JP2016142914A - Wiener filter design device, speech enhancement device, wiener filter design method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a Wiener filter design device capable of designing a Wiener filter realizing speech enhancement that only aimed sound in a narrow range is collected.SOLUTION: The Wiener filter design device comprises: a sudden PSD estimation unit 203 which estimates power spectral density of sudden target sound on the basis of time change of spot power spectral density of an observation signal; and a Wiener filtering unit 204 which designs a Wiener filter using the power spectral density of sudden target sound.SELECTED DRAWING: Figure 4

Description

  The present invention relates to a Wiener filter design device that designs a filter for clearly extracting a specific sound source signal, a speech enhancement device that clearly extracts a specific sound source signal, a Wiener filter design method, and a program.

  For example, Non-Patent Document 1 is a speech enhancement method that combines a beam forming technique for enhancing an arbitrary direction and a nonlinear Wiener filter. The method of Non-Patent Document 1 is called a speech enhancement method based on local PSD (power spectral density) estimation.

<Speech enhancement device 10 of Non-Patent Document 1 (speech enhancement method based on local PSD estimation)>
Hereinafter, a speech enhancement device of Non-Patent Document 1 that realizes speech enhancement based on local PSD estimation will be described with reference to FIGS. 1 and 2. FIG. 1 is a block diagram illustrating a configuration of a speech enhancement device 10 of Non-Patent Document 1. FIG. 2 is a flowchart showing the operation of the speech enhancement device 10 of Non-Patent Document 1. As shown in FIG. 1, the speech enhancement device 10 of Non-Patent Document 1 includes a microphone array 100 configured with M (M ≧ 2, m = 1,..., M) microphone elements, a frequency domain conversion unit 101, A beam forming unit 102, a local PSD estimation unit 103, a Wiener filtering unit 104, and a time domain conversion unit 105 are included. The microphone array 100 observes M channel observation signals from K (K ≧ 1, k = 1,..., K) sound sources (S100). The frequency domain conversion unit 101 converts the observation signal from the microphone array 100 into the frequency domain by means such as FFT (S101). When the transfer characteristic between the m-th microphone and the k-th sound source is expressed as Am _{, k} (ω), and the k-th sound source signal is expressed as S _k (ω, τ), the m-th received signal is Modeled with

Here, ω represents a frequency, and τ represents a frame. The beam forming unit 102 outputs a voice enhanced for each direction of arrival by multiplying the observation signal converted into the frequency domain by a linear filter (S102). The beam forming unit 102 includes not only a filter that emphasizes the target sound but also a total of L (L ≧ 2, l = 1,..., L) filters for estimating the PSD of noise, Calculate multiple filter outputs. When the l-th filter has a sensitivity characteristic that emphasizes the direction θ _l , and the L filters for each arrival direction are applied, the l-th output of the beamforming unit 102 is calculated by the following equation.

W _{l, m} (ω) is the filter coefficient of the l-th beam forming, and D _{l, k} (ω) is the sensitivity of the l-th beam forming to the k-th sound source. Assuming that the sound source signals are uncorrelated with each other, the PSD of the l-th beamforming output is modeled by the following equation.

Here, <•> represents an expected value operator, and φ _Sk (ω) represents the PSD of the kth sound source signal. Assuming that the relationship of equation (3) holds true for the sound source signal group (local sound source signal) coming from the local space and the beamforming output, φ _Yθl (ω) is expressed by equation (4).

Here, D _{l, θl} (ω) represents the average sensitivity of beam forming with respect to the local space with the direction θ _l as the center, and φ _Sθl (ω) represents the PSD (local PSD) of the l-th local sound source signal. The relationship between L φ _Sθl (ω) and φ _Yθl (ω) is expressed by the following equation.

Hereinafter, the non-negative matrix D (ω) is referred to as a sensitivity matrix. The local PSD estimation unit 103 estimates L local PSDs (per arrival direction) by solving the inverse problem of Equation (5) (S103). Specifically, the local PSD estimation unit 103 estimates local PSDΦ ^ _S (ω, τ) for each frame as shown in Equation (6) in order to improve noise suppression performance.

here,

It is. Hereinafter, G (ω) is referred to as a PSD estimation matrix. The Wiener filtering unit 104 uses the estimated local PSD to design a Wiener filter H _l (ω, τ) for enhancing the l-th local sound source signal based on the equation (8) (S104).

The Wiener filtering unit 104 calculates an output signal Z _θl (ω, τ) in which noise is suppressed based on Expression (9) using a Wiener filter designed using the estimated local PSD (S104).

The time domain conversion unit 105 converts the frequency domain signal into the time domain by means of IFFT or the like, and outputs the converted signal (S105).

K. Niwa, Y. Hioka, and K. Kobayashi, “Post-filter design for speech enhancement in various noisy environments,” in Proc. IWAENC 2014, pp. 36-40, 2014.

  With reference to FIG. 3, the problem in the speech enhancement method based on local PSD estimation will be considered. FIG. 3 is a diagram for considering a problem in the speech enhancement method based on local PSD estimation. Now, consider applying the speech enhancement method based on local PSD estimation to signals observed using about 2 to 4 microphones. In this case, it is known that speech enhancement can be performed so as to separate a target local space having an angular width of about 60 to 80 degrees from other local spaces. However, with this method, it is difficult to distinguish and collect sound sources coming from a range narrower than the angular width of the target local space from other sound sources. For example, as shown in FIG. 3, when it is desired to separate a sound (for example, kick sound) and other noise (for example, cheers of surrounding supporters) in a soccer field by using a microphone array 100 composed of about 2 to 4 microphones. think about. In this configuration, the supporter's cheering 5b mixed in the other local space 8 (dark dot-hatched region, space) can be suppressed, but the target local space 7 (horizontal line hatched region) including the target player ), It was difficult to suppress the supporter's cheering 5f mixed at a high level. Therefore, it is difficult to extract only the target sound in a narrow local space 6 (a thin dot-hatched area or space) including only the target player.

  Therefore, an object of the present invention is to provide a Wiener filter design device capable of designing a Wiener filter that realizes speech enhancement in which only a target sound in a narrow range is clearly picked up.

  The Wiener filter design apparatus of the present invention includes a sudden PSD estimation unit and a Wiener filtering unit.

  The sudden PSD estimation unit estimates the power spectral density of the target sound having suddenness based on the temporal change of the local power spectral density of the observation signal. The Wiener filtering unit designs a Wiener filter using the power spectral density of the target sound having suddenness.

  According to the Wiener filter design apparatus of the present invention, it is possible to design a Wiener filter that realizes speech enhancement in which only targeted sounds in a narrow range are clearly picked up.

The block diagram which shows the structure of the audio | voice emphasis apparatus of a nonpatent literature 1. FIG. 10 is a flowchart showing the operation of the speech enhancement device of Non-Patent Document 1. The figure which considers the subject in the speech enhancement method based on local PSD estimation. 1 is a block diagram illustrating a configuration of a speech enhancement device and a Wiener filter design device according to Embodiment 1. FIG. 3 is a flowchart showing the operation of the speech enhancement apparatus and Wiener filter design apparatus according to the first embodiment. FIG. 6 is a diagram for explaining an example of operation in the sudden PSD estimation unit according to the first embodiment. The figure which illustrates notionally the difference of the conventional local PSD estimation and the sudden PSD estimation of a present Example. The block diagram which shows the structure of the audio | voice emphasis apparatus of Example 2, and a Wiener filter design apparatus. 9 is a flowchart showing the operation of the speech enhancement apparatus and Wiener filter design apparatus according to the second embodiment.

  Hereinafter, embodiments of the present invention will be described in detail. In addition, the same number is attached | subjected to the structure part which has the same function, and duplication description is abbreviate | omitted.

  Hereinafter, the speech enhancement device 20 and the Wiener filter design device 2 according to the first embodiment will be described with reference to FIGS. 4 and 5. FIG. 4 is a block diagram illustrating the configuration of the speech enhancement device 20 and the Wiener filter design device 2 according to the present embodiment. FIG. 5 is a flowchart showing the operations of the speech enhancement device 20 and the Wiener filter design device 2 of this embodiment.

  In conventional array signal processing, the target sound and other noises are separated and collected by using only spatial information. In this method, unless the number of microphones is increased, it is difficult to improve spatial resolution and to clearly collect a target sound in a narrow range. In the present invention, attention has been paid to the fact that the target sound is often a sudden sound (for example, a soccer kick sound in the example of FIG. 3). As used herein, the term “sudden sound” or “sudden sound” refers to an acoustic signal that is a precursor observed over a certain time period preceding the sound in terms of power spectral density. In addition, a sound that produces a change amount (increase in level) above a certain threshold in an arbitrary frequency band, and a steep time change so that the change amount decreases in a short time width that is less than the predetermined time width. Refers to the signal. For example, a sound whose level difference in power spectral density between the previous time frame and the current time frame has a change (increase and decrease) of 6 dB or more in an arbitrary frequency band. The larger the change amount, the higher the suddenness. In the present embodiment, the speech enhancement device 20 that can clearly collect the target sound in a narrow range (suddenly) based on the time information without increasing the number of microphones, and the Wiener filter used for such speech enhancement are provided. The Wiener filter design device 2 that can be designed was realized. As shown in FIG. 4, the speech enhancement apparatus 20 of the present embodiment includes a microphone array 100 configured with M (M ≧ 2, m = 1,..., M) microphone elements, a frequency domain conversion unit 101, a beam. It includes a forming unit 102, a local PSD estimation unit 103, an abrupt PSD estimation unit 203, a Wiener filtering unit 204, and a time domain conversion unit 105. The configuration other than the abrupt PSD estimation unit 203 and the Wiener filtering unit 204 is described in Non-Patent Document 1. This is the same as the speech enhancement apparatus 10. Of the constituent requirements of the speech enhancement device 20, the Wiener filter design device 2 may be configured as another device including the sudden PSD estimation unit 203 and the Wiener filtering unit 204. In this case, the first device including the configuration requirements of the microphone array 100, the frequency domain conversion unit 101, the beam forming unit 102, and the local PSD estimation unit 103, the Wiener filter design device 2 (second device), and the time domain conversion The speech enhancement device 20 may be configured by being connected to a third device including the unit 105. Alternatively, the first device and the third device described above are configured as one device (fourth device), and the fourth device and the Wiener filter design device 2 are connected to form the speech enhancement device 20. It is good. Hereinafter, only configuration requirements different from the speech enhancement device 10 of Non-Patent Document 1 will be described.

<Sudden PSD Estimation Unit 203>
The sudden PSD estimation unit 203 estimates the PSD (sudden PSD) of the target sound having suddenness based on the temporal change of the local power spectral density for each arrival direction of the observation signal (S203). More specifically, the sudden PSD estimation unit 203 converts the target sound PSDφ ^ _Sθl (ω, τ) estimated using only spatial information into the sudden PSDφ ^ _SSθl (ω, τ) and other local PSDφ. ^ _Divide into _SNθl (ω, τ) and output. An example of operation in the sudden PSD estimation unit 203 will be described with reference to FIGS. 6 and 7. FIG. 6 is a diagram illustrating an example of operation in the sudden PSD estimation unit 203 of the present embodiment. FIG. 7 is a diagram conceptually illustrating the difference between the conventional local PSD estimation and the sudden PSD estimation of the present embodiment.

FIG. 6 is a graph in which the output level of local PSD observed during a certain soccer game is plotted with the horizontal axis representing time and the vertical axis representing local PSD output level. In the example of FIG. 6, the observed signal includes not only the (sudden) target sound but also noises coming from other directions. It is assumed that the target sound (suddenness, for example, kick sound) at the output level p1 is observed at time t1 in FIG. 6, and the time interval from immediately before the time t1 when the target sound is observed until the predetermined number of frames is ts1. The maximum output level in the time interval ts1 is ps1. Further, it is assumed that the maximum value of the output level (for example, the maximum value such as a cheer) is observed at the output level p2 at time t2 in FIG. 6, and the time interval from the time immediately before time t2 to the predetermined number of frames is ts2. The maximum value of the output level in the time interval ts2 is set to ps2. As shown in FIG. 7, according to the local PSD estimation unit 103, PSD (hereinafter, φ ^ _Sθl (ω, Estimate by dividing PSD ( _{denoted as} Σ ^L _{i = 1, i ≠ l} φ ^ _Sθi (ω, τ)) in other spaces (dark dot-hatched space, region) It is possible, but the PSD (hereinafter referred to as φ ^ _SSθl (ω, τ)) of the space containing the (sudden) target sound (lightly dot-hatched space, region) and 60 to 80 except this It is difficult to estimate the PSD (hereinafter referred to as φ ^ _SNθl (ω, τ)) of a local space (space hatched in the right figure). Therefore, the sudden PSD estimation unit 203 determines that a sudden signal has arrived based on the time change of the signal (described later), and _changes φ ^ _Sθl (ω, τ) to φ ^ _SSθl (ω, τ) and φ ^ _Divide into _SNθl (ω, τ) (S203). As shown in FIG. 6, the sudden PSD estimation unit 203 generates a local PSD maximum value (for example, ps1, ps2) in a certain interval (for example, ts1, ts2) immediately before the target time, and a local PSD (for example, ps1, ps2). For example, the ratio Λ (ω, τ) of p1, p2) is calculated, and when Λ (ω, τ) exceeds a certain threshold Thr (ω), it is determined that a sudden signal has arrived (S203). Λ (ω, τ) can be expressed as follows:

Here, Δ represents a frame index of a certain fixed section (for example, ts1, ts2) up to immediately before the target time, and a section of about 3 to 5 seconds may be set depending on the state of the sound field. Note that the abrupt PSD estimation unit 203 is suitable because when comparing Λ (ω, τ) and Thr (ω), using a value averaged in a frequency band having a certain width makes the noise more robust. is there. The frequency width may be selected by selecting a band in which the value of Λ (ω, τ) tends to appear large. Hereinafter, the comparison operation in the frequency band is simply expressed as Λ (ω, τ) <Thr (ω).

(i) In the case of Λ (ω, τ) <Thr (ω) In this case, the sudden PSD estimation unit 203 determines that no sudden sound source is detected,

And
(ii) When Λ (ω, τ) ≧ Thr (ω) In this case, the sudden PSD estimation unit 203 determines that a sudden sound source is detected,

Perform the split as Here, mean is an operator for calculating an average value.

<Wiener filtering unit 204>
The Wiener filtering unit 204 designs a Wiener filter using the PSD of the target sound having suddenness (sudden PSD, φ ^ _SSθl (ω, τ)) (S204). More specifically, the Wiener filtering unit 204 designs a Wiener filter of the following equation using the three types of PSD described above (S204).

The Wiener filtering unit 204 suppresses noise or emphasizes a sudden target sound by using the Wiener filter designed by Expression (11) for Y _θl (ω, τ) that is the output of the beamforming unit 102. (S204).

<Effects of the speech enhancement device 20 and the Wiener filter design device 2 of this embodiment>
In a conventional speech enhancement device using about 2 to 4 microphones, the target sound and other noises are separated using only spatial information. For this reason, it is difficult to clearly collect only the targeted sound in a narrow range. According to the speech enhancement apparatus 20 of the present embodiment, by using the property of the target sound (the property of high suddenness), only the targeted sound in a narrow range can be clearly seen without increasing the number of microphones. Can collect sound. Further, according to the Wiener filter design device 2 of the present embodiment, it is possible to design a Wiener filter that realizes speech enhancement in which only a target sound in a narrow range is clearly collected without increasing the number of microphones.

  Hereinafter, the speech enhancement device 30 and the Wiener filter design device 3 according to the second embodiment will be described with reference to FIGS. 8 and 9. FIG. 8 is a block diagram illustrating the configuration of the speech enhancement device 30 and the Wiener filter design device 3 according to the present embodiment. FIG. 9 is a flowchart showing the operations of the speech enhancement device 30 and the Wiener filter design device 3 of the present embodiment. As shown in FIG. 8, the speech enhancement apparatus 30 according to the present exemplary embodiment includes a microphone 300, a frequency domain conversion unit 101, an abrupt PSD estimation unit 203, a Wiener filtering unit 304, and a time domain conversion unit 105. is there. The Wiener filter design device 3 according to the present exemplary embodiment includes a sudden PSD estimation unit 203 and a Wiener filtering unit 304. The speech enhancement apparatus 30 according to the present embodiment replaces the microphone array 100 according to the conventional example and the first embodiment with a single microphone 300, so that the beam forming unit 102 and the local PSD estimation unit 103 according to the conventional example and the first embodiment are replaced with each other. This is an omitted example. The operation of the frequency domain transforming unit 101 (S101), the operation of the sudden PSD estimating unit 203 (S203), and the operation of the time domain transforming unit 105 (S105) are as follows. Since the operation is the same as that of the first embodiment except that the number of filters is eliminated, the description is omitted. Unlike the first embodiment, the Wiener filtering unit 304 multiplies the frequency-transformed received sound signal X by a Wiener filter (S304).

  According to the speech enhancement device 30 of the present embodiment, the target sound (sudden target sound) in a narrow range can be obtained even when the sound is picked up by only one microphone by focusing on the time information. Only can be clearly picked up. Further, according to the Wiener filter design apparatus 3 of the present embodiment, even when sound is picked up with only one microphone, only a target sound (sudden target sound) in a narrow range is clearly collected. It is possible to design a Wiener filter that achieves sound enhancement.

<Supplementary note>
The apparatus of the present invention includes, for example, a single hardware entity as an input unit to which a keyboard or the like can be connected, an output unit to which a liquid crystal display or the like can be connected, and a communication device (for example, a communication cable) capable of communicating outside the hardware entity. Can be connected to a communication unit, a CPU (Central Processing Unit, may include a cache memory or a register), a RAM or ROM that is a memory, an external storage device that is a hard disk, and an input unit, an output unit, or a communication unit thereof , A CPU, a RAM, a ROM, and a bus connected so that data can be exchanged between the external storage devices. If necessary, the hardware entity may be provided with a device (drive) that can read and write a recording medium such as a CD-ROM. A physical entity having such hardware resources includes a general-purpose computer.

  The external storage device of the hardware entity stores a program necessary for realizing the above functions and data necessary for processing the program (not limited to the external storage device, for example, reading a program) It may be stored in a ROM that is a dedicated storage device). Data obtained by the processing of these programs is appropriately stored in a RAM or an external storage device.

  In the hardware entity, each program stored in an external storage device (or ROM or the like) and data necessary for processing each program are read into a memory as necessary, and are interpreted and executed by a CPU as appropriate. . As a result, the CPU realizes a predetermined function (respective component requirements expressed as the above-described unit, unit, etc.).

  The present invention is not limited to the above-described embodiment, and can be appropriately changed without departing from the spirit of the present invention. In addition, the processing described in the above embodiment may be executed not only in time series according to the order of description but also in parallel or individually as required by the processing capability of the apparatus that executes the processing. .

  As described above, when the processing functions in the hardware entity (the apparatus of the present invention) described in the above embodiments are realized by a computer, the processing contents of the functions that the hardware entity should have are described by a program. Then, by executing this program on a computer, the processing functions in the hardware entity are realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used. Specifically, for example, as a magnetic recording device, a hard disk device, a flexible disk, a magnetic tape or the like, and as an optical disk, a DVD (Digital Versatile Disc), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc Read Only). Memory), CD-R (Recordable) / RW (ReWritable), etc., magneto-optical recording medium, MO (Magneto-Optical disc), etc., semiconductor memory, EEP-ROM (Electronically Erasable and Programmable-Read Only Memory), etc. Can be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In this embodiment, a hardware entity is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.

Claims (6)

Based on the temporal change of the local power spectral density of the observation signal, the sudden PSD estimation unit for estimating the power spectral density of the target sound having suddenness,
A Wiener filter design apparatus including a Wiener filtering unit that designs a Wiener filter using the power spectrum density of the target sound having suddenness. A microphone array composed of a plurality of microphone elements;
Based on the temporal change of the local power spectral density of the observation signal from the microphone array, the sudden PSD estimation unit for estimating the power spectral density of the target sound having suddenness,
A speech enhancement apparatus including a Wiener filtering unit that designs a Wiener filter using the power spectrum density of the target sound having suddenness and emphasizes the target sound having suddenness using the designed Wiener filter. The speech enhancement apparatus according to claim 2,
A frequency domain converter that converts the observation signals from the microphone array into a frequency domain;
A beam forming unit that enhances the converted observation signal for each direction of arrival;
A local PSD estimation unit for estimating a local power spectral density for each direction of arrival;
The sudden PSD estimator is
A speech enhancement device that estimates the power spectrum density of the target sound having the suddenness based on a temporal change of the local power spectrum density for each arrival direction. Sudden PSD estimation step for estimating the power spectral density of the target sound having suddenness based on the temporal change of the local power spectral density of the observation signal;
A Wiener filter design method including a Wiener filtering step of designing a Wiener filter using a power spectral density of a target sound having suddenness.   A program that causes a computer to function as the Wiener filter design device according to claim 1.   A program for causing a computer to function as the speech enhancement apparatus according to claim 2 or 3.

Images