calculation unit 3023 uses the mixed signal power spectrum corresponding to the square of the mixed signal amplitude spectrum input from the conversion unit 101, and the presence probability of the first signal (first signal existence probability: first probability). Is calculated. As a representative example showing the probability that the first signal exists in the mixed signal including the first signal and the second signal, a method of setting the variance of the mixed signal power spectrum as the emphasized signal existence probability will be described.

Assuming that the expected value of the mixed signal power spectrum | X (k, n) | ² is E [| X (k, n) | ² ], the existence probability calculation unit 3023 follows the definition of variance and the first signal existence probability p (k, n) is calculated as shown in Equation 14.

Expected values can be approximated by average values. Therefore, the existence probability calculation unit 3023 according to the present embodiment calculates an average value by a moving average or a primary leak integration used in the smoothing performed by the determination unit 202 according to the second embodiment. For example, when the moving average is used, the existence probability calculation unit 3023 calculates the expected value E [| X (k, n) | ² ] as shown in Equation 15. In Formula 15, M2 is an integer of 1 or more.

The existence probability calculation unit 3023 also calculates the expected value E [| X (k, n) | ² ] as shown in Equation 16 when using the primary leak integral. In Equation 16, α2 is a real number satisfying 0 <α2 <1.

  The existence probability calculation unit 3023 may calculate the first signal existence probability using a standard deviation that is the square root of the variance instead of the variance described above.

  Alternatively, the presence probability calculation unit 3023 can use “Speech Presence Likelihood” described in Non-Patent Document 8 as the first signal presence probability. In this case, the existence probability calculation unit 3023 calculates the first signal existence probability based on the mixed signal power spectrum for one frame. Thus, the existence probability calculation unit 3023 can obtain the first signal existence probability in a short time compared to a method using dispersion that requires a power spectrum of a plurality of frames for calculation.

  The determination execution unit 3021 uses the value indicating the relative relationship calculated by the relative relationship calculation unit 2022 and the first signal existence probability calculated by the existence probability calculation unit 3023 as a reference for the estimated second signal power spectrum. It is determined whether or not it is satisfied. The determination execution unit 3021 according to the present embodiment is different from the determination execution unit 2021 according to the second embodiment in the configuration in which determination is also performed using the first signal existence probability. When the relative relationship D (k, n) is larger than the threshold Th_D and the first signal existence probability p (k, n) is smaller than the threshold Th_p, the determination execution unit 3021 has an estimated second signal power spectrum. It is determined that the standard is not satisfied (underestimation).

  Further, the determination execution unit 3021 compares the value calculated using the linear mapping function having the relative relationship D (k, n) and the first signal existence probability p (k, n) as variables with a threshold value. You may determine by. As an example, a case where the determination execution unit 3021 uses a weighted sum that is a typical linear mapping function will be described. In this case, when the weight for the relative relationship D (k, n) is β1 (k, n) and the weight for the first signal existence probability p (k, n) is β2 (k, n), the determination execution unit 3021 , If the value indicated by “β1 (k, n) D (k, n) −β2 (k, n) p (k, n)” is larger than the threshold, the estimated second signal power spectrum does not satisfy the standard ( It is underestimated). Note that β1 (k, n) and β2 (k, n) are positive real numbers. The value indicated by “β1 (k, n) D (k, n) −β2 (k, n) p (k, n)” is larger as D (k, n) is larger and p (k, n) is larger. The smaller, the larger.

  Since the determination execution unit 3021 makes a determination based on a value calculated using a function having both the value indicating the relative relationship and the first signal existence probability as a variable, the value indicating the relative relationship and the first signal existence probability Compared to the case where the separately determined results are integrated and determined, it is difficult to overlook underestimation. The determination execution unit 3021 may use a function represented by another form such as a higher-order polynomial function or a nonlinear function instead of the above-described weighted sum.

  In the mixed signal including the first signal and the second signal, the signal processing device 100B according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. can do. The reason is as described for the first embodiment.

  Further, the determination unit 302 according to the present embodiment performs determination by using not only the relative relationship between the mixed signal power spectrum and the estimated second signal power spectrum but also the first signal existence probability. The determination execution unit 3021 refers to the first signal existence probability to clarify the relationship between the estimated second signal power spectrum and the mixed signal power spectrum. For example, if the first signal existence probability is high, the power (magnitude) of the first signal is large. Therefore, when the estimated second signal power spectrum is smaller than the mixed signal power spectrum, the determination unit 302 is underestimated. Judge that there is no. Accordingly, the determination unit 302 according to the present embodiment determines that the estimated second signal power spectrum is underestimated if the estimated second signal power spectrum is smaller than the mixed signal power spectrum, even when the first signal is large. More accurate determination can be performed than the unit 202. Thereby, the signal processing device 100B according to the present embodiment can further improve the quality of the output signal.

<Fourth Embodiment>
Next, a signal processing device 100C according to a fourth embodiment of the present invention will be described. As illustrated in FIG. 1, the signal processing device 100C according to the present embodiment includes a conversion unit 101, a determination unit 402, an estimation unit 103, an initialization unit 104, a suppression unit 105, and an inverse conversion unit 106. The signal processing apparatus 100C according to the present embodiment has the same configuration as that of the first embodiment except for the determination unit 402, and thus the description thereof is omitted.

  FIG. 9 is a block diagram conceptually showing the configuration of the determination unit 402 according to this embodiment. As illustrated in FIG. 9, the determination unit 402 includes a determination execution unit 3021, a relative relationship calculation unit 2022, and an existence probability calculation unit 4023. Since the operation of the relative relationship calculation unit 2022 is the same as that of the second embodiment, and the operation of the determination execution unit 3021 is the same as that of the third embodiment, the description thereof is omitted.

  The existence probability calculation unit 4023 calculates the first signal existence probability (second probability) using the value indicating the relative relationship input from the relative relationship calculation unit 2022. The existence probability calculation unit 4023 is different from the existence probability calculation unit 3023 according to the third embodiment in a configuration in which the first signal existence probability is calculated using a value indicating a relative relationship instead of the mixed signal power spectrum. The existence probability calculation unit 4023 is different from the existence probability calculation unit 3023 according to the third embodiment in terms of information used for calculation, but the calculation method such as a calculation formula uses the same method as the existence probability calculation unit 3023. It is possible. The existence probability calculation unit 4023 calculates the first signal existence probability p (k, n) as shown in Equation 17. In Equation 17, E [D (k, n)] is an expected value of the relative relationship D (k, n).

  The expected value E [D (k, n)] is approximated by moving average or first-order leak integration as in the third embodiment. For example, when the expected value E [D (k, n)] is approximated using a moving average, the existence probability calculating unit 4023 calculates the expected value E [D (k, n)] as shown in Equation 18. In Equation 18, M3 is an integer of 1 or more.

  The existence probability calculation unit 4023 calculates the expected value E [D (k, n)] as shown in Equation 19 when approximating the expected value E [D (k, n)] using, for example, primary leak integration. In Equation 19, α3 is a real number satisfying 0 <α3 <1.

  The existence probability calculation unit 4023 may calculate the first signal existence probability using a standard deviation that is a square root of the variance instead of the variance described above.

  The determination execution unit 3021 according to the present embodiment uses the value indicating the relative relationship calculated by the relative relationship calculation unit 2022 and the first signal existence probability calculated by the existence probability calculation unit 4023 to estimate the second signal Determine whether the power spectrum meets the criteria.

  In the mixed signal including the first signal and the second signal, the signal processing device 100C according to the present embodiment can accurately estimate the second signal even when the second signal suddenly increases. it can. The reason is as described for the first embodiment.

  In addition, the determination unit 402 according to the present embodiment calculates the first signal existence probability using the relative relationship between the mixed signal power spectrum and the estimated second signal power spectrum. Since the determination unit 402 according to the present embodiment calculates the first signal existence probability using not only the mixed signal power spectrum but also the estimated second signal power spectrum, the first signal existence probability is calculated based on the mixed signal power spectrum. The first signal existence probability can be calculated more accurately than the determination unit 302 according to the third embodiment. For example, the determination unit 402 can determine which of the first signal and the second signal has increased by referring to the estimated second signal power spectrum when the mixed signal power spectrum has increased. Therefore, the determination unit 402 according to the present embodiment can perform more accurate determination. Thereby, the signal processing device 100C according to the present embodiment can further improve the quality of the output signal.

<Fifth Embodiment>
Next, a signal processing device 100D according to a fifth embodiment of the present invention will be described. As illustrated in FIG. 1, the signal processing device 100D according to the present embodiment includes a conversion unit 101, a determination unit 502, an estimation unit 103, an initialization unit 104, a suppression unit 105, and an inverse conversion unit 106. The signal processing apparatus 100D according to the present embodiment is the same as the first embodiment with respect to the configuration excluding the determination unit 502, and thus the description thereof is omitted.

  FIG. 10 is a block diagram conceptually showing the structure of the determination unit 502 according to this embodiment. As illustrated in FIG. 10, the determination unit 502 includes a determination execution unit 3021, a relative relationship calculation unit 2022, and existence probability calculation units 5023 and 5024. Since the operation of the relative relationship calculation unit 2022 is the same as that of the second embodiment, and the operation of the determination execution unit 3021 is the same as that of the third embodiment, the description thereof is omitted.

  The existence probability calculation unit 5024 according to the present embodiment uses the mixed signal power spectrum, which is a value obtained by squaring the mixed signal amplitude spectrum input from the conversion unit 101, to generate a first first signal existence probability (first Probability). The existence probability calculation unit 5024 uses the same method as the existence probability calculation unit 3024 according to the third embodiment in calculating the first first signal existence probability. The existence probability calculation unit 5024 calculates the first first signal existence probability p1 (k, n) as shown in Equation 20.

In Equation 20, the expected value E [| X (k, n) | ² ] of the mixed signal power spectrum is approximated by moving average or first-order leak integration, as in the third embodiment.

  The existence probability calculation unit 5023 uses the value indicating the relative relationship input from the relative relationship calculation unit 2022 and the first first signal existence probability input from the existence probability calculation unit 5024 to generate the second first The signal existence probability (third probability) is calculated. The existence probability calculation unit 5023 according to the present embodiment is different from the existence probability calculation unit 4023 according to the fourth embodiment in the configuration in which the first first signal existence probability is used in addition to the value indicating the relative relationship.

  In the present embodiment as well, as in the second and third embodiments, the case where the variance of the mixed signal power spectrum is set as the emphasized signal existence probability will be described. The existence probability calculation unit 5023 calculates the second first signal existence probability p2 (k, n) as shown in Equation 21.

  In Equation 21, the expected value is approximated by an average value. The existence probability calculation unit 5023 uses the first first signal existence probability p1 (k, n) to calculate an average value by moving average or first-order leak integration. For example, when using a moving average, the existence probability calculation unit 5023 calculates an expected value E [D (k, n)] as shown in Equation 22. In Formula 22, M4 is an integer of 0 or more.

  The existence probability calculation unit 5023 calculates the expected value E [D (k, n)] as shown in Expression 23 when using, for example, a primary leak integral. In Equation 23, α4 is a real number satisfying 0 <α4 <1.

  The determination execution unit 3021 according to the present embodiment estimates using the value indicating the relative relationship calculated by the relative relationship calculation unit 2022 and the second first signal presence probability calculated by the presence probability calculation unit 5023. It is determined whether or not the second signal power spectrum satisfies the criterion.

  In the mixed signal including the first signal and the second signal, the signal processing device 100D according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. can do. The reason is as described for the first embodiment.

  In addition, the determination unit 502 according to the present embodiment weights the value indicating the relative relationship with the first first signal existence probability p1 (k, n), so that the probability that the first signal is included is determined. The low relative relationship is used preferentially in the expected value calculation. Thus, the determination unit 502 according to the present embodiment increases the expected value of the relative relationship under the influence of the first signal component in the first signal interval, and immediately after the first signal interval, the variance, that is, the first It is possible to avoid the problem that the probability of existence of one signal is erroneously increased.

  That is, when calculating the second first signal existence probability used for underestimation determination, the determination unit 502 according to the present embodiment also calculates the first first signal existence probability in addition to the value indicating the relative relationship. By using this, it is possible to improve the accuracy of the first signal existence probability. Therefore, the determination unit 502 according to the present embodiment can determine the first signal existence probability more accurately immediately after the first signal interval. Thereby, the signal processing device 100D according to the present embodiment can further improve the quality of the output signal.

<Sixth Embodiment>
Next, a signal processing device 100E according to a sixth embodiment of the present invention will be described. As illustrated in FIG. 1, the signal processing device 100E according to the present embodiment includes a conversion unit 101, a determination unit 602, an estimation unit 103, an initialization unit 104, a suppression unit 105, and an inverse conversion unit 106. Since the signal processing apparatus 100E according to the present embodiment is the same as the first embodiment except for the determination unit 602, the description thereof is omitted.

  FIG. 11 is a block diagram conceptually showing the structure of the determination unit 602 according to this embodiment. As illustrated in FIG. 11, the determination unit 602 includes a determination execution unit 3021, a relative relationship calculation unit 2022, existence probability calculation units 6023 and 6024, and a suppression unit 6025. Since the operation of the relative relationship calculation unit 2022 is the same as that of the second embodiment, and the operation of the determination execution unit 3021 is the same as that of the third embodiment, the description thereof is omitted.

  The suppression unit 6025 according to the present embodiment uses the mixed signal amplitude spectrum input from the conversion unit 101 and the estimated second signal power spectrum input from the estimation unit 103, and includes the second signal included in the mixed signal amplitude spectrum. Suppresses signal components. The suppression unit 6025 inputs the output signal amplitude spectrum as a result of the suppression to the existence probability calculation unit 6024. Since the suppression unit 6025 receives the mixed signal amplitude spectrum and the estimated second signal power spectrum as in the case of the suppression unit 105, the suppression unit 6025 suppresses the second signal component using the same method as the suppression unit 105. Can do. Therefore, the signal processing apparatus 100E according to the present embodiment may be configured to input the signal output from the suppression unit 105 to the existence probability calculation unit 6024. In this case, since the suppression processing performed by the signal processing device 100E according to the present embodiment is reduced, the amount of calculation performed by the signal processing device 100E can be reduced.

  The existence probability calculation unit 6024 according to the present embodiment uses the output signal power spectrum, which is a value obtained by squaring the output signal amplitude spectrum input from the suppression unit 6025, to generate a first first signal existence probability (fourth Probability). The existence probability calculation unit 6024 according to the present embodiment uses the same method as the existence probability calculation method, although the input signal is different compared to the existence probability calculation unit 5024 according to the fifth embodiment. be able to.

  The existence probability calculation unit 6023 according to the present embodiment uses the value indicating the relative relationship input from the relative relationship calculation unit 2022 and the first first signal existence probability input from the existence probability calculation unit 6024, A second first signal existence probability (fifth probability) is calculated. The existence probability calculation unit 6023 according to the present embodiment uses the same method as the existence probability calculation method, although the input signal is different compared to the existence probability calculation unit 5023 according to the fifth embodiment. be able to.

  The determination execution unit 3021 according to the present embodiment estimates using the value indicating the relative relationship calculated by the relative relationship calculation unit 2022 and the second first signal presence probability calculated by the existence probability calculation unit 6023. It is determined whether or not the second signal power spectrum satisfies the criterion.

  In the mixed signal including the first signal and the second signal, the signal processing device 100E according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. can do. The reason is as described for the first embodiment.

  In addition, the determination unit 602 according to the present embodiment calculates the first signal existence probability using the output signal power spectrum in which the component of the second signal is smaller than the mixed signal power spectrum in the existence probability calculation unit 6024. Therefore, the determination unit 602 can more accurately calculate the first signal existence probability by reducing the influence of the second signal, and perform an accurate determination. Thereby, the signal processing device 100E according to the present embodiment can further improve the quality of the output signal.

<Seventh Embodiment>
FIG. 12 is a block diagram conceptually showing the structure of a signal processing device 700 according to the seventh embodiment of the present invention. The signal processing device 700 estimates the second signal from the mixed signal including the first signal (the desired audio signal to be noticed) and the second signal (noise), so that the first signal exists. It is an apparatus for determining whether or not.

  The signal processing apparatus 700 according to the present embodiment includes a conversion unit 101, a determination unit 102, an estimation unit 103, an initialization unit 104, and a second determination unit 707. Since the signal processing apparatus 700 according to the present embodiment is the same as the first embodiment except for the configuration of the second determination unit 707, the description thereof is omitted.

  FIG. 13 is a block diagram conceptually showing the configuration of the second determination unit 707 according to this embodiment. The second determination unit 707 includes a relative relationship calculation unit 7071 and a determination execution unit 7072.

The relative relationship calculation unit 7071 includes the mixed signal power spectrum | X (k, n) | ² corresponding to the square of the mixed signal amplitude spectrum input from the conversion unit 101 and the estimated second signal input from the estimation unit 103. A value indicating a relative relationship with the power spectrum | N (k, n) | ² is calculated. Here, the relative relationship is not only a simple magnitude relationship but also a difference or ratio between the two. For example, when the relative relationship is represented by a magnitude relationship, the relative relationship calculation unit 7071 calculates the relative relationship D2 (n) as shown in Equation 24.

  In Equation 24, a2 and b2 are real numbers. When the relative relationship is indicated by a difference, the relative relationship calculation unit 7071 calculates the relative relationship D2 (n) as shown in Equation 25.

  The relative relationship calculation unit 7071 may also calculate the relative relationship D2 (n) as a logarithmic difference as shown in Equation 26. In this case, the value calculated by the relative relationship calculation unit 7071 corresponds to the ratio between the mixed signal power spectrum and the estimated second signal power spectrum.

  The determination execution unit 7072 determines whether or not the first signal exists, using the value indicating the relative relationship calculated by the relative relationship calculation unit 7071, and outputs the determination result. For example, the determination execution unit 7072 determines that the first signal exists when the mixed signal power spectrum is larger than the estimated second signal power spectrum. Therefore, for example, when the relative relationship calculated by the relative relationship calculation unit 7071 is indicated by the magnitude relationship as shown in Expression 24, the determination execution unit 7072 determines that the value indicating the relative relationship is “a2”. It is determined that one signal exists. The determination execution unit 7072 determines that the first signal is present when the relative relationship D (k, n) exceeds a predetermined threshold when the relative relationship is as shown in Equation 25 or Equation 26. .

  In the mixed signal including the first signal and the second signal, the signal processing device 700 according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. can do. The reason is as described for the first embodiment. And the signal processing apparatus 700 which concerns on this embodiment is provided with the 2nd determination part 707, Even if it is a case where the magnitude | size of a 2nd signal becomes large suddenly, it is determined whether a 1st signal exists. Accurate judgment can be made.

  Below, the effect implement | achieved by the signal processing apparatus 700 which concerns on this embodiment is demonstrated in detail.

  In general, the voice detection technology used in various voice terminals such as mobile phones does not assume that the magnitude of the second signal, which is a noise component, suddenly increases. For this reason, when the magnitude | size of the 2nd signal contained in a mixed signal becomes large suddenly, there exists a problem that a 2nd signal cannot be estimated correctly but a determination result is mistaken. As a result, there arises a problem that the bit rate of the voice encoding is erroneously increased or the voice recognition result is wrong.

  For such a problem, in the signal processing device 700 according to the present embodiment, the estimation unit 103 is a mixed signal in which the first signal and the second signal are included and the second signal is input up to now. Estimation is performed based on information indicating the signal state, and an estimated second signal is generated as an estimation result. Based on the mixed signal and the estimated second signal, the determining unit 102 determines whether or not the magnitude of the estimated second signal satisfies a criterion (whether it is too small). When the determination unit 102 determines that the size of the estimated second signal does not satisfy the standard, the initialization unit 104 starts processing for estimating the second signal with information indicating the state of the mixed signal. An initialization process is performed to change the current state or a state related to that state. Then, the second determination unit 707 determines whether or not the mixed signal includes the first signal using the estimated second signal.

  That is, the signal processing apparatus 700 according to the present embodiment uses the second information indicating the state of the mixed signal used when estimating the second signal so that the second signal whose magnitude has suddenly increased can be followed. It returns (changes) to the state at the time of starting the process of estimating the signal or the state related to the state. Thereby, the signal processing apparatus 700 according to the present embodiment accurately determines whether or not the first signal is included even in the case where the size of the second signal suddenly increases in the mixed signal. be able to.

<Eighth Embodiment>
FIG. 14 is a block diagram conceptually showing the structure of a signal processing device 800 according to the eighth embodiment of the present invention.

  The signal processing apparatus 800 according to the present embodiment includes a determination unit 802 and an estimation unit 803.

  The estimation unit 803 estimates the second signal mixed in the mixed signal using the mixed signal in which the first signal and the second signal are mixed (estimation processing), and provides the second signal as the estimated second signal.

  The determination unit 802 determines whether the estimated second signal is excessively small using the mixed signal and the estimated second signal.

  Then, the estimation unit 803 initializes the estimation process based on the determination result in the determination unit 802.

  In the mixed signal including the first signal and the second signal, the signal processing device 800 according to the present embodiment accurately estimates the second signal even when the magnitude of the second signal suddenly increases. can do. The reason is that the signal processing apparatus 800 determines whether or not the estimated second signal is too small, and initializes the estimation process based on the determination result.

<Hardware configuration example>
In each of the above-described embodiments, each unit in the signal processing device illustrated in FIGS. 1 to 5 and 7 to 14 can be realized by a dedicated HW (HardWare) (electronic circuit). 1 to 5 and 7 to 14, at least the following configuration can be regarded as a function (processing) unit (software module) of the software program.
Determination units 102, 202, 302, 402, 502, 602, and 802,
・ Estimators 103 and 803,
-Initialization unit 104,
Second determination unit 707
-Suppression unit 105.

  However, the division of each part shown in these drawings is a configuration for convenience of explanation, and various configurations can be assumed for mounting. An example of the hardware environment in this case will be described with reference to FIG.

  FIG. 15 is a diagram illustratively explaining the configuration of an information processing apparatus 900 (computer) that can execute the signal processing apparatus according to each embodiment of the present invention. That is, FIG. 15 shows a configuration of a computer (information processing apparatus) that can realize the signal processing apparatus shown in FIGS. 1 to 5 and FIGS. 7 to 14, and realizes each function in the above-described embodiment. Represents a possible hardware environment.

The information processing apparatus 900 illustrated in FIG. 15 includes the following as constituent elements.
CPU (Central_Processing_Unit) 901,
ROM (Read_Only_Memory) 902,
RAM (Random_Access_Memory) 903,
-Hard disk (storage device) 904,
A communication interface 905 with an external device,
・ Bus 906 (communication line),
A reader / writer 908 capable of reading and writing data stored in a recording medium 907 such as a CD-ROM (Compact_Disc_Read_Only_Memory)
An input / output interface 909;

  In other words, the information processing apparatus 900 including the above-described components is a general computer in which these configurations are connected via the bus 906. The information processing apparatus 900 may include a plurality of CPUs 901 or may include a CPU 901 configured by a multi-core.

  The present invention described by taking the above embodiment as an example supplies a computer program capable of realizing the following functions to the information processing apparatus 900 shown in FIG. The function is the above-described configuration in the block configuration diagrams (FIGS. 1 to 5 and FIGS. 7 to 14) referred to in the description of the embodiment, or the function of the flowchart (FIG. 6). The present invention is then achieved by reading the computer program to the CPU 901 of the hardware, interpreting it and executing it. The computer program supplied to the apparatus may be stored in a readable / writable volatile memory (RAM 903) or a nonvolatile storage device such as the ROM 902 or the hard disk 904. The input / output interface 909 may have an A (Analog) / D (Digital) conversion function and a D / A conversion function.

  In the above case, a general procedure can be adopted as a method for supplying the computer program into the hardware. The procedure includes, for example, a method of installing in the apparatus via various recording media 907 such as a CD-ROM, a method of downloading from the outside via a communication line such as the Internet, and the like. In such a case, it can be understood that the present invention is configured by a code constituting the computer program or a recording medium 907 in which the code is stored.

  The present invention has been described above using the above-described embodiment as an exemplary example. However, the present invention is not limited to the embodiment described above. That is, the present invention can apply various modes that can be understood by those skilled in the art within the scope of the present invention.

In addition, a part or all of each embodiment mentioned above can be described also as the following additional remarks. However, the present invention described by way of example with the above-described embodiments is not limited to the following. That is,
(Appendix 1)
Using a mixed signal in which the first signal and the second signal are mixed, estimating a second signal mixed in the mixed signal, and providing an estimated second signal;
First determination means for determining whether or not the estimated second signal is too small using the mixed signal and the estimated second signal;
With
The signal processing apparatus characterized in that the estimation means initializes estimation processing based on a result of determination by the first determination means.
(Appendix 2)
The first determination means determines whether the estimated second signal is too small by obtaining a relative relationship between the first signal and the second signal.
The signal processing apparatus according to appendix 1.
(Appendix 3)
The first determination means includes
Based on the mixed signal, calculate a first probability that the first signal exists in the mixed signal;
Determining whether the estimated second signal is too low based on the first probability;
The signal processing device according to attachment 2.
(Appendix 4)
The first determination means includes
Based on the relative relationship, calculate a second probability that the first signal exists in the mixed signal,
Based on the second probability, it is determined whether the estimated second signal is too small;
The signal processing device according to attachment 2.
(Appendix 5)
The first determination means includes
Based on the relative relationship and the first probability, calculate a third probability that the first signal exists in the mixed signal;
Based on the third probability, it is determined whether the estimated second signal is too small,
The signal processing device according to attachment 3.
(Appendix 6)
The first determination means includes
Using the mixed signal and the estimated second signal, generating a suppression signal by suppressing the second signal included in the mixed signal,
Based on the suppression signal, calculate a fourth probability that the first signal exists in the mixed signal,
Based on the calculated fourth probability and the relative relationship, a fifth probability that the first signal exists in the mixed signal is calculated,
Determining whether the estimated second signal is too low based on the fifth probability;
The signal processing device according to attachment 2.
(Appendix 7)
Further comprising suppression means for suppressing the second signal included in the mixed signal using the estimated second signal;
The signal processing device according to any one of appendices 1 to 6.
(Appendix 8)
Using the estimated second signal, further comprising second determination means for determining whether or not the first signal is included in the mixed signal;
The signal processing device according to any one of appendices 1 to 6.
(Appendix 9)
Depending on the information processing device,
Using the mixed signal in which the first signal and the second signal are mixed, the second signal mixed in the mixed signal is estimated and provided as the estimated second signal,
Using the mixed signal and the estimated second signal, determine whether the estimated second signal is too small,
A signal processing method for initializing an estimation process based on a result of the determination.
(Appendix 10)
Using a mixed signal in which the first signal and the second signal are mixed, estimating a second signal mixed in the mixed signal, and providing an estimated second signal;
A determination process for determining whether or not the estimated second signal is too small using the mixed signal and the estimated second signal;
To the computer,
The signal processing program characterized in that the estimation process initializes the estimation process based on a result of determination by the first determination unit.
(Appendix 11)
The second signal in the mixed signal including the first signal and the second signal is estimated based on the information indicating the state of the mixed signal that has been input so far, and an estimated second signal that represents the estimation result is generated. An estimation means to
First determination means for determining whether the magnitude of the estimated second signal satisfies a criterion based on the mixed signal and the estimated second signal;
When the first determination unit determines that the size of the estimated second signal does not satisfy the criterion, the estimation unit estimates the second signal with information indicating the state of the mixed signal Initialization means for performing initialization processing to change to a state at the time of starting or a state related to the state;
A signal processing apparatus comprising:
(Appendix 12)
The first determination means obtains a value indicating a relative relationship between the magnitude of the first signal and the magnitude of the second signal, thereby determining whether the magnitude of the estimated second signal satisfies the criterion. judge,
The signal processing apparatus according to appendix 11.
(Appendix 13)
The first determination means includes
Based on the mixed signal, calculate a first probability that the first signal exists in the mixed signal;
Determining whether the magnitude of the estimated second signal satisfies the criterion when the first probability is less than or equal to a threshold;
The signal processing device according to attachment 12.
(Appendix 14)
The first determination means includes
Based on the value indicating the relative relationship, a second probability that the first signal exists in the mixed signal is calculated,
If the second probability is less than or equal to a threshold, determine whether the magnitude of the estimated second signal satisfies the criterion;
The signal processing device according to attachment 12.
(Appendix 15)
The first determination means includes
Based on the value indicating the relative relationship and the first probability, a third probability that the first signal exists in the mixed signal is calculated;
If the third probability is less than or equal to a threshold, determine whether the magnitude of the estimated second signal satisfies the criterion;
The signal processing device according to attachment 13.
(Appendix 16)
The first determination means includes
Using the mixed signal and the estimated second signal to generate a suppression result signal by suppressing the second signal included in the mixed signal,
Based on the generated suppression result signal, calculate a fourth probability that the first signal exists in the mixed signal,
Based on the calculated fourth probability and the value indicating the relative relationship, a fifth probability that the first signal exists in the mixed signal is calculated,
If the fifth probability is less than or equal to a threshold, determine whether the magnitude of the estimated second signal satisfies the criterion;
The signal processing device according to attachment 12.
(Appendix 17)
Further comprising suppression means for suppressing the second signal included in the mixed signal using the estimated second signal;
The signal processing device according to any one of appendices 11 to 16.
(Appendix 18)
Using the estimated second signal, further comprising second determination means for determining whether or not the first signal is included in the mixed signal;
The signal processing device according to any one of appendices 11 to 16, characterized by:
(Appendix 19)
The determination means inputs, to a predetermined mapping function, a value indicating an average value or a minimum value of the estimated second signal generated in a predetermined period until now, indicating the current mixed signal size. It is determined whether or not the value obtained by
The signal processing device according to any one of appendices 11 to 18.
(Appendix 20)
The initialization unit is based on at least one of a setting value for initialization and a value indicating a state of the estimated second signal generated by the estimation unit in a predetermined period until initialization is performed. Initializing information indicating the state of the mixed signal;
The signal processing device according to any one of appendices 11 to 19.
(Appendix 21)
Depending on the information processing device,
The second signal in the mixed signal including the first signal and the second signal is estimated based on the information indicating the state of the mixed signal that has been input so far, and an estimated second signal that represents the estimation result is generated. And
Based on the mixed signal and the estimated second signal, determine whether the magnitude of the estimated second signal satisfies a criterion,
When it is determined that the size of the estimated second signal does not satisfy the criterion, the information indicating the state of the mixed signal is related to the state when the process of estimating the second signal is started or the state. Perform initialization processing to change to the state,
Signal processing method.
(Appendix 22)
The second signal in the mixed signal including the first signal and the second signal is estimated based on the information indicating the state of the mixed signal that has been input so far, and an estimated second signal that represents the estimation result is generated. An estimation process to
Based on the mixed signal and the estimated second signal, a determination process for determining whether the magnitude of the estimated second signal satisfies a criterion;
When the determination process determines that the size of the estimated second signal does not satisfy the criterion, the estimation process starts the process of estimating the second signal using information indicating the state of the mixed signal An initialization process to change to a state of time or a state related to that state,
Signal processing program for causing a computer to execute.

DESCRIPTION OF SYMBOLS 100 Signal processing apparatus 101 Conversion part 1011 Frame division part 1012 Window function multiplication part 1013 Fourier transform part 102 Determination part 1021 Determination execution part 103 Estimation part 104 Initialization part 105 Suppression part 1051 Gain calculation part 1052 Multiplication part 106 Inverse conversion part 1061 Inverse Fourier transform unit 1062 Window function multiplication unit 1063 Frame composition unit 202 Judgment unit 2021 Judgment execution unit 2022 Relative relation calculation unit 302 Judgment unit 3021 Judgment execution unit 3023 Presence probability calculation unit 402 Judgment unit 4023 Presence probability calculation unit 502 Judgment unit 5023 Presence probability Calculation unit 5024 Existence probability calculation unit 602 Determination unit 6023 Existence probability calculation unit 6024 Existence probability calculation unit 6025 Suppression unit 700 Signal processing device 707 Second determination unit 800 Signal processing device 802 Determination unit 803 Estimation unit 9 0 information processing device 901 CPU
902 ROM
903 RAM
904 Hard disk (storage device)
905 Communication interface 906 Bus 907 Recording medium 908 Reader / writer 909 Input / output interface

Claims (10)

Using a mixed signal in which the first signal and the second signal are mixed, estimating a second signal mixed in the mixed signal, and providing an estimated second signal;
First determination means for determining whether or not the estimated second signal is too small using the mixed signal and the estimated second signal;
With
The signal processing apparatus characterized in that the estimation means initializes estimation processing based on a result of determination by the first determination means. The first determination means determines whether the estimated second signal is too small by obtaining a relative relationship between the first signal and the second signal.
The signal processing apparatus according to claim 1. The first determination means includes
Based on the mixed signal, calculate a first probability that the first signal exists in the mixed signal;
Determining whether the estimated second signal is too low based on the first probability;
The signal processing apparatus according to claim 2. The first determination means includes
Based on the relative relationship, calculate a second probability that the first signal exists in the mixed signal,
Based on the second probability, it is determined whether the estimated second signal is too small;
The signal processing apparatus according to claim 2. The first determination means includes
Based on the relative relationship and the first probability, calculate a third probability that the first signal exists in the mixed signal;
Based on the third probability, it is determined whether the estimated second signal is too small,
The signal processing apparatus according to claim 3. The first determination means includes
Using the mixed signal and the estimated second signal, generating a suppression signal by suppressing the second signal included in the mixed signal,
Based on the suppression signal, calculate a fourth probability that the first signal exists in the mixed signal,
Based on the calculated fourth probability and the relative relationship, a fifth probability that the first signal exists in the mixed signal is calculated,
Determining whether the estimated second signal is too low based on the fifth probability;
The signal processing apparatus according to claim 2. Further comprising suppression means for suppressing the second signal included in the mixed signal using the estimated second signal;
The signal processing apparatus according to claim 1. Using the estimated second signal, further comprising second determination means for determining whether or not the first signal is included in the mixed signal;
The signal processing apparatus according to claim 1. Depending on the information processing device,
Using the mixed signal in which the first signal and the second signal are mixed, the second signal mixed in the mixed signal is estimated and provided as the estimated second signal,
Using the mixed signal and the estimated second signal, determine whether the estimated second signal is too small,
A signal processing method for initializing an estimation process based on a result of the determination. Using a mixed signal in which the first signal and the second signal are mixed, estimating a second signal mixed in the mixed signal, and providing an estimated second signal;
A determination process for determining whether or not the estimated second signal is too small using the mixed signal and the estimated second signal;
To the computer,
The estimation process initializes a process of estimating the second signal based on a result of the determination in the determination process.

Images