JP2010220087A - Sound processing apparatus and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To suppress feedback sound while suppressing the generation of musical noise caused by spectrum subtraction. <P>SOLUTION: A feedback sound estimation unit 24 generates an estimated feedback sound component n2 (t) obtained by estimating a feedback sound component n1 (t) reaching a sound collection apparatus 14 from a sound emission apparatus 12. A feedback sound suppression unit 26 suppresses the estimated feedback sound component n2 (t) from a sound signal w1 (t) generated by the sound collection apparatus 14. A kurtosis calculation unit 52A calculates kurtosis KA (m) in a frequency distribution of intensity of an estimated residual sound component e2 (t) obtained by estimating a residual component within a feedback sound component n1 (t) after processing by the feedback sound suppression unit 26. A coefficient setter 54 sets a subtraction coefficient α(m) in accordance with the kurtosis KA (m) calculated by the kurtosis calculation unit 52A. A spectrum subtraction unit 42 adjusts a spectrum E2 (m, f) of the estimated residual sound component e2 (t) in accordance with a subtraction coefficient α(m) to be subtracted from a spectrum W2 of a sound signal w2 (t) after processing by the feedback sound suppression unit 26. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a technique for suppressing the influence of sound (hereinafter referred to as “return sound”) that arrives at a sound collecting device from the sound emitting device in an acoustic system including the sound emitting device and the sound collecting device.

  Conventionally, a technique for suppressing the influence (echo or howling) of feedback sound coming from a sound emitting device to a sound collecting device has been proposed. For example, in Patent Document 1, a component simulating feedback sound (hereinafter referred to as “estimated feedback sound component”) is subtracted from the input speech in the time domain, and then the spectrum of the estimated feedback sound component is subtracted from the subtracted spectrum ( A technique for spectral subtraction) is disclosed.

JP 2004-56453 A

  However, in the technique of suppressing the estimated feedback sound component in the frequency domain as in Patent Document 1, components scattered in the time axis and the frequency axis after spectrum subtraction are received as artificial and annoying musical noise. There is a problem that it can be perceived by the listener. In view of the above circumstances, an object of the present invention is to suppress feedback sound while suppressing the generation of musical noise.

  In order to solve the above problems, the acoustic processing device according to the first aspect of the present invention includes a feedback sound estimation unit that generates an estimated feedback sound component that estimates a feedback sound that arrives at a sound collection device from a sound emitting device. , Feedback sound suppression means for suppressing the estimated feedback sound component from the acoustic signal generated by the sound collection device (for example, the acoustic signal w1 (t) and the acoustic signal w2 (t)), and processing by the feedback sound suppression means of the feedback sound Kurtosis calculation means (for example, kurtosis calculation unit 52A) for calculating kurtosis (for example, kurtosis KA (m)) in the frequency distribution of the intensity of the estimated residual sound component obtained by estimating the remaining component, and kurtosis calculation means Coefficient setting means for setting a subtraction coefficient according to the calculated kurtosis, and an acoustic signal after processing by the feedback sound suppression means by adjusting the spectrum of the estimated residual sound component according to the subtraction coefficient (for example, the acoustic signal w2 (t) Spectral subtraction means to subtract from the spectrum of) Comprising. In addition, the specific example of the above aspect is later mentioned as 1st Embodiment, 2nd Embodiment, and 5th Embodiment.

  In the acoustic processing device according to the first aspect, since the spectrum of the estimated residual sound component is subtracted from the spectrum of the acoustic signal processed by the feedback sound suppression means, only the suppression by the feedback sound suppression means is executed. In comparison, there is an advantage that the feedback sound component can be effectively suppressed from the acoustic signal generated by the sound collection device. In addition, since the subtraction coefficient set according to the kurtosis in the frequency distribution of the intensity of the estimated residual sound component is applied to the adjustment of the spectrum of the estimated residual sound component, the subtraction coefficient does not depend on the kurtosis (for example, the subtraction coefficient Can be suppressed as compared with a configuration in which is fixed to a predetermined value). The “kurtosis in the frequency distribution of the estimated residual sound component intensity” is the intensity in the frequency distribution of a plurality of signal values (intensities) arranged in time series in the time domain signal representing the waveform of the estimated residual sound component. And a kurtosis in a frequency distribution of a plurality of intensities in the spectrum of the estimated residual sound component.

  In a preferred aspect of the present invention, the spectrum subtracting means multiplies the spectrum of the estimated residual sound component by a subtraction coefficient and subtracts it from the spectrum of the acoustic signal processed by the feedback sound suppressing means. In the above aspect, musical noise resulting from spectral subtraction can be effectively suppressed by setting the subtraction coefficient to a larger value as the kurtosis increases.

  In a preferred aspect of the present invention, the kurtosis calculation means calculates the kurtosis of the estimated residual sound component and the kurtosis in the frequency distribution of the intensity of the acoustic signal generated by the sound collection device, and the coefficient setting means When the kurtosis of the estimated residual sound component approximates the kurtosis of the acoustic signal, a subtraction coefficient is set according to at least one of the kurtosis of the estimated residual sound component and the kurtosis of the acoustic signal, and the estimated residual sound component If the kurtosis of the sound does not approximate the kurtosis of the acoustic signal, the update of the subtraction coefficient is stopped. In the above aspect, the update of the subtraction coefficient stops when the kurtosis of the estimated residual sound component and the kurtosis of the acoustic signal are not approximate (that is, when the acoustic signal includes the target sound component). Not affected by the target sound component. Therefore, there is an advantage that musical noise can be appropriately suppressed. A specific example of the above aspect will be described later as a fifth embodiment. If the kurtosis of the estimated residual sound component and the kurtosis of the acoustic signal are not approximated, the coefficient setting means stops updating the subtraction coefficient and sets the subtraction coefficient to a predetermined value (for example, reduces the degree of spectrum subtraction). A configuration in which the value is initialized to a numerical value) is also preferable. According to the above configuration, it is possible to prevent the spectral subtraction for the section including the target sound component from being excessive due to the influence of the section not including the target sound component.

  The acoustic processing device according to the second aspect of the present invention includes a feedback sound estimation unit that generates an estimated feedback sound component that estimates a feedback sound that arrives at a sound collection device from a sound emitting device, and a sound that is generated by the sound collection device. Feedback sound suppression means for suppressing the estimated feedback sound component from the signal (for example, the acoustic signal w1 (t) or the acoustic signal w2 (t)), and the acoustic signal generated by the sound collecting device (for example, the acoustic signals w1 (t) to w3) (t)) kurtosis calculation means for calculating the kurtosis (for example, kurtosis KB (m) or kurtosis KC (m)) in the intensity distribution of intensity, and subtraction according to the kurtosis calculated by the kurtosis calculation means A coefficient setting means for setting a coefficient, and an acoustic signal after the processing by the feedback sound suppression means by adjusting the spectrum of the estimated residual sound component estimated from the feedback sound after the processing by the feedback sound suppression means according to the subtraction coefficient Spectral subtraction means for subtracting from the spectrum of the signal (eg acoustic signal w2 (t)) It comprises. Specific examples of the above aspects will be described later as third to seventh embodiments.

  In the acoustic processing device according to the second aspect, since the spectrum of the estimated residual sound component is subtracted from the spectrum of the acoustic signal processed by the feedback sound suppression unit, only the suppression by the feedback sound suppression unit is executed. In comparison, there is an advantage that the feedback sound component can be effectively suppressed from the acoustic signal generated by the sound collection device. Further, since the subtraction coefficient set according to the kurtosis in the frequency distribution of the intensity of the acoustic signal is applied to the adjustment of the spectrum of the estimated residual sound component, the subtraction coefficient does not depend on the kurtosis (for example, the subtraction coefficient is set to a predetermined value). Compared with a configuration fixed to a value), it is possible to suppress musical noise caused by spectral subtraction. Note that “the kurtosis in the frequency distribution of the intensity of the acoustic signal” means the intensity distribution in the frequency distribution of a plurality of signal values (intensities) arranged in time series in the acoustic signal and the frequency distribution of the plurality of intensities in the spectrum of the acoustic signal. It is a concept that includes kurtosis.

  In a specific example of the second mode (for example, the third embodiment, the fourth embodiment, the fifth embodiment, and the seventh embodiment), the kurtosis calculation means is the kurtosis of the acoustic signal before being processed by the spectrum subtraction means. The spectrum subtracting means multiplies the spectrum of the estimated residual sound component by the subtraction coefficient and subtracts it from the spectrum of the acoustic signal processed by the feedback sound suppressing means. In the above configuration, for example, musical noise caused by spectrum subtraction can be effectively suppressed by setting the subtraction coefficient to a larger numerical value as the kurtosis increases. Note that the “acoustic signal before processing by the spectrum subtracting means” is not limited to the acoustic signal immediately before processing by the spectrum subtracting means, and the acoustic signal at any stage from the generation by the sound collecting device to before the processing by the spectrum subtracting means. A signal (for example, an acoustic signal before processing by the feedback sound suppression unit or an acoustic signal after processing by the feedback sound suppression unit).

  In another specific example of the second mode (for example, the sixth embodiment or the seventh embodiment), the kurtosis calculating unit calculates the kurtosis of the acoustic signal processed by the spectrum subtracting unit, and the spectrum subtracting unit Then, the spectrum of the estimated residual sound component is multiplied by a subtraction coefficient and subtracted from the spectrum of the acoustic signal after processing by the feedback sound suppression means. In the above configuration, for example, musical noise caused by spectrum subtraction can be effectively suppressed by setting the subtraction coefficient to a smaller value as the kurtosis increases. The “acoustic signal after processing by the spectrum subtracting means” is not limited to the acoustic signal immediately after the processing by the spectral subtracting means, and includes, for example, an acoustic signal that has been subjected to other processing after being processed by the spectral subtracting means. To do.

  An acoustic processing device according to a preferred aspect of the present invention (first aspect and second aspect) includes a target sound determination unit that determines the presence or absence of a target sound component for an acoustic signal generated by a sound collection device, The kurtosis calculating means calculates the kurtosis when the target sound determining means determines that the target sound component does not exist, and stops calculating the kurtosis when the target sound determining means determines that the target sound component exists. . In the above aspect, since the influence of the target sound component on the subtraction coefficient is reduced (ideally excluded), there is an advantage that musical noise caused by spectral subtraction can be effectively suppressed. Specific examples of the above aspects will be described later as, for example, the second embodiment and the fourth embodiment.

  The acoustic processing device according to each of the above aspects is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to processing of an acoustic signal, or a general-purpose calculation such as a CPU (Central Processing Unit). This is also realized by cooperation between the processing device and the program. The program according to the first aspect of the present invention includes a feedback sound estimation process for generating an estimated feedback sound component that estimates a feedback sound that arrives at a sound collection device from a sound emitting device, and an acoustic signal generated by the sound collection device. A feedback sound suppression process for suppressing the estimated feedback sound component, and a kurtosis calculation process for calculating the kurtosis in the frequency distribution of the intensity of the estimated residual sound component estimated after the feedback sound suppression process of the feedback sound is performed. , The coefficient setting process for setting the subtraction coefficient according to the kurtosis calculated by the kurtosis calculation process, and the spectrum of the acoustic signal after executing the feedback sound suppression process by adjusting the spectrum of the estimated residual sound component according to the subtraction coefficient And causing the computer to execute spectral subtraction processing for subtracting from. According to the above program, the same operation and effect as the sound processing apparatus according to the first aspect of the present invention are exhibited.

  Further, the program according to the second aspect of the present invention includes a feedback sound estimation process for generating an estimated feedback sound component obtained by estimating a feedback sound arriving at a sound collection device from a sound emitting device, and a sound generated by the sound collection device. Feedback sound suppression processing to suppress the estimated feedback sound component from the signal, kurtosis calculation processing to calculate the kurtosis in the frequency distribution of the intensity of the sound signal generated by the sound collection device, and the kurtosis calculated by the kurtosis calculation processing Of the feedback sound suppression processing by adjusting the spectrum of the estimated residual sound component that estimates the component remaining after execution of the feedback sound suppression processing of the feedback sound according to the subtraction coefficient. The computer executes a spectrum subtraction process for subtracting from the spectrum of the acoustic signal after the execution. According to the above program, the same operation and effect as the sound processing apparatus according to the second aspect of the present invention are exhibited.

  The program according to each of the above aspects is provided to the user in a form stored in a computer-readable recording medium and installed in the computer, and is also provided from the server device in the form of distribution via a communication network. Installed on the computer.

1 is a block diagram of a sound processing apparatus according to a first embodiment of the present invention. It is a conceptual diagram for demonstrating the musical noise resulting from spectrum subtraction. It is a block diagram of the sound processing apparatus which concerns on 2nd Embodiment of this invention. It is a block diagram of the sound processing apparatus which concerns on 3rd Embodiment of this invention. It is a block diagram of the sound processing apparatus which concerns on 4th Embodiment of this invention. It is a block diagram of the sound processing apparatus which concerns on 5th Embodiment of this invention. It is a block diagram of the sound processing apparatus which concerns on 6th Embodiment of this invention. It is a block diagram of the sound processing apparatus which concerns on 7th Embodiment of this invention.

<A: First Embodiment>
FIG. 1 is a block diagram of a sound processing apparatus 100A according to the first embodiment of the present invention. The acoustic processing device 100A is preferably used for a telephone (typically a hands-free telephone) that exchanges acoustic signals with other communication devices via a communication network.

  A sound emitting device (for example, a speaker) 12 and a sound collecting device 14 are connected to the sound processing apparatus 100A. A sound signal (far end signal) v (t) in the time domain (time t) received from the communication network is supplied to the sound emitting device 12 via the sound processing device 100A. The sound emitting device 12 emits a sound wave corresponding to the acoustic signal v (t). The sound collection device 14 generates a time-domain sound signal w1 (t) corresponding to the surrounding sound and outputs it to the sound processing apparatus 100A. It should be noted that a D / A converter that converts the acoustic signal v (t) into an analog signal and supplies it to the sound emitting device 12 and an A / D converter that converts the acoustic signal w1 (t) into a digital signal are shown for convenience. Are omitted.

The target sound and the return sound arrive at the sound collecting device 14. The target sound is the sound generated by the user of the sound processing apparatus 100A (that is, the sound that is the original purpose of communication), and the return sound is directly or indirectly (that is, after reflection) from the sound emitting device 12. The sound (typically an echo sound) that reaches the sound collection device 14. Therefore, the acoustic signal w1 (t) is expressed by the following equation (1): the target sound component s (t) corresponding to the target sound and the feedback sound component n1 (t) corresponding to the feedback sound. It corresponds to addition. The feedback sound component n1 (t) is a component obtained by adding a transfer function h1 corresponding to a sound wave path from the sound emitting device 12 to the sound collecting device 14 to the acoustic signal v (t) (n1 (t) = h1 · v ( t)).
w1 (t) = s (t) + n1 (t) (1)

  The acoustic processing apparatus 100A is an echo suppression apparatus that generates an acoustic signal (near-end signal) z (t) by suppressing the feedback sound component n1 (t) from the acoustic signal w1 (t) generated by the sound collection device 14. Echo canceller). The acoustic signal z (t) generated by the acoustic processing device 100A is transmitted to another communication device via the communication network. Each element of FIG. 1 constituting the sound processing apparatus 100A is realized by a general-purpose computer (CPU) that executes a program or a dedicated electronic circuit (DSP).

  The target sound determination unit 22 of FIG. 1 is an element (VAD: voice activity detection) that determines the presence or absence of the target sound component s (t) in the acoustic signal w1 (t) and distinguishes the voiced and silent sections. . The voiced section is a section in which the target sound component s (t) is included in the acoustic signal w1 (t) (regardless of the presence or absence of the feedback sound component n1 (t)), and the silent section is the acoustic signal w1 (t). The target sound component s (t) is not included (or the intensity is sufficiently low). A known technique is arbitrarily adopted to distinguish between the sounded section and the silent section.

  The feedback sound estimation unit 24 generates an estimated feedback sound component n2 (t) obtained by estimating the feedback sound component n1 (t). For example, an AEC (acoustic echo canceller) using an adaptive filter is preferably used as the feedback sound estimation unit 24. The feedback sound suppression unit 26 generates the acoustic signal w2 (t) by suppressing the estimated feedback sound component n2 (t) generated by the feedback sound estimation unit 24 from the acoustic signal w1 (t). For example, as shown in FIG. 1, a subtractor that subtracts the estimated feedback sound component n2 (t) from the acoustic signal w1 (t) is used as the feedback sound suppression unit 26 (w2 (t) = w1 (t) − n2 (t)).

  The feedback sound estimation unit 24 generates the transfer function h2 by estimating the transfer function h1 so that the intensity of the acoustic signal w2 (t) is minimized, and multiplies the acoustic signal v (t) by the transfer function h2. To generate an estimated feedback sound component n2 (t) (n2 (t) = h2 · v (t)). The feedback sound estimation unit 24 sequentially calculates and calculates the estimated feedback sound component n2 (t) (transfer function h2) within the silent section (the section where the target sound component s (t) does not exist) determined by the target sound determination unit 22. The update of the estimated feedback sound component n2 (t) is stopped within the sound section determined by the target sound determination unit 22. Therefore, there is an advantage that the estimated feedback sound component n2 (t) can be estimated with high accuracy by suppressing the influence of the target sound component s (t).

However, since strict estimation of the transfer function h1 is difficult in practice, the actual transfer function h1 and the transfer function h2 estimated by the feedback sound estimation unit 24 do not always match. Therefore, as expressed by the following formula (2), one of the residual sound component (feedback sound component n1 (t) corresponding to the difference between the feedback sound component n1 (t) and the estimated feedback sound component n2 (t). Part) e1 (t) remains in the acoustic signal w2 (t) processed by the feedback sound suppression unit 26 (e1 (t) = n1 (t) −n2 (t)). Therefore, the acoustic processing apparatus 100A obtains the spectrum E2 (m, f) of the estimated residual sound component e2 (t) obtained by estimating the residual sound component e1 (t) from the spectrum W2 (m, f) of the acoustic signal w2 (t). By subtracting (spectrum subtraction), the residual sound component e1 (t) remaining in the acoustic signal w2 (t) is suppressed.
w2 (t) = w1 (t) -n2 (t)
= S (t) + n1 (t) -n2 (t)
= S (t) + e1 (t) (2)

  The frequency analysis unit 32 in FIG. 1 has a spectrum (frequency spectrum) N2 (m,) for each of a plurality of frames obtained by dividing the time series of the estimated feedback sound component n2 (t) generated by the feedback sound estimation unit 24 on the time axis. generates f). Symbol m indicates a frame (frame number), and symbol f indicates a frequency or frequency band (frequency bin) on the frequency axis. Similar to the frequency analysis unit 32, the frequency analysis unit 34 generates a spectrum (frequency spectrum) W2 (m, f) for each of a plurality of frames obtained by dividing the acoustic signal w2 (t) on the time axis. For the generation of the spectrum N2 (m, f) and the spectrum W2 (m, f), known frequency analysis such as fast Fourier transform and wavelet transform is arbitrarily employed.

  The residual sound estimation unit 36 estimates the residual sound component e1 (t) processed by the feedback sound suppression unit 26 from the feedback sound component n1 (t), and the spectrum E2 (m, f) of the estimated residual sound component e2 (t). ) (Ie, the residual sound component e1 (t) is estimated). In consideration of the tendency that the frequency characteristic of the feedback sound component n1 (t) (estimated feedback sound component n2 (t)) is reflected in the residual sound component e1 (t), the residual sound estimation unit 36 of this embodiment is A spectrum E2 (m, f) is generated from the spectrum N2 (m, f) of the estimated feedback sound component n2 (t) generated by the frequency analysis unit 32. FIG. Specifically, the residual sound estimation unit 36 multiplies the spectrum N2 (m, f) of the estimated feedback sound component n2 (t) by a coefficient δ (0 ≦ δ ≦ 1) to thereby estimate the residual sound component e2 (t ) Spectrum E2 (m, f). The coefficient δ can be set to a numerical value common to each frequency or a different numerical value for each frequency.

The spectrum subtracting unit 42 subtracts the spectrum E2 (m, f) of the estimated residual sound component e2 (t) from the spectrum W2 (m, f) of the acoustic signal w2 (t) to obtain the spectrum W3 (m, f). Calculate. The spectrum W3 (m, f) is a frequency spectrum of the acoustic signal z (t) processed by the acoustic processing apparatus 100A, and the phase spectrum θw2 (of the amplitude spectrum A (m, f) and the spectrum W2 (m, f). m, f) and expressed by the following formula (3).
W3 (m, f) = A (m, f) ・ e ^{jθw2 (m, f)} (3)

The spectrum subtraction unit 42 calculates the power spectrum | A (m, f) | ² of the spectrum W3 (m, f) by executing the following expressions (4a) and (4b), and the power spectrum | From the amplitude spectrum A (m, f) calculated from A (m, f) | ² and the phase spectrum θw2 (m, f) of the spectrum W2 (m, f), the spectrum W3 (m , f).

As shown in Expression (4a), when the intensity (power) | W2 (m, f) | ² of the spectrum W2 (m, f) exceeds a predetermined value TH, the spectrum subtracting unit 42 calculates the estimated residual sound component e2 ( t) The intensity of the spectrum E2 (m, f) | E2 (m, f) | ² is adjusted according to the subtraction coefficient α (m), and then the spectrum W2 (m, f) of the acoustic signal w2 (t) is adjusted. By subtracting (spectral subtraction) from the intensity | W2 (m, f) | ² , the power spectrum | A (m, f) | ² of the spectrum W3 (m, f) is calculated. Specifically, the product of the subtraction coefficient α (m) and the intensity | E2 (m, f) | ² is subtracted from the intensity | W2 (m, f) | ² . The predetermined value TH is set to, for example, a multiplication value of the subtraction coefficient α (m) and the intensity | E2 (m, f) | ² (TH = α (m) · | E2 (m, f) | ² ). . On the other hand, when the intensity | W2 (m, f) | ² is lower than the predetermined value TH, the spectrum subtraction unit 42 converts the intensity | E2 (m, f) | ² to the flooring coefficient β as shown in the equation (4b). The power spectrum | A (m, f) | ² is calculated by adjusting according to (m) (specifically, multiplying the intensity | E2 (m, f) | ² by the flooring coefficient β (m)) To do.

  1 converts the spectrum W3 (m, f) of each frame generated by the spectrum subtracting unit 42 into a time domain signal, and converts the converted signals of the preceding and succeeding frames to each other on the time axis. The acoustic signal z (t) is generated by the connection. Since the spectrum E2 (m, f) of the estimated residual sound component e2 (t) is subtracted from the spectrum W2 (m, f) of the acoustic signal w2 (t) as shown in Equation (4a), the feedback sound component n1 (t It is possible to generate an acoustic signal z (t) in which the residual sound component e1 (t) is effectively reduced. The acoustic signal z (t) generated by the inverse conversion unit 44 is transmitted to the communication network.

  By the way, in the spectrum W3 (m, f) generated by subtracting the spectrum E2 (m, f) as described above, a high-intensity component (isolated point) resulting from the spectrum subtraction is on the time axis and the frequency axis. May be perceived by the listener as artificial and annoying musical noise. The greater the degree of spectral subtraction (the degree of change in the spectrum before and after spectral subtraction), the more marked the musical noise in the spectrum W3 (m, f). Specifically, musical noise becomes noticeable when the subtraction coefficient α (m) is set to a large value (when the degree of spectrum subtraction is increased) or the flooring coefficient β (m) is set to a small value. .

  Part (A) in FIG. 2 is generated by counting the frequency of intensity at each frequency of the spectrum W2 (m, f) (acoustic signal w2 (t)) before spectrum subtraction for a predetermined number of frames. Is a graph of the frequency distribution (probability density function with intensity as a random variable) F1. As shown in part (A) of FIG. 2, the frequency (probability) of each intensity before spectrum subtraction is non-linearly distributed so as to decrease as the intensity increases from zero.

  Part (B) of FIG. 2 shows a frequency distribution generated by counting the frequency of intensity at each frequency of the spectrum W3 (m, f) (acoustic signal z (t)) after spectrum subtraction for a predetermined number of frames. (Probability density function) is a graph of F2. Since the frequency (probability) of the intensity close to zero is increased by spectrum subtraction, the distribution in the region where the intensity is close to zero in the frequency distribution F2 after the spectrum subtraction is compared with the frequency distribution F1 before the spectrum subtraction. It becomes a steep shape.

When kurtosis is introduced as a measure of the shape of the frequency distribution (steepness of inclination), the kurtosis KC of the frequency distribution F2 after the spectrum subtraction is higher than the kurtosis KB of the frequency distribution F1 before the spectrum subtraction. Become. The kurtosis κ is a higher-order statistic calculated from the n-th moment μn by the following equation (5).

  Focusing on the nature of kurtosis κ as a non-Gaussian index, it can be understood that the non-Gaussianity of the frequency distribution increases due to spectral subtraction. Since musical noise is highly non-Gaussian noise, the change in kurtosis κ before and after spectral subtraction (the relative ratio KC / KB of kurtosis KC to kurtosis KB, or the difference between kurtosis KC and kurtosis KB As (KC-KB)) increases, there is a tendency that the musical noise becomes obvious. That is, the lower the kurtosis KB of the frequency distribution F1 of the spectrum W2 (m, f) (acoustic signal w2 (t)) before spectrum subtraction, the more the musical noise caused by the spectrum subtraction becomes the spectrum W3 (m, f) (sound Signal z (t)).

  The spectrum E2 (m, f) of the estimated residual sound component e2 (t) from the spectrum E1 (m, f) of the residual sound component e1 (t) in the spectrum W2 (m, f) of the acoustic signal w2 (t). Focusing on the original purpose of subtracting (ie, ignoring the target sound component s (t) for convenience), the lower the kurtosis of the intensity distribution of the residual sound component e1 (t), the lower the spectral subtraction. The tendency that the musical noise resulting from this is easy to generate is grasped. Since the estimated residual sound component e2 (t) is an estimated value of the residual sound component e1 (t), the kurtosis in the frequency distribution of the intensity of the estimated residual sound component e2 (t) (spectrum E2 (m, f)) As KA is lower, there is a tendency that musical noise due to subtraction of the spectrum E2 (m, f) of the estimated residual sound component e2 (t) is more likely to occur. That is, the kurtosis KA can be used as a quantitative index indicating the degree of occurrence of musical noise. In consideration of the above tendency, in the present embodiment, the subtraction coefficient α (m) in Expression (4a) and the flooring coefficient β (m) in Expression (4b) (that is, the degree of spectrum subtraction) are estimated residual sound. Control is variably performed according to the kurtosis KA (m) in the frequency distribution of the intensity of the component e2 (t) (spectrum E2 (m, f)).

  The kurtosis calculation unit 52A in FIG. 1 calculates the kurtosis KA (m) in the frequency distribution of the intensity of the spectrum E2 (m, f) (estimated residual sound component e2 (t)) generated by the residual sound estimation unit 36 for each frame. To calculate. A specific example of a method for calculating the kurtosis κ for the frequency distribution of M (M is a natural number of 2 or more) intensities x1 to xM will be described in detail below.

数式(6)の係数Ｃは、ガンマ関数Γ(k)を利用して以下のように定義される。

The frequency distribution of M intensities x1 to xM is approximated by a function Ga (x; k, θ) of the following formula (6).

The coefficient C in Equation (6) is defined as follows using the gamma function Γ (k).

The following equation (7) is derived by replacing the distribution function (probability density function) P (x) in the definition equation of the second moment μ2 with the function Ga (x; k, θ) of the equation (6). .

Similar to the derivation of the second-order moment μ2, the distribution function (probability density function) P (x) in the definition equation of the fourth-order moment μ4 is replaced with the function Ga (x; k, θ) of the equation (6). The following formula (8) is derived.

Substituting the second-order moment μ2 in equation (7) and the fourth-order moment μ4 in equation (8) into equation (5) yields the following equation (9) that defines kurtosis κ.

The kurtosis calculation unit 52A in FIG. 1 performs M number of spectrums E2 (m, f) over a predetermined number of frames including the mth frame (for example, a plurality of frames with the mth frame as the last). The kurtosis κ in the equation (9) when the intensity | E2 (m, f) | ² is the intensity x1 to xM is calculated as the kurtosis KA (m).

  The coefficient setting unit 54 in FIG. 1 variably sets the subtraction coefficient α (m) and the flooring coefficient β (m) according to the kurtosis KA (m) calculated by the kurtosis calculation unit 52A. Each time the kurtosis KA (m) is calculated for each frame, the subtraction coefficient α (m) and the flooring coefficient β (m) are sequentially updated. As described with reference to FIG. 2, the smaller the kurtosis KA (m) (that is, the smaller the kurtosis in the frequency distribution of the intensity of the residual sound component e1 (t)), the more noise is generated after subtracting the spectrum. There is a tendency to be easy to do. In consideration of the above tendency, the coefficient setting unit 54 reduces the degree of spectrum subtraction as the kurtosis KA (m) is smaller (that is, the musical noise due to spectrum subtraction is reduced). A subtraction coefficient α (m) and a flooring coefficient β (m) are set.

  Specifically, since the musical noise of the acoustic signal z (t) is suppressed as the subtraction coefficient α (m) is smaller, the coefficient setting unit 54 determines that the subtraction coefficient α (m) is smaller as the kurtosis KA (m) is smaller. Is set to a small value (that is, a value that suppresses the amount of subtraction from the spectrum W2 (m, f)). The specific relationship between the subtraction coefficient α (m) and the kurtosis KA (m) is arbitrary. For example, the subtraction coefficient α (m) is obtained by multiplying the kurtosis KA (m) by a predetermined positive number. The structure to calculate is suitable.

  Further, since the musical noise of the acoustic signal z (t) is suppressed as the flooring coefficient β (m) is larger, the coefficient setting unit 54 sets the flooring coefficient β (m) as the kurtosis KA (m) is smaller. A large numerical value (that is, a numerical value that suppresses the difference between the spectrum W2 (m, f) and the spectrum W3 (m, f)) is set. For example, a configuration in which a numerical value obtained by subtracting the kurtosis KA (m) from a predetermined value is calculated as the flooring coefficient β (m) is preferable. The subtraction coefficient α (m) and the flooring coefficient β (m) are stored for each numerical value (or range) of the kurtosis KA (m), and the subtraction coefficient α corresponding to the kurtosis KA (m) is stored. A configuration in which the coefficient setting unit 54 searches for (m) and the flooring coefficient β (m) is also suitable.

  In the above embodiment, the subtraction coefficient α (m) applied to the adjustment of the spectrum E2 (m, f) subtracted from the spectrum W2 (m, f) of the acoustic signal w2 (t) is the estimated residual sound component e2 ( It is variably set according to the kurtosis KA (m) in the frequency distribution of the intensity of t) (that is, the degree of musical noise that may occur due to spectral subtraction). Therefore, compared with a configuration in which the subtraction coefficient α (m) does not depend on the kurtosis KA (m) (for example, a configuration in which the subtraction coefficient α (m) is fixed to a predetermined value), the musical noise in the acoustic signal z (t) is reduced. There is an advantage that the feedback sound component n1 (t) (particularly the residual sound component e1 (t)) can be suppressed while being effectively suppressed. Further, since the flooring coefficient β (m) is also variably set according to the kurtosis KA (m), the subtraction coefficient α (m) and the flooring coefficient β (m) do not depend on the kurtosis KA (m). Compared to the configuration (for example, the configuration in which the subtraction coefficient α (m) and the flooring coefficient β (m) are fixed to predetermined values) and the configuration in which only the subtraction coefficient α (m) is set according to the kurtosis KA (m) The effect of reducing musical noise due to spectral subtraction is particularly remarkable.

<B: Second Embodiment>
Next, a second embodiment of the present invention will be described. In addition, about the element in which an effect | action and a function are equivalent to 1st Embodiment in each following form, the same code | symbol as the above is attached | subjected and each detailed description is abbreviate | omitted suitably.

  The feedback sound estimation unit 24 of the first embodiment stops the operation (update of the estimated feedback sound component n2 (t)) in the silent section specified by the target sound determination unit 22 to obtain the estimated feedback sound component n2 (t). Estimate with high accuracy. However, an error may occur in the estimated feedback sound component n2 (t) applied to the voiced section due to various factors (for example, the influence of the estimation error in the previous silent section). The kurtosis calculation unit 52A of the first embodiment calculates the kurtosis KA (m) for the spectrum E2 (m, f) of the estimated residual sound component e2 (t) corresponding to the estimated feedback sound component n2 (t). Therefore, an error caused by the estimated feedback sound component n2 (t) occurs in the kurtosis KA (m). Therefore, the subtraction coefficient α (m) and the flooring coefficient β (m) may not be set to appropriate values. The second embodiment is a form for solving the above problem.

  FIG. 3 is a block diagram of the sound processing apparatus 100B according to the second embodiment. As indicated by the dashed arrows in FIG. 3, the determination result by the target sound determination unit 22 is notified to the kurtosis calculation unit 52 </ b> A together with the feedback sound estimation unit 24. The kurtosis calculation unit 52A calculates the kurtosis KA (m) for each frame in the silent section specified by the target sound determination unit 22 (section in which the target sound component s (t) does not exist) as in the first embodiment. However, the calculation of the kurtosis KA (m) is stopped within the sound section. Therefore, in the voiced section, the coefficient setting unit 54 is instructed continuously with the kurtosis KA (m) calculated at the end of the immediately preceding silent section. The operation of the coefficient setting unit 54 calculating the subtraction coefficient α (m) and the flooring coefficient β (m) according to the kurtosis KA (m) is the same as that in the first embodiment.

  In the above embodiment, the update of the kurtosis KA (m) is stopped in the sounded section, so the error of the estimated feedback sound component n2 (t) in the sounded section is the subtraction coefficient α (m) or the flooring coefficient. Does not affect β (m). Therefore, there is an advantage that musical noise caused by spectrum subtraction can be more effectively suppressed than in the first embodiment. In addition, since the calculation of the kurtosis KA (m) is stopped in the voiced section, the kurtosis is compared with the first embodiment in which the kurtosis KA (m) is calculated in both the voiced section and the silent section. There is also an advantage that the processing amount of the calculation unit 52A is reduced.

<C: Third Embodiment>
As described above with reference to FIG. 2, the lower the kurtosis KB (m) of the frequency distribution F1 of the intensity of the spectrum W2 (m, f) (acoustic signal w2 (t)) before the spectrum subtraction, the lower the spectrum subtraction. The resulting musical noise tends to occur easily in the spectrum W3 (m, f) (acoustic signal z (t)). In consideration of the above tendency, in the third embodiment, the degree of spectrum subtraction (subtraction coefficient α (m) and the like) according to the kurtosis KB (m) of the frequency distribution F1 of the intensity of the spectrum W2 (m, f). The flooring coefficient β (m)) is controlled.

  FIG. 4 is a block diagram of a sound processing apparatus 100C according to the third embodiment. As illustrated in FIG. 4, the acoustic processing device 100C has a configuration in which the kurtosis calculation unit 52A in the acoustic processing device 100A of the first embodiment is replaced with a kurtosis calculation unit 52B. The kurtosis calculation unit 52B calculates the kurtosis KB (m) in the frequency distribution F1 of the intensity of the spectrum W2 (m, f) (acoustic signal w2 (t)) generated by the frequency analysis unit 34 for each frame. For the calculation of the kurtosis KB (m), the same method as the calculation of the kurtosis KA (m) in the first embodiment is adopted.

  As in the first embodiment, the coefficient setting unit 54 variably sets the subtraction coefficient α (m) and the flooring coefficient β (m) according to the kurtosis KB (m) calculated by the kurtosis calculation unit 52B. To do. Considering the tendency that musical noise is more likely to occur as the kurtosis KB (m) is lower, the coefficient setting unit 54 reduces the subtraction coefficient α so that the degree of spectral subtraction is reduced as the kurtosis KB (m) is smaller. Set (m) and flooring coefficient β (m). Specifically, the coefficient setting unit 54 sets the subtraction coefficient α (m) to a smaller value as the kurtosis KB (m) is smaller, and sets the flooring coefficient β (m) as the kurtosis KB (m) is smaller. Set to a large number.

  In the third embodiment, the subtraction coefficient α (m) and flooring coefficient β (m) are controlled in accordance with the kurtosis KB (m) of the spectrum W2 (m, f) of the acoustic signal w2 (t). As in the first embodiment, there is an advantage that the feedback sound component n1 (t) (particularly the residual sound component e1 (t)) can be suppressed while effectively suppressing the musical noise in the acoustic signal z (t). In addition, since the kurtosis KB (m) of the acoustic signal w2 (t) is used, an appropriate subtraction coefficient α (m) and flooring coefficient are not affected by the estimation accuracy of the estimated residual sound component e2 (t). There is also an advantage that β (m) can be calculated.

<D: Fourth Embodiment>
The acoustic signal w2 (t) may or may not include the target sound component s (t). Since the feedback sound component n1 (t) is a reverberant sound that reaches the sound collecting device 14 through reflection or diffusion, most of the feedback sound component n1 (t) is a target sound component s (t) that reaches the sound collecting device 14 directly from the sound source. In comparison, kurtosis (non-Gaussian) is low. That is, the kurtosis KB (m) when the acoustic signal w2 (t) does not include the target sound component s (t) is the kurtosis KB when the acoustic signal w2 (t) includes the target sound component s (t). Lower than (m).

  Therefore, the kurtosis KB (m) is subtracted from the kurtosis KB (m) so that the musical noise is effectively suppressed under the kurtosis KB (m) when the acoustic signal w2 (t) does not include the target sound component s (t). Assuming that the relationship with the coefficient α (m) is determined, the kurtosis KB (m) of the acoustic signal w2 (t) is the subtraction coefficient α (m) regardless of the presence or absence of the target sound component s (t). In the configuration of the third embodiment reflected in the above, the kurtosis KB is compared with the case where the acoustic signal w2 (t) includes the target sound component s (t) (that is, as compared with the case where the target sound component s (t) is not included). When (m) is high), the subtraction coefficient α (m) is set to a large value, and the spectrum E (m, f) of the estimated residual sound component e2 (t) becomes the spectrum W2 (m, There is a possibility of excessive subtraction from f). Therefore, in the fourth embodiment, the calculation of the kurtosis KB (m) is stopped when the acoustic signal w2 (t) includes the target sound component s (t).

  FIG. 5 is a block diagram of a sound processing apparatus 100D according to the fourth embodiment. As indicated by the dashed arrows in FIG. 5, the determination result by the target sound determination unit 22 is notified to the kurtosis calculation unit 52 </ b> B together with the feedback sound estimation unit 24. Similar to the kurtosis calculation unit 52A of the second embodiment, the kurtosis calculation unit 52B executes or stops the calculation of the kurtosis KB (m) according to the determination result by the target sound determination unit 22. That is, the kurtosis calculation unit 52B calculates and updates the kurtosis KB (m) for each frame in the silent section specified by the target sound determination unit 22, as in the third embodiment. Stop calculating kurtosis KB (m). The operation of the coefficient setting unit 54 to calculate the subtraction coefficient α (m) and the flooring coefficient β (m) according to the kurtosis KB (m) is the same as in the third embodiment.

  In the above embodiment, since the calculation of the kurtosis KB (m) is stopped within the sound section where the acoustic signal w2 (t) includes the target sound component s (t), the subtraction coefficient α (m) and the flooring coefficient β (m) is not affected by the target sound component s (t). Therefore, the effect that the musical noise resulting from spectrum subtraction can be suppressed appropriately is realized. Also, since the calculation of kurtosis KB (m) stops in the voiced section, the kurtosis calculation is compared with the third embodiment in which the kurtosis KB (m) is calculated in both the voiced and silent sections. There is also an advantage that the processing amount of the unit 52B is reduced.

<E: Fifth Embodiment>
FIG. 6 is a block diagram of a sound processing apparatus 100E according to the fifth embodiment. As shown in FIG. 6, the sound processing device 100E adds the kurtosis calculating unit 52B of the third embodiment (FIG. 4) to the sound processing device 100A of the first embodiment (FIG. 1) and adds it to the coefficient setting unit 54. The determination unit 56 is added. The kurtosis calculation unit 52A calculates the kurtosis KA (m) of the estimated residual sound component e2 (t) (spectrum E2 (m, f)) for each frame, and the kurtosis calculation unit 52B calculates the acoustic signal w2 (t) ( The kurtosis KB (m) of the spectrum W2 (m, f)) is calculated for each frame.

  The estimated residual sound component e2 (t) (spectrum E2 (m, f)) for which the kurtosis calculation unit 52A calculates the kurtosis KA (m) is the acoustic for which the kurtosis calculation unit 52B calculates the kurtosis KB (m). This is an estimated value of the residual sound component e1 (t) in the signal w2 (t). Therefore, when the acoustic signal w2 (t) includes only the residual sound component e1 (t) (that is, when it does not include the target sound component s (t)), the kurtosis KA (m) and the kurtosis KB (m ) Approximate. On the other hand, as described above for the fourth embodiment, the kurtosis KB (m) is obtained when the acoustic signal w2 (t) includes the target sound component s (t) and does not include the target sound component s (t). Therefore, when the acoustic signal w2 (t) includes the target sound component s (t), the kurtosis KA (m) and the kurtosis KB (m) are different.

  Considering the above tendency, the determination unit 56 determines whether or not the kurtosis KA (m) calculated by the kurtosis calculation unit 52A approximates the kurtosis KB (m) calculated by the kurtosis calculation unit 52B. Accordingly, the voiced section and the silent section are distinguished. That is, the determination unit 56 determines the presence or absence of the target sound component s (t) by a method different from that of the target sound determination unit 22. Specifically, the determination unit 56 determines that the acoustic signal w2 (t) does not include the target sound component s (t) when the kurtosis KA (m) and the kurtosis KB (m) are approximated (silent section). If the kurtosis KA (m) and the kurtosis KB (m) are not approximate, the acoustic signal w2 (t) includes the target sound component s (t) (which is a voiced section). Is determined. For example, if the difference value (absolute value) between the kurtosis KA (m) and the kurtosis KB (m) is below a predetermined threshold, it is determined as a silent section, and the kurtosis KA (m) and the kurtosis KB (m ) Is greater than a predetermined threshold value, it is determined as a sound section.

  The coefficient setting unit 54 executes or stops the update of the subtraction coefficient α (m) and the flooring coefficient β (m) according to the determination result by the determination unit 56. Specifically, the coefficient setting unit 54, for each frame in the silent section specified by the determination unit 56, according to the kurtosis KA (m) calculated by the kurtosis calculation unit 52A, as in the first embodiment. To set the subtraction coefficient α (m) and the flooring coefficient β (m). On the other hand, the calculation of the subtraction coefficient α (m) and the flooring coefficient β (m) is stopped for each frame in the sound section specified by the determination unit 56. When the calculation of the subtraction coefficient α (m) and the flooring coefficient β (m) is stopped, the coefficient setting unit 54 calculates the subtraction coefficient α (m) and the flooring coefficient β (m) before the calculation is stopped (sound period). Is initialized to a predetermined value irrelevant to the subtraction coefficient α (m) and the flooring coefficient β (m) (for example, a numerical value for reducing the degree of spectrum subtraction). Specifically, the subtraction coefficient α (m) is set to a value close to zero, and the flooring coefficient β (m) is set to a value close to 1. Therefore, it is possible to prevent the spectral subtraction for each frame in the sounded section from becoming excessive due to the influence of the acoustic signal w2 (t) (feedback sound component n1 (t)) in the silent section. However, even if the updated subtraction coefficient α (m) and flooring coefficient β (m) in the silent section are applied to the frames in the voiced section, the degree of spectrum subtraction may not be excessive. Accordingly, a configuration in which the subtraction coefficient α (m) and the flooring coefficient β (m) updated in the silent section are continuously instructed from the coefficient setting unit 54 to the spectrum subtracting section 42 even in the immediately following voiced section can be adopted. .

  In the above embodiment, the update of the subtraction coefficient α (m) and the flooring coefficient β (m) applied to the spectral subtraction stops in the sound section, so that the characteristic of the target sound component s (t) is spectral subtraction. (For example, excessive spectrum subtraction due to the target sound component s (t)) is prevented. In addition, since the calculation of the subtraction coefficient α (m) and flooring coefficient β (m) is stopped in the sounded section, the subtraction coefficient α (m) and flooring coefficient β ( Compared with the configuration for calculating m), there is also an advantage that the processing amount of the coefficient setting unit 54 is reduced. The configuration in which the coefficient setting unit 54 calculates the subtraction coefficient α (m) and the flooring coefficient β (m) from the kurtosis KA (m) has been exemplified above, but the subtraction coefficient α (m) and the flooring coefficient β (m) is calculated from the kurtosis KB (m) as in the third embodiment, and according to both the kurtosis KA (m) and the kurtosis KB (m) (for example, the kurtosis KA (m ) And the kurtosis KB (m) (based on the average value) are also used.

<F: Sixth Embodiment>
FIG. 7 is a block diagram of a sound processing apparatus 100F according to the sixth embodiment. As illustrated in FIG. 7, the sound processing device 100F includes a kurtosis calculation unit 52B (the kurtosis calculation unit 52A in the sound processing device 100A of the first embodiment) in the sound processing device 100C (FIG. 4) of the third embodiment. In this configuration, the kurtosis calculation unit 52C is replaced. The kurtosis calculation unit 52C spectrums the kurtosis KC (m) (part (B) of FIG. 2) in the frequency distribution F2 of the intensity of the acoustic signal z (t) (spectrum W3 (m, f)) after the spectrum subtraction. Calculate for each frame from W3 (m, f). For the calculation of the kurtosis KC (m), the same method as the calculation of the kurtosis KB (m) in the third embodiment is adopted.

  The coefficient setting unit 54 sequentially sets the subtraction coefficient α (m) and the flooring coefficient β (m) for each frame according to the calculation result by the kurtosis calculation unit 52C. Since the kurtosis KC (m) is calculated using the result of spectrum subtraction already performed (spectrum W3 (m, f)), the subtraction coefficient α (m) and flooring coefficient β ( m) is set according to the kurtosis KC (m-1) calculated from the spectrum W3 (m-1, f) of the (m-1) th frame. As described with reference to FIG. 2, musical noise is more likely to occur as the kurtosis KC (m) increases. Therefore, the coefficient setting unit 54 sets the subtraction coefficient α (m) to a smaller numerical value and the flooring coefficient β (m) to a larger numerical value as the kurtosis KC (m−1) increases. In the sixth embodiment, the same effects as those of the first and third embodiments are realized.

  In addition, the structure which stops calculation and update of kurtosis KC (m) within a sound area similarly to 4th Embodiment is also employ | adopted. In the above, the subtraction coefficient α (m) and the flooring coefficient β (m) are calculated according to the kurtosis KC (m−1), but the kurtosis KC (m) of the m-th frame A configuration for calculating the subtraction coefficient α (m) and the flooring coefficient β (m) is also employed. For example, the spectral subtraction using the subtraction coefficient α (m) and the flooring coefficient β (m) set according to the kurtosis KC (m) after the spectral subtraction of the mth frame is performed for the mth frame. It is performed on the spectrum W2 (m, f) (ie, the spectral subtraction for calculating the kurtosis KC (m) and the original target spectral subtraction are performed for one frame).

<G: Seventh Embodiment>
FIG. 8 is a block diagram of a sound processing apparatus 100G according to the seventh embodiment of the present invention. As shown in FIG. 8, the acoustic processing device 100G calculates the kurtosis calculation unit 52B of the third embodiment for calculating the kurtosis KB (m) before spectrum subtraction, and the kurtosis KC (m) after spectrum subtraction. The kurtosis calculating unit 52C of the sixth embodiment is configured.

  The coefficient setting unit 54 varies the subtraction coefficient α (m) and the flooring coefficient β (m) of the mth frame according to both the kurtosis KB (m−1) and the kurtosis KC (m−1). Set to. Specifically, the subtraction coefficient α (m) and the flooring coefficient β are set such that the greater the difference between the kurtosis KB (m−1) and the kurtosis KC (m−1), the lower the degree of spectral subtraction. (m) is calculated. For example, the coefficient setting unit 54 calculates the relative ratio KC (m-1) / KB (m-1) of the kurtosis KC (m-1) to the kurtosis KB (m-1) or the kurtosis KC (m-1). ) And kurtosis KB (m-1), the larger the difference value (KC (m-1) -KB (m-1)), the smaller the subtraction coefficient α (m) and the flooring coefficient β Set (m) to a large number.

  In the above embodiment, both the kurtosis KB (m) before execution of spectral subtraction and the kurtosis KC (m) after execution are reflected in the subtraction coefficient α (m) and the flooring coefficient β (m). Therefore, compared with the configuration using only one of kurtosis KB (m) and kurtosis KC (m) (the third embodiment and the sixth embodiment), the accuracy of suppressing musical noise caused by spectral subtraction is improved. It is possible to improve.

<H: Modification>
Each form illustrated above can be variously modified. Specific modes of deformation are exemplified below. Note that two or more aspects arbitrarily selected from the following examples may be appropriately combined.

(1) Modification 1
The object for which the kurtosis calculating unit 52B calculates the kurtosis KB (m) is changed as appropriate. For example, the kurtosis KB (m) can be calculated from the intensity (x1 to xM) of each spectrum of the acoustic signal w1 (t) generated by the sound collection device 14. A configuration for calculating kurtosis KB (m) from a signal in the time domain is also suitable. For example, the mathematical expression (9) applying the kurtosis κ (that is, M intensities x1 to xM arranged in time series) in the frequency distribution of the intensity (each signal value) of the acoustic signal w1 (t) and the acoustic signal w2 (t). )) Can be calculated as the kurtosis KB (m) before spectral subtraction.

  Similarly, the target in which the kurtosis calculation unit 52C calculates the kurtosis KC (m) in the sixth embodiment and the seventh embodiment is appropriately changed. For example, in a configuration in which the kurtosis κ in the frequency distribution of each signal value of the acoustic signal z (t) in the time domain is calculated as the kurtosis KC (m) after spectrum subtraction, or by a predetermined process for the acoustic signal z (t) A configuration is employed in which the kurtosis KC (m) is calculated from the generated acoustic signal (or spectrum).

(2) Modification 2
The position of transformation from the time domain to the frequency domain is arbitrarily changed. In other words, except that the processing by the spectrum subtracting unit 42 is performed in the frequency domain, whether the processing of each element of the acoustic processing device 100 (100A to 100F) is performed in the frequency domain or the time domain is described in the present invention. Is unquestionable. For example, in a configuration in which the acoustic signal w1 (t) immediately after generation by the sound collection device 14 (before processing by the feedback sound suppression unit 26) is converted to the frequency domain, processing by the feedback sound estimation unit 24 and the feedback sound suppression unit 26 is performed. Run in the frequency domain. Further, for example, in a configuration in which the frequency analysis unit 32 is arranged at the subsequent stage of the residual sound estimation unit 36, the residual sound estimation unit 36 generates an estimated residual sound component e2 (t) in the time domain.

(3) Modification 3
The residual sound estimation unit 36 in each of the above forms is not an essential element. For example, in each of the above embodiments, the spectrum E2 (m, f) of the estimated residual sound component e2 (t) is generated by multiplying the spectrum N2 (m, f) of the estimated feedback sound component n2 (t) by the coefficient δ. However, if the subtraction coefficient α (m) and the flooring coefficient β (m) are set in consideration of the coefficient δ, the spectrum N2 (m, f) of the estimated feedback sound component n2 (t) is estimated as the estimated residual sound component e2. It is also possible to supply the spectrum subtraction unit 42 as the spectrum E2 (m, f) of (t).

(4) Modification 4
The method for calculating the kurtosis κ (KA (m), KB (m), KC (m)) is not limited to the illustration of each form. For example, a configuration for approximating the frequency distribution of the intensities x1 to xM with a predetermined function (for example, Equation (6)) is not essential. Further, the calculation and update cycle of kurtosis κ (KA (m), KB (m), KC (m)) is arbitrary. For example, a configuration in which kurtosis κ is calculated in units of a plurality of frames, or a configuration in which kurtosis κ is calculated at a period unrelated to the frame are also employed.

(5) Modification 5
The acoustic processing device 100 (100A to 100F) in each of the above embodiments is also suitable as a device that suppresses howling caused by feedback sound (howling suppression device). For example, a configuration in which the acoustic signal z (t) generated by the inverse transform unit 44 is supplied to the sound emitting device 12 after being amplified by an amplifier (typically, the sound is emitted by adjusting the sound volume around the sound collecting device 14. A loudspeaker radiating from the device 12 is employed. In the above configuration, the feedback sound arriving at the sound collecting device 14 from the sound emitting device 12 is suppressed by the feedback sound suppressing unit 26 and the spectrum subtracting unit 42, so that the musical noise in the radiated sound from the sound emitting device 12 is reduced. While being suppressed, howling due to a loop formed by the sound emitting device 12, the sound collecting device 14, and the sound processing device 100 is effectively prevented.

100A, 100B, 100C, 100D, 100E, 100F, 100G... Acoustic processing device, 12... Sound emission device, 14... Sound collection device, 22 .. target sound determination unit, 24. …… Feedback sound suppression unit 32, 34 Frequency analysis unit 36 Residual sound estimation unit 42 Spectral subtraction unit 44 Inverse transformation unit 52A 52B 52C Kurtosis calculation unit 54... Coefficient setting unit, 56.

Claims (9)

Feedback sound estimation means for generating an estimated feedback sound component that estimates the feedback sound coming from the sound emitting device to the sound collecting device;
Feedback sound suppression means for suppressing the estimated feedback sound component from the acoustic signal generated by the sound collection device;
Kurtosis calculating means for calculating the kurtosis in the frequency distribution of the intensity of the estimated residual sound component obtained by estimating the component remaining after the processing by the feedback sound suppression means of the feedback sound;
Coefficient setting means for setting a subtraction coefficient according to the kurtosis calculated by the kurtosis calculating means;
An acoustic processing apparatus comprising: a spectrum subtracting unit that adjusts a spectrum of the estimated residual sound component according to the subtraction coefficient and subtracts it from a spectrum of the acoustic signal processed by the feedback sound suppressing unit. The spectrum subtracting unit multiplies the spectrum of the estimated residual sound component by the subtraction coefficient and subtracts it from the spectrum of the acoustic signal processed by the feedback sound suppression unit,
The sound processing apparatus according to claim 1, wherein the coefficient setting unit sets the subtraction coefficient to a larger numerical value as the kurtosis increases. The kurtosis calculation means calculates the kurtosis of the estimated residual sound component and the kurtosis in the frequency distribution of the intensity of the acoustic signal generated by the sound collection device,
When the kurtosis of the estimated residual sound component approximates the kurtosis of the acoustic signal, the coefficient setting means, according to at least one of the kurtosis of the estimated residual sound component and the kurtosis of the acoustic signal, The acoustic processing apparatus according to claim 1, wherein a subtraction coefficient is set, and updating of the subtraction coefficient is stopped when the kurtosis of the estimated residual sound component and the kurtosis of the acoustic signal are not approximated. Feedback sound estimation means for generating an estimated feedback sound component that estimates the feedback sound coming from the sound emitting device to the sound collecting device;
Feedback sound suppression means for suppressing the estimated feedback sound component from the acoustic signal generated by the sound collection device;
Kurtosis calculating means for calculating kurtosis in the frequency distribution of the intensity of the acoustic signal generated by the sound collecting device;
Coefficient setting means for setting a subtraction coefficient according to the kurtosis calculated by the kurtosis calculating means;
Of the feedback sound, the spectrum of the estimated residual sound component obtained by estimating the component remaining after the processing by the feedback sound suppression means is adjusted according to the subtraction coefficient and subtracted from the spectrum of the acoustic signal after the processing by the feedback sound suppression means And a spectral subtracting means. The kurtosis calculation means calculates the kurtosis of the acoustic signal before processing by the spectrum subtraction means,
The spectrum subtracting means multiplies the spectrum of the estimated residual sound component by the subtraction coefficient and subtracts it from the spectrum of the acoustic signal processed by the feedback sound suppression means,
The sound processing apparatus according to claim 4, wherein the coefficient setting unit sets the subtraction coefficient to a larger numerical value as the kurtosis increases. The kurtosis calculation means calculates the kurtosis of the acoustic signal after processing by the spectrum subtraction means,
The spectrum subtracting unit multiplies the spectrum of the estimated residual sound component by the subtraction coefficient and subtracts it from the spectrum of the acoustic signal processed by the feedback sound suppression unit,
The sound processing apparatus according to claim 4, wherein the coefficient setting unit sets the subtraction coefficient to a smaller numerical value as the kurtosis is larger. A target sound determining means for determining the presence or absence of a target sound component for the acoustic signal generated by the sound collecting device;
The kurtosis calculating unit calculates the kurtosis when the target sound determining unit determines that the target sound component does not exist, and the kurtosis when the target sound determining unit determines that the target sound component exists. The sound processing apparatus according to any one of claims 1 to 6, wherein the calculation of is stopped. A feedback sound estimation process for generating an estimated feedback sound component estimating a feedback sound arriving at the sound collecting device from the sound emitting device;
Feedback sound suppression processing for suppressing the estimated feedback sound component from the acoustic signal generated by the sound collection device;
A kurtosis calculation process for calculating a kurtosis in the frequency distribution of the intensity of the estimated residual sound component obtained by estimating the component remaining after execution of the feedback sound suppression process in the feedback sound;
A coefficient setting process for setting a subtraction coefficient according to the kurtosis calculated in the kurtosis calculation process;
A program that causes a computer to execute spectrum subtraction processing that adjusts a spectrum of the estimated residual sound component according to the subtraction coefficient and subtracts it from a spectrum of an acoustic signal after execution of the feedback sound suppression processing. A feedback sound estimation process for generating an estimated feedback sound component estimating a feedback sound arriving at the sound collecting device from the sound emitting device;
Feedback sound suppression processing for suppressing the estimated feedback sound component from the acoustic signal generated by the sound collection device;
Kurtosis calculation processing for calculating the kurtosis in the frequency distribution of the intensity of the acoustic signal after generation by the sound collection device;
A coefficient setting process for setting a subtraction coefficient according to the kurtosis calculated in the kurtosis calculation process;
A spectrum of an estimated residual sound component obtained by estimating a component remaining after execution of the feedback sound suppression processing in the feedback sound is adjusted according to the subtraction coefficient and subtracted from a spectrum of the acoustic signal after execution of the feedback sound suppression processing. A program that causes a computer to execute spectral subtraction processing.

Images