JP2015070315A - Sound source separation device, sound source separation method, and sound source separation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To separate a target sound arriving from a sound source in a specific direction, regardless of a position of a sound source generating non-target sounds.SOLUTION: A sound source separation device 100 is configured to separate between a target sound arriving from a sound source which is present in a target direction, and non-target sounds arriving from sound sources which are present in any other directions than the target direction. The sound source separation device 100 includes: unidirectional formation parts 31 and 32 which use acoustic signals Xand Xoutputted by microphones M1 and M2 to generate two acoustic signals Aand Ahaving unidirectionality so as to invert the dead angles thereof; non-target signal extraction parts 41 and 42 which perform spectrum subtraction using the generated two acoustic signals having unidirectionality and extract signals Nand Nof non-target sounds arriving from the dead angles; and a target signal extraction part 51 which performs spectrum subtraction using the extracted signals Nand Nof non-target sounds and extracts a signal Y of the target sound.

Description

  The present invention relates to a sound source separation device, a sound source separation method, and a sound source separation program, and is particularly applicable to a case where only a sound source in a specific direction is separated and extracted in an environment where a plurality of sound sources exist.

Conventionally, as a technique for separating and extracting only a sound signal (hereinafter referred to as a target sound) in a specific direction (hereinafter referred to as a target sound) in an environment where a plurality of sound sources exist, a beam former (Beam Former; “BF”). BF is a technique for forming directivity using the time difference between signals reaching each microphone (see Non-Patent Document 1).
BF is roughly divided into two types, an addition type and a subtraction type. In particular, the subtraction type BF has an advantage that directivity can be formed with a smaller number of microphones than the addition type BF. FIG. 6 is a block diagram showing a configuration related to the subtraction type BF when the number of microphones is two.

  The subtraction type BF first calculates a time difference between signals that sound signals from a certain sound source arrive at each microphone by a delay unit, and adds a delay of the calculated time difference to a signal that arrives at one microphone, thereby generating sound from the sound source. Match the signal phase. Hereinafter, the direction in which the sound source exists is referred to as “sound source direction”.

The time difference is calculated by the following equation (1).
τ = (d · sin θ) / c (1) where “d” is the distance between the microphones (m) and “c”. Is the speed of sound (m / s), and “τ” is the delay time (s). “Θ” is an angle (°) between a perpendicular to a straight line connecting the microphones and a sound source direction.

When the sound source direction exists in the direction of the first microphone M1 with respect to the center of the first microphone M1 and the second microphone M2, a delay process is performed on the input x ₁ (t) of the _first microphone M1. . As a result, the input of the first microphone M1 after the delay processing is “x ₁ (t−τ)”. Thereafter, the subtraction type BF performs subtraction processing by a subtracter according to the following equation (2).
a (t) = x ₂ (t) −x ₁ (t−τ) (2)

The subtraction process can be similarly performed in the frequency domain. Here, it is known that the Fourier transform of a signal whose time axis is delayed by τ is obtained by multiplying the result of Fourier transform of the original signal by “exp (−jωτ)”. In that case, the equation (2) is changed to the following equation (3).
A (ω) = X ₂ (ω) − [exp (−jωτ)] X ₁ (ω) (3)

  Here, in the case of θ = ± π / 2, the formed directivity is a cardioid unidirectionality as shown in FIG. Further, in the case of θ = 0, π, the “8” -shaped bidirectionality as shown in FIG. Here, a filter that forms unidirectionality from an input signal is referred to as “unidirectional filter”, and a filter that forms bidirectionality is referred to as “bidirectional filter”.

  Using these techniques, techniques for forming stronger directivity in the target direction have been developed. For example, Patent Document 1 discloses a method of forming directivity in two overlapping regions of unidirectivity (shaded portions in FIG. 7C) formed so that the blind spot is directed left and right with respect to the target direction. Proposed. In this method, an adaptive filter is formed using the difference between the outputs of the left and right unidirectional filters, and components included in both directivities are extracted.

Patent Document 2 proposes a method of forming strong directivity in the target direction by using two types of directivity, unidirectionality and bi-directionality. In this method, first, unidirectionality in which a blind spot is directed to the left and right with respect to the target direction and bi-directionality in which the blind spot is directed to the target direction are formed. After that, the larger one of the outputs of the two unidirectional filters is selected, and spectral subtraction (hereinafter referred to as “SS”) is performed on the bi-directional filter output from the selected unidirectional filter output. Thus, the sound existing in the direction other than the target direction (hereinafter, non-target sound) is suppressed and the target sound is extracted.
Thus, if these existing technologies are used, a strong directivity can be formed in the target direction.

Japanese Patent Laid-Open No. 11-205900 JP 2008-131183 A

Asano Tadashi, "Sound Array Signal Processing", Corona Publishing, February 25, 2011 p. 70-p. 79

  However, as shown in FIG. 3, in the situation where the microphones M1 and M2 are arranged in the middle of the two speakers (sound sources), if the same sound (non-target sound) is reproduced from the left and right speakers L and R at the same time, The existing technology has a problem that it is impossible to separate and extract only the voice (target sound) of the speaker located in front.

  In this situation, the two microphones M1 and M2 pick up exactly the same sound, and the left and right unidirectivity includes not only the target sound component but also the reproduced sound of the speakers L and R at the same time. It will be. Further, since not only the target sound but also the reproduced sound of the speakers L and R have the same phase, the bidirectional sound does not include both the target sound and the reproduced sound of the speakers L and R.

That is, in the technique of Patent Document 1, since the reproduction sound of the speaker is included in either unidirectionality, the difference becomes 0 and learning is not performed, and the reproduction sound of the speaker is output as it is together with the target sound. Become.
In addition, in the technique of Patent Document 2, in addition to including the reproduced sound of the speaker regardless of which of the left and right unidirectional filters is selected, there is no time difference between the reproduced sounds of the speakers picked up by the left and right microphones. Therefore, the reproduced sound of the speaker cannot be extracted even if the bidirectionality is formed. Therefore, the reproduced sound of the speaker cannot be suppressed even after performing spectral subtraction (SS).

  The present invention has been made in view of the above problems, and a sound source separation device and a sound source that can separate a target sound that arrives from a sound source in a specific direction regardless of the position of the sound source that generates the non-target sound. It is an object to provide a separation method and a sound source separation program.

In order to solve the above-described problem, the sound source separation device according to the present invention is an object that reaches from a sound source that exists in the target direction among sounds that reach a plurality of microphones located in a plane whose normal direction is the target direction. A sound source separation device (100) that separates a sound from a non-target sound that arrives from a sound source that exists in the other direction, and an acoustic signal (X) output from any two of the plurality of microphones ₁ , X ₂ ), an acoustic signal (A _R , A _L ) having a unidirectional directivity in which the blind spot is directed in any of the directions connecting the two microphones is set so that the blind spot is reversed. Unidirectionality forming units (31, 32) to be generated, and acoustic signals (X ₁ , X ₂ ) output from any one of the plurality of microphones, or acoustic signals output from two or more microphones are averaged did No. respect (X _DS), said unidirectional said generated by forming unit with two acoustic signals having a unidirectional performs spectral subtraction, a non-target sound signals arriving from the respective blind spot ( N _UDL , N _UDR ) for extracting non-target signals (41, 42), acoustic signals (X ₁ , X ₂ ) output from any one of the plurality of microphones, or output from two or more microphones Spectrum signal subtraction is performed on the signal (X _DS ) obtained by averaging the sound signals to be _processed using the non-target sound signals (N _UDL , N _UDR ) extracted by the non-target signal extraction unit, and the target sound signal (Y And a target signal extraction unit (51) for extracting. However, the reference numerals in parentheses are examples.

  According to the present invention, it is possible to separate a target sound that arrives from a sound source in a specific direction regardless of the position of the sound source that generates the non-target sound.

It is a block diagram which shows the structure of the sound source separation apparatus which concerns on embodiment. It is a figure for demonstrating the process in which the sound source separation apparatus which concerns on embodiment forms the sharp directivity in a target direction, FIG.2 (a) is a figure for demonstrating the process of a unidirectional formation part, FIG. 2B is a diagram for explaining the processing of the non-target signal extracting unit, and FIG. 2C is a diagram for explaining the processing of the target signal extracting unit. It is a figure which shows an example of the usage condition of the sound source separation apparatus which concerns on embodiment. It is a figure for demonstrating the result of the performance evaluation experiment of the sound source separation device which concerns on embodiment, and the conventional sound source separation device by bidirectionality. FIG. 5A is a diagram for explaining a configuration of a sound source separation device according to a modified example. FIG. 5A is a diagram illustrating an arrangement of microphones included in the sound source separation device according to the modification, and FIG. It is a figure which shows the example of a combination of the microphone for forming directivity. It is a block diagram which shows the structure of subtraction type BF in case the number of microphones in a prior art example is two. It is a figure which shows the directional characteristic formed by subtraction type BF in case the number of microphones in a prior art example is two.

DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the drawings as appropriate.
Each figure is only schematically shown so that the invention can be fully understood. Therefore, the present invention is not limited to the illustrated example. In the drawings to be referred to, dimensions of members constituting the present invention may be exaggerated for clarity of explanation. In addition, in each figure, about the same component or the same component, the same code | symbol is attached | subjected and those overlapping description is abbreviate | omitted.

[Embodiment]
<< Configuration of Sound Source Separation Device According to Embodiment >>
With reference to FIG. 1, the structure of the sound source separation device according to the embodiment will be described. FIG. 1 is a block diagram illustrating a configuration of a sound source separation device 100 according to the embodiment.
The sound source separation device 100 arrives from a target sound that arrives from a sound source that exists in a specific direction (target direction), and a sound source that exists in other directions, among the sounds that arrive from a plurality of sound sources that exist in the vicinity. Separates non-target sounds.

  The sound source separation device 100 is mounted on, for example, a multi-function mobile phone (smart phone) having a videophone function, and enables improvement of call sound quality when talking under noise using the videophone function. The sound source separation device 100 is also effective for a video conference system, an interphone, and the like. The sound source separation device 100 is mounted on a car navigation system having a voice search function, and prevents erroneous recognition of information to be searched when searching for information under noise using the voice search function.

As shown in FIG. 1, the sound source separation device 100 according to the embodiment includes a first microphone M1, a second microphone M2, signal input units 11 and 12, a signal addition unit 21, and a unidirectional formation. Units 31, 32, non-target signal extraction units 41, 42, and a target signal extraction unit 51.
The sound source separation device 100 may be configured exclusively by a combination of discrete components, a semiconductor chip, or the like, or may be realized by a program execution process by a CPU (Central Processing Unit). Good. That is, the method for realizing the sound source separation device 100 is not particularly limited. Below, each component is demonstrated in detail.

(Microphone)
The first microphone M1 and the second microphone M2 convert sound (sound wave) into an acoustic signal. Here, the perpendicular direction to the front side with respect to the line connecting the two microphones M1 and M2 is “0 °”, the clockwise direction is represented by a positive angle, and the counterclockwise direction is represented by a negative angle. . Further, the front (0 °) is assumed as the target direction. That is, the first microphone M1 and the second microphone M2 are respectively arranged horizontally, and the perpendicular of the axis connecting the first microphone M1 and the second microphone M2 is parallel to the target direction. In other words, the normal direction of the vertical plane including the axis connecting the first microphone M1 and the second microphone M2 is the target direction.

The microphones M1 and M2 are omnidirectional (omnidirectional). For this reason, the microphones M1 and M2 collect non-target sound that arrives from other than the target direction in addition to target sound that arrives from the target direction. The microphones M1 converts the target sound picked up and non-target sound to the acoustic signals x _1, and outputs the acoustic signals x ₁ to the signal input unit 11. Further, the microphone M2 converts the target sound picked up and non-target sound to the acoustic signal x _2, and outputs the acoustic signal x ₂ to the signal input portion 12.

(Signal input section)
The signal input units 11 and 12 collect sound with the microphones M1 and M2, and convert the output acoustic signals x ₁ and x ₂ from analog signals to digital signals. For example, the signal input units 11 and 12 perform analog / digital conversion by sampling at a frequency twice the maximum audio frequency.
The signal input units 11 and 12 perform frequency analysis on the converted digital signal. The signal input units 11 and 12 convert the digital signal from the time domain to the frequency domain using, for example, fast Fourier transform. In the following, the amplitude component of the acoustic signal represented in the frequency domain is referred to as “amplitude spectrum” and is represented by capital letters. Then, the signal input units 11 and 12 output the amplitude spectra X ₁ and X ₂ of the acoustic signal to the signal adding unit 21 and the unidirectional forming units 31 and 32.

(Signal adder)
The signal adding unit 21 cooperates with the target sound component by adding the amplitude spectra X ₁ and X ₂ of the acoustic signals input from the signal input units 11 and 12 and then multiplying by half. The signal addition unit 21 outputs the amplitude spectrum X _DS coordinated acoustic signal non-target signal extraction unit 41, and the target signal extractor 51.

(Unidirectionality forming part)
The unidirectional formation units 31 and 32 use the amplitude spectra X ₁ and X ₂ of the acoustic signals from the signal input units 11 and 12 to provide unidirectionality that directs the dead angle to ± 90 ° with respect to the target direction. Amplitude spectra A _L and A _R having the same are formed. Formation of amplitude spectra A _L and A _R having unidirectionality is calculated as θ = + π / 2 and θ = −π / 2, respectively, according to the above-described equations (1) and (3). Here, the unidirectional forming units 31 and 32 perform gain correction for each frequency on the amplitude spectra A _R and A _L.
τ = (d · sin θ) / c (1) Formula A (ω) = X ₂ (ω) − [exp (−jωτ)] X ₁ (ω) (3)

Specifically, unidirectional forming unit 31, as θ = + π / 2, the subtraction processing after multiplying the "e ^{-Jeiomegatau"} from the signal input unit 11 to the amplitude spectrum X ₁ output acoustic signals it is carried out, the _(amplitude spectrum a _R forming the unidirectional to the "-90 °" direction relative to the intended _direction) amplitude spectrum a _R of unidirectional directing blind spots + 90 ° with respect to the intended direction It forms (refer Fig.2 (a)). The amplitude spectrum A _R, include a non-target sound arriving from which is the side of objective sound and unidirectional reaching the intended direction is formed "-90 °" direction.

Further, the unidirectional forming unit 32 performs subtraction processing after multiplying the amplitude spectrum X ₂ of the acoustic signal output from the signal input unit 12 by “e ^−jωτ ”, with θ = −π / 2. in form _(amplitude spectrum a _L forming the unidirectional to "+ 90 °" direction relative to the intended _direction) amplitude spectrum a _L of unidirectional directing blind spots -90 ° with respect to the intended direction (See FIG. 2 (a)). The amplitude spectrum _AL includes a target sound arriving from the target direction and a non-target sound arriving from the “+ 90 °” direction on the side where the unidirectionality is formed.

(Non-purpose signal extraction unit)
The non-target signal extraction units 41 and 42 perform spectral subtraction (SS) on the outputs A _R and A _L of the unidirectional forming units 31 and 32 from the output X _DS of the signal addition unit 21, respectively, and ±± The non-target sounds N _UDL and N _UDR existing in the 90 ° direction (dead angle) are extracted. Extraction of the non-target sounds N _UDL and N _UDR is performed according to the equations (4) and (5).
N _UDL = X _DS -A _R (4) Formula N _UDR = X _DS -A _L ...・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ ・ (5)

That is, non-target signal extraction unit 41, by spectral subtraction (SS) the amplitude spectrum _{A R} from the amplitude spectrum _{X DS,} to extract the non-target sound _{N UDL} present in "+ 90 °" direction relative to the intended direction (See FIG. 2 (b)). Then, the non-purpose signal extraction unit 41 outputs the extracted non-purpose sound N _UDL to the target signal extraction unit 51.
Also, non-target signal extraction unit 42, by spectral subtraction (SS) the amplitude spectrum _{A L} from the amplitude spectrum _{X DS,} extracted non-target sound _{N UDR} present in "-90 °" direction relative to the intended direction (See FIG. 2 (b)). Then, the non-purpose signal extraction unit 42 outputs the extracted non-purpose sound N _UDR to the target signal extraction unit 51.

Here, the non-target signal extraction unit 41 and 42, the use of the output _{X DS} signal addition unit 21 and used as an output _X 1, _{X 2} of the signal input unit 11 or the signal input unit 12 You can also The same applies to the target signal extraction unit 51 described later.

(Target signal extraction unit)
Target signal extraction part 51, a non-target sound present in the ± 90 ° direction with respect to the destination direction extracted from the output _{X DS} signal addition unit 2 in a non-target sound extraction section 41 and 42 _{N UDL,} spectral subtraction the _{N UDR} (SS) and the amplitude spectrum Y of the target sound is extracted. Extraction of the amplitude spectrum Y of the target sound is performed according to the equation (6). Here, “β _L ” and “β _R ” are coefficients for adjusting the intensity of spectral subtraction (SS).
_{Y = X DS -β L N DUL} -β R N DUR ············· (6) formula

Accordingly, the target signal extraction unit 51 forms an amplitude spectrum Y of the target sound having a sharp directivity with respect to the target direction (see FIG. 2C). Then, the target signal extraction unit 51 outputs the amplitude spectrum Y of the target sound.
This is the end of the description of the configuration of the sound source separation device 100 according to the embodiment.

<< Effect of the sound source separation apparatus according to the embodiment >>
Next, the effect of the sound source separation device 100 according to the embodiment will be described with reference to FIG. FIG. 3 is a diagram illustrating an example of a usage state of the sound source separation device according to the embodiment.
In FIG. 3, the first microphone M1 is disposed on the left side when viewed from the target direction, and the second microphone M2 is disposed on the right side when viewed from the target direction. Here, for convenience, it is assumed that the distance between the first microphone M1, the speaker, and the left speaker L and the distance between the second microphone M2, the speaker, and the right speaker R are equal.

Assuming that τ is the time difference between the sounds (non-target sounds) reproduced from the speakers L and R reaching the two microphones M1 and M2, the acoustic signal x ₁ (t) collected by the first microphone M1 at a certain time t. ) Include the speaker's voice s (t) as the target sound, the playback sound p _L (t) of the left speaker L as the non-target sound, and the playback sound p _R (of the right speaker R as the non-target sound. t-τ). Further, the acoustic signal x ₂ (t) picked up by the second microphone M2 includes the speaker's voice s (t) as the target sound and the reproduced sound p _L (t of the left speaker L as the non-target sound. ), The reproduction sound p _R (t−τ) of the right speaker R, which is a non-target sound, is included.

When the same sound is reproduced from the left and right speakers L and R with the same volume, the components included in the acoustic signal x ₁ (t) and the acoustic signal x ₂ (t) are equal and centered at time t. The amplitude spectra X ₁ (t) and X ₂ (t) obtained by performing the FFT operation in the time frame are both expressed as Equation (7). Here, “S” is the amplitude spectrum of the target sound, and “P” is the amplitude spectrum of the non-target sound.
X (t) = S (t) + P (t) + P (t−τ) (7)

Further, when the time difference between the sound reproduced from the speakers L and R reaching the two microphones M1 and M2 (the delay time τ described above) is sufficiently small, it can be approximated as P (t) ≈P (t−τ). Therefore, equation (7) can be expressed as equation (8).
X (t) = S (t) + 2P (t) (8)

Next, the unidirectional formation units 31 and 32 direct the blind spots in the “+ 90 °” direction from the amplitude spectra X ₁ (t) and X ₂ (t) of the acoustic signals collected by the microphones M1 and M2. An amplitude spectrum A _R (t) having unidirectionality and an amplitude spectrum A _L (t) having unidirectionality in which a blind spot is directed in the “−90 °” direction are formed. In the case of the amplitude spectrum X ₁ (t) = X ₂ (t), the amplitude spectra A _L (t) and A _R (t) output from the unidirectional forming units 31 and 32 are also equal (A _L (t ) = A _R (t)). The amplitude spectrum A (t) that is the output of the unidirectional forming units 31 and 32 is shown in the equation (9). Since A (t) shown in Equation (9) suppresses the left or right non-target sound P (t) due to the unidirectional blind spot, P (t) included in A (t) is X It becomes half compared to (t). It is assumed that A (t) is gain correction for each frequency.
A (t) = S (t) + P (t) (9)

Subsequently, the non-target signal extraction units 41 and 42 perform spectral subtraction (SS) of A (t) from X (t) according to the equation (10). As a result, S (t) is suppressed, and the non-target sound P (t) existing in a direction (dead angle) of ± 90 ° with respect to the target direction is extracted. As a result, only P (t) is included in the outputs N (t) of the non-purpose signal extraction units 41 and 42.
N (t) = X (t) −A (t) (10)

Finally, the target signal extraction unit 51 performs spectral subtraction (SS) of N (t) from X (t) according to the equation (11). Here, “β” is a coefficient (parameter) for adjusting the intensity of spectral subtraction (SS). If the equation (8) is satisfied, if “β = 2” is set, P (t) included in X (t) is suppressed, and the final output Y (t) = S (t) is obtained to extract the target sound. be able to.
Y (t) = X (t) −βN (t) (11)

  Next, the result of the performance evaluation experiment of the sound source separation device 100 according to the embodiment will be described. FIG. 4 is a result of simulation of the amount of suppression of reproduced sound when it is assumed that different sounds from the left and right speakers L and R or the same sound are simultaneously reproduced in the arrangement shown in FIG. As experimental conditions, both the target sound and the non-target sound used were speech, and NRR (Noise Reduction Rate) was used as a suppression amount index. For comparison with the present invention, the method of Patent Document 2 was used as a conventional method.

  When different sounds were reproduced from the left and right speakers L and R, the NRR in the conventional method was “36 dB”, and the NRR in the present invention was “32 dB”. It can be seen that both can sufficiently suppress the non-target sound. However, if the same sound is reproduced from the left and right speakers L and R, it can be seen that the NRR in the conventional method is “5 dB” and the suppression amount is degraded. This is because only the non-target sound of one of the left and right speakers L and R can be suppressed. In comparison, the NRR in the present invention is proud of a high suppression amount of “23 dB”, and it can be seen that the non-target sounds of the left and right speakers L and R can be suppressed.

  As described above, the sound source separation device 100 according to the embodiment forms unidirectivity in the left and right directions with respect to the target direction, and extracts the non-target sound existing in the blind direction of each unidirectional. A sharp directivity can be formed in the target direction by spectral subtraction (SS) of the non-target sound extracted from the input signal. Therefore, the sound source separation device 100 according to the embodiment can suppress the non-target sound even when the sound source is arranged in the middle of two sound sources that generate the same sound (non-target sound) with the same magnitude, and the target direction It is possible to separate and extract the target sound that reaches from the sound source.

[Modification]
As mentioned above, although embodiment of this invention was described, this invention is not limited to this, It can implement in the range which does not change the meaning of a claim. The modification of embodiment is shown below.

  In the embodiment, the case where the sound source separation device 100 includes two microphones M1 and M2 has been described. However, the sound source separation device 100 receives an acoustic signal collected by an external microphone using a wired line or a wireless line, Sharp directivity may be formed in the target direction using the received acoustic signal.

  Further, in the embodiment, the case where the sound source separation device 100 forms sharp directivity in the target direction using the acoustic signals collected from the two microphones M1 and M2 has been described. However, the sound source separation device 100 may use acoustic signals collected from a plurality of microphones arranged horizontally with respect to the target direction, and the number of microphones is not particularly limited. With reference to FIG. 5, a description will be given of a case where sharp directivity is formed in a target direction using acoustic signals collected by the four microphones M1, M2, M3, and M4.

FIG. 5A is a diagram illustrating an arrangement of microphones included in a sound source separation device according to a modification. FIG. 5B is a diagram illustrating a combination example of microphones for forming unidirectionality.
As shown in FIG. 5A, when the target sound comes from the front (z-axis direction), four omnidirectional microphones M1, M2, M3, and M4 are arranged on the xy plane.

  The sound source separation device 200 (not shown) according to the modified example is configured such that two of the four microphones M1, M2, M3, and M4 are “fourth microphone M4 and second microphone” as shown in FIG. "A pair B1 of one microphone M1", "a pair B2 of a first microphone M1 and a second microphone M2", "a pair B3 of a second microphone M2 and a third microphone M3", "a third microphone M3 and Four patterns of “pair B4 of the fourth microphone M4” are created. Subsequently, the sound source separation device 200 (not shown) uses these four pairs B1 to B4 to form unidirectionalities in four directions, up, down, left, and right. The formation of unidirectionality in the four directions of up, down, left, and right is performed, for example, according to the following equations (12) to (15).

A ₁ (ω) = X ₁ (ω) − [exp (−jωτ)] X ₄ (ω) (12) Formula A ₂ (ω) = X ₂ (ω) − [exp (−jωτ)] X ₁ (ω) (13) Formula A ₃ (ω) = X ₃ (ω) − [exp (−jωτ)] X ₂ (ω) (14) Formula A ₄ (ω) = X ₄ (ω) − [exp (−jωτ)] X ₃ (ω) (15)

  Subsequently, the sound source separation device 200 (not shown) according to the modified example extracts the non-target sound existing in the blind spot direction of each unidirectional using the unidirectionality in the four directions of up, down, left, and right. A sharp directivity can be formed in the target direction by spectral subtraction (SS) of the non-target sound extracted from the input signal. Therefore, even when the sound source separation device 200 (not shown) according to the modified example is arranged in the middle of four sound sources that are arranged vertically and horizontally that generate the same sound (non-target sound) with the same magnitude, The non-target sound can be suppressed, and only the target sound that arrives from the sound source in the target direction can be separated and extracted.

  In the embodiment, the case where the directivity sharp in the target direction is calculated using the amplitude spectrum of the acoustic signal has been described. However, the calculation may be performed using the power spectrum.

DESCRIPTION OF SYMBOLS 11, 12 Signal input part 21 Signal addition part 31, 32 Unidirectional formation part 41, 42 Non-target signal extraction part 51 Target signal extraction part 100 Sound source separation device M1, M2, M3, M4 Microphone

Claims (3)

Of the sound that reaches a plurality of microphones located in the plane whose normal direction is the target direction, the target sound that arrives from the sound source that exists in the target direction and the non-purpose that arrives from the sound source that exists in the other direction A sound source separation device for separating sound from
Using the acoustic signal output by any two microphones of the plurality of microphones, the dead angle is reverse to the acoustic signal having a unidirectionality in which the blind angle is directed to any of the directions connecting the two microphones. A unidirectional forming unit that generates two so that
The unidirectionality generated by the unidirectional forming unit with respect to an acoustic signal output from any one of the plurality of microphones or an averaged acoustic signal output from two or more microphones. A non-target signal extraction unit that performs spectral subtraction using two acoustic signals and extracts a signal of a non-target sound that arrives from each blind spot;
A non-target sound signal extracted by the non-target signal extraction unit is used for an acoustic signal output from any one of the plurality of microphones or an average signal of an acoustic signal output from two or more microphones. A target signal extraction unit that performs spectral subtraction and extracts a target sound signal;
A sound source separation device comprising: Of the sound that reaches a plurality of microphones located in the plane whose normal direction is the target direction, the target sound that arrives from the sound source that exists in the target direction and the non-purpose that arrives from the sound source that exists in the other direction A sound source separation method executed by a sound source separation device that separates sound,
Using the acoustic signal output by any two microphones of the plurality of microphones, the dead angle is reverse to the acoustic signal having a unidirectionality in which the blind angle is directed to any of the directions connecting the two microphones. Generate two so that
A spectrum obtained by using two acoustic signals having the unidirectionality generated with respect to an acoustic signal output from any one of the plurality of microphones or an average of acoustic signals output from two or more microphones. Subtract, extract the signal of the non-target sound that arrives from each blind spot,
Spectral subtraction is performed on the acoustic signal output from any one of the plurality of microphones, or the average of the acoustic signals output from two or more microphones, using the extracted non-target sound signal. Extract the signal of the
A sound source separation method characterized by the above.   A sound source separation program for causing a computer to execute the sound source separation method according to claim 2.

Images