JP2022065565A - Acoustic signal processing apparatus, acoustic signal processing method, and program - Google Patents

Abstract

To provide a technique for removing noises from monaurally recorded audio signals.SOLUTION: An acoustic signal processing apparatus includes: a compensated spectrogram acquisition unit for acquiring a processing target spectrogram that satisfies an origin matching condition in which a start position of a removal target sound in a processing target sound time series indicating a monaurally recorded processing target sound matches a time origin in a reference time series indicating a pre-recorded removal target sound, and satisfies a sampling frequency matching condition in which a sampling frequency in the processing target sound time series matches the sampling frequency in the reference time series, and for acquiring a reference spectrogram that satisfies the origin matching condition and the sampling frequency matching condition; and a target complex spectrogram estimation unit for estimating a time series that maximizes likelihood for a predetermined probability distribution of a target complex spectrogram obtained as a difference in complex spectrum at the same time between the processing target spectrogram and the reference spectrogram multiplied by a frequency transfer function.SELECTED DRAWING: Figure 1

Description

Patent Law Article 30, Paragraph 2 Application Applicable Proceedings of the Acoustical Society of Japan 2020 Spring Research Presentation (at Saitama University) on March 2, 2nd year of the Ordinance

The present invention relates to an acoustic signal processing device, an acoustic signal processing method and a program.

Various techniques have been proposed to generate a time series showing a sound in which the influence of a sound other than the target sound (that is, noise) is reduced from a time series showing the sound when the target sound is recorded (non-patent documents). 1 and 2).

R. Martin, “Noise Power Spectral Density Optimization Based on Optimal Smoothing and Minimum Statistics” IEEE Trans. On SAP, vol. 9, no. 5, pp.504-512, 2001. T. Gerkmann, et al., “Unbiased MMSE-Based Noise Power Optimization With Low Complexity and Low Tracking Delay” IEEE / ACM Trans. On ASLP, vol. 20, no. 4, pp.1383 | 1393, 2012.

Such a proposal includes a technique for removing unsteady noise from monaural recording. However, it cannot be said that the technique sufficiently removes noise. As described above, it is difficult to generate a time series of sounds in which the influence of noise is reduced from the time series of sounds recorded by one microphone (that is, monaural recording).

In view of the above circumstances, it is an object of the present invention to provide a technique for generating a time series of sounds in which the influence of noise is further reduced from the time series of sounds recorded in monaural.

One aspect of the present invention is a reference time series indicating a start position at which the removal target sound to be removed in the processing target sound time series indicating the monaurally recorded processing target sound is started and the pre-recorded removal target sound. When the complex spectrum of the processing target sound time series satisfies the origin matching condition that the time origin matches and the sampling frequency matching condition that the sampling frequency of the processing target sound time series matches the sampling frequency of the reference time series. A compensated spectrogram acquisition unit for acquiring a processed spectrogram that is a series, a reference spectrogram that is a time series of a complex spectrum of the reference time series that satisfies the origin matching condition and the sampling frequency matching condition, and the processed spectrogram. The likelihood of the objective complex spectrum, which is the complex spectrum in the objective complex spectrogram obtained as the difference of the complex spectrum at the same time from the reference spectrogram multiplied by the frequency transfer function, with respect to a predetermined probability distribution is maximized. An acoustic signal processing device including an objective complex spectrogram estimation unit that estimates the time series to be used, and a target sound time series generation unit that generates a time series of sounds having the objective complex spectrogram estimated by the objective complex spectrogram estimation unit. Is.

INDUSTRIAL APPLICABILITY According to the present invention, it is possible to generate a time series of sounds in which the influence of noise is further reduced from the time series of sounds recorded in monaural.

Explanatory drawing explaining the outline of the acoustic signal processing apparatus 1 of embodiment. The figure which shows an example of the functional structure of the acoustic signal processing apparatus 1 of embodiment. The figure which shows an example of the functional structure of the control part 10 in an embodiment. The flowchart which shows an example of the flow of the process executed by the acoustic signal processing apparatus 1 of embodiment. Explanatory drawing explaining the experimental environment of the evaluation experiment in embodiment. The figure which shows an example of the result of the evaluation experiment in an embodiment.

(Embodiment)
FIG. 1 is an explanatory diagram illustrating an outline of the acoustic signal processing device 1 of the embodiment. The acoustic signal processing device 1 uses a time series indicating the recorded sound (hereinafter referred to as “processing target sound”) when the target sound (hereinafter referred to as “target sound”) is monaurally recorded, and uses noise. Generates a time series of sounds with reduced effects. Hereinafter, the time series indicating the processing target sound is referred to as a processing target sound time series. Monaural recording means recording with one microphone.

Noise is a sound other than the target sound. The noise may be any sound other than the target sound. The noise includes not only random sounds but also sounds of classical music played from speakers and other than the target sound. Reducing the influence of sound specifically means reducing the amplitude of sound.

Specifically, the acoustic signal processing device 1 uses a reference time series to form a time series (hereinafter referred to as “target sound time series”) indicating a sound in which the influence of the sound to be removed is suppressed from the time series of the sound to be processed. Generate. The sound to be removed is a sound other than the target sound among the sounds included in the sound to be processed, and is a sound to be removed from the sound to be processed. The reference time series is a time series indicating a reference sound. The reference sound is a sound to be removed that has been pre-recorded on a sound recording medium such as a compact disc. Since music data in a predetermined format such as MP3 (MPEG-1 Audio Layer-3) indicates a pre-recorded sound, the reference sound may be a pre-recorded removal target sound in the form of music data. ..

The sound to be processed is, for example, a conversation between two speakers recorded in monaural while a classical song recorded on a compact disc is played from a speaker. In such a case, the target sound is a conversation between two speakers, the sound to be removed is the sound of a classical song flowing from a speaker, and the reference sound is a sound recorded on a compact disc.

The processing target sound time series is a signal in which an analog signal representing the processing target sound is discretized at a predetermined sampling frequency. Therefore, in the processing target sound time series, x (n) in FIG. 1 represents the amplitude x indicated by the nth sample of the processing target sound time series.

The reference time series is a signal in which an analog signal representing a reference sound is discretized at a predetermined sampling frequency. Therefore, in the reference time series, o (n) in FIG. 1 represents the amplitude o indicated by the nth sample of the reference time series.

The processing target sound time series and the reference time series are not always discretized at the same sampling frequency. Rather, in general, the sampling frequencies of the processed sound time series and the reference time series are not the same. This is because even if the sampling frequency set at the time of recording is the same, the sampling frequency deviates from the set value due to a change in the performance of the hardware used for recording such as a vibrator depending on the environment. As described above, the processing target sound time series and the reference time series do not always satisfy the sampling frequency matching condition. The sampling frequency matching condition is a condition that the sampling frequency of the sound time series to be processed and the sampling frequency of the reference time series are the same.

Further, the sound to be removed does not necessarily originate from the timing at which the recording of the sound to be processed is started. Therefore, the position where the removal target sound starts in the processing target sound time series (hereinafter referred to as “removal target sound start position”) does not necessarily coincide with the time origin (that is, n = 0) of the processing target sound time series. Not necessarily. Therefore, the processing target sound time series and the reference time series do not always satisfy the origin matching condition. The origin matching condition is a condition that the start position of the sound to be removed and the time origin of the reference time series match.

A more detailed flow of processing executed by the acoustic signal processing device 1 will be described. The acoustic signal processing device 1 includes a compensated spectrogram acquisition unit 110, a target complex spectrogram estimation unit 120, and a target sound time series generation unit 130.

The compensated spectrogram acquisition unit 110 acquires the processing target sound time series and the reference time series, and acquires the processing target spectrogram and the reference spectrogram based on the processing target sound time series and the reference time series. The processing target spectrogram is a time series of complex spectra of the processing target sound time series that satisfy the sampling frequency matching condition and the origin matching condition. A reference spectrogram is a time series of complex spectra of a reference time series that satisfies the sampling frequency matching condition and the origin matching condition.

The compensated spectrogram acquisition unit 110 acquires the processed spectrogram and the reference spectrogram by executing, for example, a time origin compensation process, an amplitude frequency conversion process, and a sampling frequency compensation process.

The time origin compensation process is a process of matching the start position of the sound to be removed with the time origin of the reference time series. In the time origin compensation process, for example, the time origin of the reference time series is removed by moving the time of the time origin of the reference time series so as to maximize the cross-correlation between the sound time series to be processed and the reference time series. Match the sound start position (see Reference 1). In such a case, the processing target sound time series does not change, but the reference time series changes. The origin matching condition is satisfied by executing the time origin compensation process.

Reference 1: Japanese Unexamined Patent Publication No. 2014-174393

The amplitude frequency conversion process includes a first sub-amplitude frequency conversion process and a second sub-amplitude frequency conversion process. The first sub-amplitude frequency conversion process is a process of acquiring a complex spectrum for each first section period of a plurality of first section periods for the process target sound time series satisfying the origin matching condition. The first division period is a continuous part of the entire period of the processing target sound time series. One first division period does not include the other first division period, and the union of all the first division periods is equal to the entire period of the sound time series to be processed. The process of acquiring a complex spectrum for each first division period of a plurality of first division periods is, for example, a short-time Fourier transform. By executing the first sub-amplitude frequency conversion process, the processing target sound time series satisfying the origin matching condition becomes a time series of the complex spectrum of the processing target sound time series satisfying the origin matching condition (hereinafter referred to as "first spectrogram"). Will be converted.

The second sub-amplitude frequency conversion process is a process of acquiring a complex spectrum for each second division period of a plurality of second division periods for a reference time series satisfying the time origin matching condition. The second division period is a continuous part of the entire period of the reference time series. One second division period does not include the other second division period, and the union of all the second division periods is equal to the entire period of the reference time series. The process of acquiring a complex spectrum for each second division period of a plurality of second division periods is, for example, a short-time Fourier transform. By executing the second sub-amplitude frequency conversion process, the reference time series satisfying the origin matching condition is converted into the time series of the complex spectrum of the reference time series satisfying the origin matching condition (hereinafter referred to as “second spectrogram”).

The sampling frequency compensation process is a process of acquiring a process target spectrogram and a reference spectrogram based on the first spectrogram and the second spectrogram. The sampling frequency compensation process is, for example, a process of applying blind synchronization (a blind compensation audio signal processing method according to Reference 1) to the first spectrogram and the second spectrogram. In blind synchronization, the sampling frequency of the reference time series changes, and the sampling frequency of the processing target sound time series does not change.

In this way, by executing the time origin compensation processing, the amplitude frequency conversion processing, and the sampling frequency compensation processing in this order for the set of the processing target sound time series and the reference time series, the processing target spectrogram and the reference spectrogram can be obtained. Be done.

The symbol of the following equation (1) in FIG. 1 represents a spectrogram to be processed.

The symbol of the equation (1) represents the amplitude and phase of the frequency component of the frequency ω in the complex spectrum represented by the m-th time frame of the spectrogram to be processed (m is an integer of 1 or more).

The symbol of the following equation (2) in FIG. 1 represents a reference spectrogram.

The symbol in equation (2) represents the amplitude and phase of the frequency component of frequency ω in the complex spectrum represented by the m-th time frame of the reference spectrogram. The symbol ε ₀ in the equation (2) represents the difference in sampling frequency between the processing target sound time series and the reference time series.

The target complex spectrogram estimation unit 120 is a time series of the target complex spectrum that maximizes the probability with respect to a predetermined predetermined probability distribution (hereinafter referred to as “reference probability distribution”) based on the processing target spectrogram and the reference spectrogram. To estimate. The target complex spectrum is a complex spectrum obtained as the difference between the complex spectra to be processed and the reference spectrogram multiplied by the frequency transfer function at the same time. Hereinafter, the time series of the target complex spectrum is referred to as a target complex spectrogram.

The target complex spectrum represented by the m-th time frame of the target complex spectrogram is represented by, for example, the following equation (3). The symbol on the left side of equation (3) represents the target complex spectrum.

H (ω) on the right side of the equation (3) represents a frequency transfer function (that is, a function representing the frequency response of the reproduction recording environment of the reference sound).

The reference probability distribution is expressed by, for example, the following equation (4). Equation (4) is a zero-mean generalized complex normal distribution.

α and β are auxiliary variables that determine the shape of the reference probability distribution, and are predetermined values. In particular, α represents the variance. β is an integer of 1 or more. When β = 2, equation (4) represents a normal distribution, and when β = 1, equation (4) represents a Laplace distribution. In equation (4), Γ represents the gamma function.

Hereinafter, the distribution of the appearance probability of the target complex spectrum in the target complex spectrogram is referred to as the target complex spectrum appearance probability distribution. Maximizing the likelihood with respect to a predetermined probability distribution is, for example, the logarithm of the frequency transfer function represented by the following equation (5) when the reference probability distribution is represented by the equation (4). It is to maximize the likelihood function.

The left side of equation (5) is the log-likelihood function of the frequency transfer function. The second term on the right side of the equation (5) represents a predetermined constant. Increasing the left side in the equation (5) means decreasing the size of the negative sign of the first term on the right side. The fact that the negative sign of the first term on the right side of the equation (5) becomes smaller means that the average of the absolute values of S (ω, m) to the β power becomes smaller. Since S (ω, m) becomes smaller as the removal target sound is removed from the processing target sound, the smaller the first term on the right side of the equation (5), the more the time series of the processing target sound in which the influence of the removal target sound is suppressed. Means that is obtained. Therefore, to acquire the objective complex spectrum that maximizes the log-likelihood function of the frequency transfer function represented by Eq. (5), the time series of the processing target sound that suppresses the influence of the removal target sound to the maximum is acquired. Means that.

The target sound time series generation unit 130 generates a time series of sounds having the target complex spectrogram estimated by the target complex spectrogram estimation unit 120 as a target sound time series based on the target complex spectrogram estimated by the target complex spectrogram estimation unit 120. ..

The target sound time series is a time series generated by using the processing target sound time series, and the component of the reference time series is suppressed from the processing target sound time series. Since the reference sound is a pre-recorded sound to be removed, the target sound time series is a time series of sounds in which the components of the sound to be removed are suppressed from the sound to be processed.

FIG. 2 is a diagram showing an example of the functional configuration of the acoustic signal processing device 1 of the embodiment. The acoustic signal processing device 1 includes a control unit 10 including a processor 91 such as a CPU (Central Processing Unit) connected by a bus and a memory 92, and executes a program. The acoustic signal processing device 1 functions as a device including a control unit 10, an input unit 11, a communication unit 12, a storage unit 13, and an output unit 14 by executing a program. More specifically, the processor 91 reads out the program stored in the storage unit 13, and stores the read program in the memory 92. By executing the program stored in the memory 92 by the processor 91, the acoustic signal processing device 1 functions as a device including a control unit 10, an input unit 11, a communication unit 12, a storage unit 13, and an output unit 14.

The control unit 10 controls the operation of various functional units included in the acoustic signal processing device 1. The control unit 10 generates a target sound time series using, for example, a processing target sound time series and a reference time series.

The input unit 11 includes an input device such as a mouse, a keyboard, and a touch panel. The input unit 11 may be configured to include an interface for connecting these input devices to its own device. The input unit 11 receives input of various information to its own device.

The communication unit 12 includes a communication interface for connecting the own device to an external device. The communication unit 12 communicates with the external device to be connected via wired or wireless. The communication unit 12 acquires a reference time series from, for example, an external device. The communication unit 12 acquires the processing target sound time series from, for example, an external device. The external device is, for example, a computer that reproduces music data of a reference sound and transmits time-series data being reproduced to the acoustic signal processing device 1. The external device is, for example, a computer that transmits a time series of sound to be processed. The external device is, for example, a microphone that records the sound time series to be processed in monaural.

The storage unit 13 is configured by using a non-temporary computer-readable storage medium device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 13 stores various information about the acoustic signal processing device 1. The storage unit 13 stores, for example, information indicating a reference probability distribution in advance. The storage unit 13 stores, for example, a frequency transfer function in advance.

The output unit 14 outputs various information. The output unit 14 includes display devices such as a CRT (Cathode Ray Tube) display, a liquid crystal display, and an organic EL (Electro-Luminescence) display. The output unit 14 may be configured as an interface for connecting these display devices to its own device. The output unit 14 outputs, for example, the information input to the input unit 11. The output unit 14 may be configured to include a sound output device such as a speaker. The output unit 14 may be configured as an interface for connecting these sound output devices to the own device.

FIG. 3 is a diagram showing an example of the functional configuration of the control unit 10 in the embodiment. The control unit 10 includes a compensated spectrogram acquisition unit 110, a target complex spectrogram estimation unit 120, a target sound time series generation unit 130, a processing target sound time series acquisition unit 140, a reference time series acquisition unit 150, a communication control unit 160, and output control. A unit 170 and a recording unit 180 are provided.

The processing target sound time series acquisition unit 140 acquires the processing target sound time series input to the communication unit 12. When the processing target sound time series is stored in the storage unit 13 in advance, the processing target sound time series acquisition unit 140 acquires the processing target sound time series by reading the processing target sound time series from the storage unit 13. May be good.

The reference time series acquisition unit 150 acquires the reference time series input to the communication unit 12. When the reference time series is stored in the storage unit 13 in advance, the reference time series acquisition unit 150 may acquire the reference time series by reading the reference time series from the storage unit 13.

The communication control unit 160 and the communication control unit 160 control the operation of the communication unit 12. The output control unit 170 controls the operation of the output unit 14. The recording unit 180 records information in the storage unit 13.

The compensated spectrogram acquisition unit 110 includes a time origin compensation unit 111, an amplitude frequency conversion unit 112, and a sampling frequency compensation unit 113.

The time origin compensation unit 111 executes the time origin compensation process for the processing target sound time series acquired by the processing target sound time series acquisition unit 140 and the reference time series acquired by the reference time series acquisition unit 150.

The amplitude frequency conversion unit 112 executes the amplitude frequency conversion process for the processing target sound time series satisfying the origin matching condition and the reference time series satisfying the origin matching condition. By executing the amplitude frequency conversion process, the processing target sound time series satisfying the origin matching condition is converted into the first spectrogram, and the reference time series satisfying the origin matching condition is converted into the second spectrogram.

The sampling frequency compensation unit 113 executes the sampling frequency compensation process. By executing the sampling frequency compensation process, the spectrogram to be processed and the reference spectrogram are generated based on the first spectrogram and the second spectrogram.

FIG. 4 is a flowchart showing an example of a processing flow executed by the acoustic signal processing device 1 of the embodiment. The processing target sound time series is input to the communication unit 12, and the processing target sound time series acquisition unit 140 acquires the input processing target sound time series (step S101). Next, the reference time series is input to the communication unit 12, and the reference time series acquisition unit 150 acquires the input reference time series (step S102). Next, the compensated spectrogram acquisition unit 110 acquires the processing target sound time series and the reference time series, and acquires the processing target spectrogram and the reference spectrogram based on the processing target sound time series and the reference time series (step S103).

In step S103, the compensated spectrogram acquisition unit 110 acquires the process target spectrogram and the reference spectrogram based on the process target sound time series and the reference time series by, for example, executing the following compensated spectrogram acquisition process. In the compensated spectrogram acquisition process, the time origin compensation unit 111 first executes the time origin compensation process for the processing target sound time series and the reference time series, and sets the start position of the processing target sound time series and the time origin of the reference time series. Match. As a result, the time origin compensation unit 111 acquires the processing target sound time series satisfying the origin matching condition and the reference time series satisfying the origin matching condition.

In the compensated spectrogram acquisition process, the amplitude frequency conversion unit 112 then executes the amplitude frequency conversion process for the processing target sound time series satisfying the origin matching condition and the reference time series satisfying the origin matching condition. By executing the amplitude frequency conversion process, the processing target sound time series satisfying the origin matching condition is converted into the first spectrogram, and the reference time series satisfying the origin matching condition is converted into the second spectrogram.

In the compensated spectrogram acquisition process, the sampling frequency compensation unit 113 then executes the sampling frequency compensation process. By executing the sampling frequency compensation process, the spectrogram to be processed and the reference spectrogram are generated based on the first spectrogram and the second spectrogram.

Following step S103, the target complex spectrogram estimation unit 120 estimates the target complex spectrogram that maximizes the likelihood with respect to the reference probability distribution based on the processed spectrogram and the reference spectrogram (step S104).

Next, the target sound time series generation unit 130 uses the time series of sounds having the target complex spectrogram estimated by the target complex spectrogram estimation unit 120 as the target sound time series based on the target complex spectrogram estimated by the target complex spectrogram estimation unit 120. Generate (step S105).

Next, the output control unit 170 controls the operation of the output unit 14, and outputs the sound indicated by the target sound time series from the output unit 14 (step S106). The target sound time series generated in step S105 may be recorded in the storage unit 13 by the recording unit 180 after step S105.

<Experimental results>
An example of the result of an experiment (hereinafter referred to as "evaluation experiment") in which a time series of sounds in which the influence of noise is reduced by using the acoustic signal processing device 1 of the embodiment is shown is shown.

FIG. 5 is an explanatory diagram illustrating the experimental environment of the evaluation experiment in the embodiment. The sound to be processed and the reference sound for the evaluation experiment are a room of 4.1 x 3.8 x 2.8 cubic meters in which two speakers 901 and 902 and a microphone 903 are installed (hereinafter referred to as "laboratory"). It was recorded by the microphone 903. The sound absorption coefficient of the laboratory was 0.2. The speaker 901 was a sound source that outputs (sounds) the target sound, and the speaker 902 is a sound source that outputs (sounds) the reference sound.

The speaker 902 was installed at a position 100 cm from one of the walls of the laboratory (hereinafter referred to as "vertical reference wall"), and the speaker 901 was installed at a position 140 cm from the vertical reference wall. The distance between the speaker 901 and the speaker 902 in the direction perpendicular to the vertical reference wall was 40 cm. The microphone 903 was installed at a position 120 cm from the vertical reference wall.

The distance between the speaker 901 and the speaker 902 from the lateral reference wall was 120 cm. The horizontal reference wall is one of the walls of the laboratory and is orthogonal to the vertical reference wall. The distance of the microphone 903 from the lateral reference wall was 290 cm. That is, the microphone 903 was installed at a position 120 cm from the wall of the laboratory facing the lateral reference wall.

In the laboratory, recording was performed by the microphone 903 in a state where only the speaker 901 was operated and only the target sound was output (hereinafter referred to as "target sound recording"). The setting value of the sampling frequency of the microphone 903 at the time of recording the target sound was 16000 Hz. Further, in the laboratory, recording was performed by the microphone 903 in a state where only the speaker 902 was operated and only the reference sound was output at a timing different from the target sound recording (hereinafter referred to as "reference sound recording"). The setting value of the sampling frequency of the microphone 903 at the time of recording the reference sound was (16000 + 1) Hz. In the evaluation experiment, the sampling frequency mismatch was simulated by shifting the sampling frequency at the time of recording the target sound and the sampling frequency at the time of recording the reference sound by one.

In the evaluation experiment, the sound recorded by the target sound recording and the sound recorded by the reference sound recording are mixed so that the input SNR (Signal to Noise Ratio) is -5, 0, 5, 10 decibels, and the evaluation experiment is performed. It was used as a sound to be processed in. The input SNR is mixed so as to be -5, 0, 5, and 10 decibels, which means that the input SNR is changed to four conditions of -5 decibels, 0 decibels, 5 decibels, and 10 decibels.

In the evaluation experiment, the short-time Fourier transform was used as the amplitude frequency conversion process. The short-time Fourier transform in the evaluation experiment was performed at 8192 points under the condition that the window length was 4096 points, the shift length was 1/2 of the window length, and the shift length was reduced to zero. A humming window was used as the window for the short-time Fourier transform.

In the evaluation experiment, the sound output from the speakers 901 and 902 is replaced with the time series of the sound recorded by the microphone 903, and the environment similar to that of the laboratory including the input SNR is modeled and generated by numerical simulation such as the finite element method. The acoustic signal processing device 1 was evaluated using the time series. Hereinafter, among the evaluation experiments, an experiment in which the sound output from the speakers 901 and 902 is evaluated by using the time series of the sound recorded by the microphone 903 is referred to as an actual experiment. Hereinafter, among the evaluation experiments, an experiment in which the acoustic signal processing device 1 is evaluated using a time series generated by a numerical simulation is referred to as a simulation experiment.

FIG. 6 is a diagram showing an example of the result of the evaluation experiment in the embodiment. Results R1 to R3 show examples of experimental results of simulation experiments, respectively. Results R4 to R6 each show an example of the experimental results of an actual experiment. The horizontal axis of the results R1 to R6 represents the input SNR, and the vertical axis represents the output SNR. The definition of the output SNR was (10 × log ₁₀ (σ _q ² / σ _o ² )) decibels. σ _q ² is the dispersion of the target sound time series and represents the dispersion of the target sound time series during the period in which the sound to be removed exists. σ _o ² represents the variance of the reference sound recorded by the microphone 903.

Results The bar graphs in R1 to R6 are the results when β = 0.2, 0.4, ..., 2.0 from the left in each graph, respectively. The filled part (Sync. (Off)) is the result when blind synchronization is not performed, and the uncolored part (Sync. (On)) is the result when blind synchronization is performed.

Results R1 to R6 indicate that the output SNR was improved by compensating for the deviation of the sampling frequency. One of the reasons for this is that when there is a deviation in the sampling frequency, the frequency response (frequency transfer function) is apparently not invariant at the time, and the model represented by the equation (3) does not hold. Another reason is that β, which maximizes the output SNR, differs depending on the type of sound to be processed and the input SNR. Therefore, the acoustic signal processing device 1 is a more flexible model than assuming that the probability distribution of the target complex spectrum is a normal distribution or a Laplace distribution, and a higher output SNR can be obtained.

<Method of obtaining the target complex spectrogram using Eq. (5)>
Here, an example of a method for estimating the target complex spectrogram using the equation (5) will be described. The target complex spectrogram is estimated by differentiating the equation (5) with H (ω) and acquiring S (ω, m) which gives a maximum value as an estimation result. However, the derivative value of the power of the absolute value cannot be obtained analytically.

Of course, since only an analytical solution cannot be obtained, the acoustic signal processing apparatus 1 may estimate the target complex spectrogram by a steady method of handling the equation (5) as it is in the numerical calculation and acquiring an approximate value. However, in such a case, since the amount of calculation is large, rounding error or the like is likely to occur, and there is a high possibility that the calculation error will be large. Therefore, instead of numerically calculating the equation (5) as it is, the acoustic signal processing apparatus 1 may transform the equation (5) into an equivalent equation with a smaller amount of calculation to estimate the target complex spectrogram.

As a method of estimating the target complex spectrogram by transforming the equation (5) into an equivalent and less computationally expensive equation, the acoustic signal processing device 1 may obtain the target complex spectrogram by applying, for example, the following auxiliary function. good. The auxiliary function method is also referred to as Majorization-Minimization Algorithm or MM Algorithm (see References 2 and 3).

Reference 2: David R Hunter & Kenneth Lange “A Tutorial on MM Algorithms”, The American Statistician, 58: 1, 30-37.
Reference 3: Ono, "Optimization Algorithm by Auxiliary Function Method and Its Application to Acoustic Signal Processing", Journal of Acoustical Society of Japan, Vol. 68, No. 11, pp. 566-571, 2012

In the auxiliary function method, it is necessary to find an appropriate auxiliary function for the objective function. Here, the objective function is a function to be maximized or minimized given for each optimization problem. Appropriate auxiliary functions in the auxiliary function method can be found, for example, by using the method described in Reference 4.

Reference 4: Nobutaka Ono and Shigeki Miyabe, “Auxiliary-Function-Based Independent
Component Analysis for Super-Gaussian Sources ”, V. Vigneron et al. (Eds.): LVA / ICA 2010, LNCS 6365, pp. 165-172, 2010, Springer-Verlag Berlin Heidelberg 2010

Here, consider the mathematical characteristics of the function G (x), which is a continuously differentiable function expressed by the amplitude x and is an even function of the amplitude x. Therefore, the amplitude x is treated as a variable x that can have a negative value, and the mathematical characteristics of the even function G (x) will be described. If the function that divides the derivative of G (x) by the variable x by the variable x is continuous within the definition range, positive at x> 0, and monotonically decreasing, the following equation (6) is It holds for any x.

The equal sign condition of the formula (6) is expressed by the following formula (7).

The first term of the formula (5) satisfies the condition of the formula (7). Therefore, an appropriate auxiliary function Q in the auxiliary function method expressed by the following equation (8) is expressed by the following equation (9).

H ₀ (ω) is an auxiliary variable. The second term on the right-hand side of equation (9) is a term that depends only on the auxiliary variable H ₀ (ω). Therefore, it is a term irrelevant to the optimization of H (ω). Equation (9) is a quadratic function of H (ω). The following updated equation (11) can be obtained by transforming the equation by substituting H ₀ (ω) = H (ω) ^(k) and setting the equation obtained by differentiating the auxiliary function Q with H (ω) as 0. .. k represents the number of repetitions.

In equation (11), the asterisk symbol indicates that it is a complex conjugate. By substituting the frequency transfer function estimated by the equation (11) into the equation (3), the target complex spectrogram is estimated. The auxiliary function application method is a method of estimating the target complex spectrogram by repeatedly executing the estimation using the equations 811) and (3) until a predetermined end condition is satisfied. The end condition is, for example, a condition that the process is repeated a predetermined number of times. The initial value of the repetition may be any condition as long as the denominator of the equation (11) is not set to 0. For example, the initial value of H (ω) may be a condition that the denominator of the equation (11) is not set to 0 under the condition that it is 1.

The acoustic signal processing apparatus 1 of the embodiment configured as described above has a purpose of minimizing the difference between the distribution of the appearance probability of the target complex spectrum and the reference probability distribution in the target complex spectrogram based on the processed target spectrogram and the reference spectrogram. Estimate a complex spectrogram. The target sound time series is a time series generated by using the processing target sound time series, and the component of the reference time series is suppressed from the processing target sound time series. The reference sound is a pre-recorded sound to be removed. Therefore, the acoustic signal processing device 1 can generate a time series of sounds in which the component of the sound to be removed is suppressed from the sound to be processed as the target sound time series. That is, the acoustic signal processing device 1 can generate a time series of sounds in which the influence of noise is further reduced from the time series of sounds recorded in monaural.

(Modification example)
The acoustic signal processing device 1 may be mounted by using a plurality of information processing devices connected so as to be communicable via a network. In this case, each functional unit included in the acoustic signal processing device 1 may be distributed and mounted in a plurality of information processing devices.

In addition, all or a part of each function of the acoustic signal processing device 1 is realized by ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), FPGA (Field Programmable Gate Array), etc. good. The program may be recorded on a computer-readable recording medium. The computer-readable recording medium is, for example, a flexible disk, a magneto-optical disk, a portable medium such as a ROM or a CD-ROM, or a storage device such as a hard disk built in a computer system. The program may be transmitted over a telecommunication line.

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs and the like within a range that does not deviate from the gist of the present invention.

1 ... Sound signal processing device, 10 ... Control unit, 11 ... Input unit, 12 ... Communication unit, 13 ... Storage unit, 14 ... Output unit, 110 ... Compensated spectrogram acquisition unit, 111 ... Time origin compensation unit, 112 ... Amplitude Frequency conversion unit, 113 ... Sampling frequency compensation unit, 120 ... Target complex spectrogram estimation unit, 130 ... Target sound time series generation unit, 140 ... Processing target sound time series acquisition unit, 150 ... Reference time series acquisition unit, 160 ... Communication control Unit, 170 ... Output control unit, 180 ... Recording unit, 91 ... Processor, 92 ... Memory

Claims (5)

The origin that the start position of the removal target sound in the processing target sound time series that indicates the monaurally recorded processing target sound and the time origin of the reference time series that indicates the pre-recorded removal target sound match. A processing target spectrogram that is a time series of a complex spectrum of the processing target sound time series that satisfies the matching condition and the sampling frequency matching condition that the sampling frequency of the processing target sound time series and the sampling frequency of the reference time series match. , The compensated spectrogram acquisition unit for acquiring the reference spectrogram, which is the time series of the complex spectrum of the reference time series, which satisfies the origin matching condition and the sampling frequency matching condition.
The probability of the objective complex spectrum, which is the complex spectrum in the objective complex spectrogram obtained as the difference between the complex spectra at the same time time between the processing target spectrogram and the reference spectrogram multiplied by the frequency transfer function, with respect to a predetermined probability distribution. Purpose of estimating the time series that maximizes the degree The complex spectrogram estimator and
An acoustic signal processing device. A target sound time series generator that generates a time series of sounds having the target complex spectrogram estimated by the target complex spectrogram estimation unit,
The acoustic signal processing apparatus according to claim 1. The compensated spectrogram acquisition unit is a time series of a complex spectrum of the processed sound time series that satisfies the origin matching condition after matching the start position of the processed sound time series with the time origin of the reference time series. The 1 spectrogram and the 2nd spectrogram which is a time series of the complex spectrum of the reference time series satisfying the origin matching condition are acquired, and the processed target spectrogram and the reference spectrogram are acquired using the 1st spectrogram and the 2nd spectrogram. do,
The acoustic signal processing apparatus according to claim 1 or 2. The origin that the start position of the removal target sound in the processing target sound time series that indicates the monaurally recorded processing target sound and the time origin of the reference time series that indicates the pre-recorded removal target sound match. A processing target spectrogram that is a time series of a complex spectrum of the processing target sound time series that satisfies the matching condition and the sampling frequency matching condition that the sampling frequency of the processing target sound time series and the sampling frequency of the reference time series match. , The compensated spectrogram acquisition step for acquiring the reference spectrogram, which is the time series of the complex spectrum of the reference time series, which satisfies the origin matching condition and the sampling frequency matching condition.
The probability of the objective complex spectrum, which is the complex spectrum in the objective complex spectrogram obtained as the difference between the complex spectra at the same time time between the processed spectrogram and the reference spectrogram multiplied by the frequency transfer function, with respect to a predetermined probability distribution. Purpose of estimating the time series that maximizes the degree Complex spectrogram estimation step and
Acoustic signal processing method having. A program for operating a computer as the acoustic signal processing device according to any one of claims 1 to 3.

Images