JP6863004B2 - Sound collectors, programs and methods - Google Patents

Description

The present invention relates to a sound collecting device, a program and a method, and can be applied to, for example, a case where only a sound in a specific area is emphasized and a sound in another area is suppressed.

A beam former using a microphone array as a technique for emphasizing sounds existing in a specific direction (voice and sound; hereinafter, voice and sound are sometimes collectively referred to as "sound") and suppressing other sounds. There is. The beam former is a technique for forming directivity and blind spots by utilizing the time difference between signals arriving at each microphone (see Non-Patent Document 1 and Non-Patent Document 2).

However, if the directivity of the beam former is simply directed to an area for sound collection purposes (hereinafter referred to as "target area"), if there is a noise source around the target area, it exists in the target area. There is a problem that not only the sound source (hereinafter referred to as "target area sound") but also the noise source existing outside the target area (hereinafter referred to as "non-purpose area sound") is picked up at the same time.

To solve this problem, conventionally, a method has been proposed in which directivity is crossed from different directions toward a target area by using a plurality of microphone arrays, and sound in the target area is picked up (Patent Document 1). In the method described in Patent Document 1, the target area is extracted by simultaneously processing the beamformer output of each microphone array.

FIG. 6 is an explanatory diagram showing an example of sound collection processing using a plurality of conventional microphone arrays.

FIG. 6 shows an example in which the directivity of the _two microphone arrays MA (MA ₁ and MA 2) is directed toward the target area.

FIG. 6A shows the positional relationship between each microphone array MA and the sound source of the target area sound when the directivity of the _two microphone arrays MA ₁ and MA 2 is directed toward the target area. Further, FIG. 6A also illustrates the directivity (directivity of the beam former) Z1 and Z2 corresponding to _{the microphone arrays MA 1} and MA _2. Further, in the example of FIG. 6A, a sound source of the non-purpose area sound exists around the sound source of the target area. Therefore, in the state of FIG. _{6A, both the beamformer outputs of the microphone arrays MA 1} and MA ₂ are not only in the target area sound by the sound source in the target area but also in the non-purpose area in the same directivity direction. The non-purpose area sound from the sound source will be included.

6 (b) and 6 (c) show the frequency components of the beamformer outputs of the _{two microphone arrays MA 1} and MA 2, respectively. Assuming the sparsity of speech, as shown in FIGS. 6 (b) and 6 (c), one frequency component includes only one sound source (target area sound or non-target area sound). Since the target area is included in the directivity of all the microphone arrays, the frequency components of the target area sound are included in all the beam former outputs in the same proportion and with the same distribution. In comparison, the frequency component of the non-purpose area sound is different for each beamformer output. From these characteristics, the frequency component commonly included in each beamformer output can be estimated to be a component of the target area sound, and based on this, the conventional target area sound described in Patent Document 1 and the like can be estimated. The sound collection method of is realized.

FIG. 7 is a block diagram showing a functional configuration of the sound collecting device 10 to which the conventional sound collecting method is applied.

The conventional sound collecting device 10 shown in FIG. 7 includes a data input unit 2, a frequency domain conversion unit 3, a directivity forming unit 4, a propagation delay difference correction unit 5, a power correction unit 6, a first subtraction unit 7, and a first subtraction unit 7. It has a subtraction unit 8 of 2.

The captured signals from the microphone arrays MA ₁ and MA ₂ are converted from analog signals to digital signals (data) in the data input unit 2 and converted from the time domain to the frequency domain in the frequency domain conversion unit 3, respectively, and captured signals. Groups X ₁ and X ₂ are obtained. Then, a beam former having directivity such as the directivity Z1 and the directivity Z2 of FIG. 6A is applied to the directivity forming unit 4, and the beamformer output signals X ma1 (f) and X _ma2 (f) are _generated. can get. Then, the propagation delay difference correction unit 5 delays one of the beamformer output signals X _ma1 (f) and X _ma2 (f) based on the distance (known information) between each microphone array and the target area to adjust the timing. together, the delay correction signal X _'ma1 (f) and X' _ma2 (f) is obtained.

In the power correction unit 6, in addition to the amplitude difference due to the distance between each microphone array and the target area, the amplitude correction coefficient α _ma1 (alpha) is calculated by the equation (1) in order to adapt to the direction of the speaker in the target area. To do. _{The operator mode f} (A (f)) in the equation (1) is an operator that obtains the most frequently occurring value (mode) of the function values A (f) whose values change depending on the variable f. .. Further, instead of the mode value, the median value may be used as in Eq. (2). _{The operator median f} (A (f)) in the equation (2) is an operator that obtains the median value of the function value A (f) whose value changes depending on the variable f.

Then, in the first subtracting unit 7, 'delay correction signal X according _ma1 from (f) to the microphone array MA ₂ obtained by correcting the amplitude by the amplitude correction coefficient alpha _{ma1' ma2} delay correction signal X according to the microphone array MA ₁ (f ) by spectral subtraction to both beamformer erased object area sound components overlapping in the output, the delay correction signal X _{'ma1 (f)} non-target area sound components contained in the N of the microphone array MA _{1 ma1} (f) is extracted. Equation (3) is a calculation equation that generally follows this way of thinking.
_{N ma1 = X 'ma1 -α ma1} · X' ma2 ... (3)

Then, in the second subtracting unit 8, by spectral subtraction of the non-target area sound component _{N ma1} (f) from the delay correction signal X _'ma1 according to the microphone array MA ₁ (f), sound object area _{Y ma1} (f ) Is extracted. Equation (4) is a calculation equation that generally follows this way of thinking. _{Note that β ma1} (beta) in Eq. (4) is a coefficient that takes a constant value that determines the removal intensity of non-purpose area sound.
Y _ma1 = _X'ma1- β _ma1 · N _ma1 ... (4)

As described above, by using the conventional sound collection method, even if a non-purpose area sound source exists around the target area, only the target area sound can be collected.

Japanese Unexamined Patent Publication No. 2014-722708

Tadashi Asano, "Sound Array Signal Processing-Localization, Tracking and Separation of Sound Source", Acoustical Society of Japan, Corona Publishing Co., Ltd., February 25, 2011 Takashi Yato, Makoto Morito, Kei Yamada, Tetsuji Ogawa, "Square Microphone Array Sound Separation Technology (<Special Feature> Efforts to Practical Use of Speech Recognition Technology)", Information Processing Society of Japan, Information Processing 51 (11) , Pp. 1410-1416.2010

However, in the conventional sound picking method, since the spectrum subtraction is performed twice in order to pick up only the target area sound, there is a possibility that the extracted target area sound has a problem of sound quality.

In spectrum subtraction, when there is an observation signal in which the target sound component and noise component are mixed and a noise component estimated by an appropriate method, the amplitude or power of the estimated noise component is extracted from the amplitude or power of the observation signal for each frequency component. Is a method of estimating the amplitude or power of the target sound by subtracting. The estimated noise component always includes an estimation error in the real environment. Therefore, spectrum subtraction attenuates even the target sound component in the frequency component where the noise component is overestimated, so that the problem that the target sound is distorted and the noise component is attenuated in the frequency component where the noise component is underestimated. Since it cannot be cut off, there is a problem that the noise component remains. Furthermore, for each frequency component, the sum of the true target sound amplitude or power and the true noise amplitude or power does not always match the amplitude or power of the observed signal, so the estimated noise component is assumed to be an estimation error. Even if the noise component is not included, the spectrum subtraction has a problem that the target sound is distorted and a problem that the noise component remains.

Since the remaining noise component is perceived as extremely unpleasant noise called musical noise, it is generally known as the greatest problem of spectrum subtraction. Musical noise is noise in which the noise component is strongly distorted.

In the conventional sound collecting method, since the spectrum subtraction having the above problems is applied twice, there is a problem that the emphasized target area sound may be distorted.

Therefore, a sound collecting device, a program, and a method that emphasize only the target area sound with less distortion are desired.

The first sound collecting device of the present invention is (1) for each of a plurality of microphone arrays composed of two microphones, which changes according to the direction of arrival of sound, with respect to sound arriving from the direction of the target area. A feature amount calculation means for calculating an arrival direction feature amount having a feature of taking a large value and taking a small value for a sound arriving from a direction other than the target area direction, and (2) each of the above-mentioned microphones for each frequency component. Using the feature quantity integration means for acquiring the area feature quantity that integrates the arrival direction feature quantity of the array and (3) the area feature quantity, the target area sound is extracted from the signal based on the captured signal output by the microphone array. possess the destination area sound extracting means for, (4) the target area sound extracting means, in accordance with the magnitude of the area feature quantity, and extract the target area sound the signal based on the acquired signal, (5) the The target area sound extraction means holds a threshold value for each frequency component in advance, and extracts the target area sound by attenuating a frequency component whose area feature amount is smaller than the threshold value from the signal based on the captured signal. It is characterized by.

The second sound collection program of the present invention is to change the computer according to the direction of arrival of sound for each of (1) a plurality of microphone arrays composed of two microphones, and the sound coming from the direction of the target area. A feature amount calculation means for calculating an arrival direction feature amount having a feature of taking a large value for the sound coming from a direction other than the target area direction and taking a small value for the sound coming from a direction other than the target area direction, and (2) for each frequency component, respectively. The target area from the signal based on the captured signal output by the microphone array by using the feature amount integrating means for acquiring the area feature amount in which the arrival direction feature amount of the microphone array is integrated and (3) the area feature amount. It functions as a target area sound extracting means for extracting sound , and (4) the target area sound extracting means extracts a target area sound from a signal based on the captured signal according to the magnitude of the area feature amount, and (5). The target area sound extraction means holds a threshold value for each frequency component in advance, and extracts a target area sound from a signal based on the captured signal by attenuating a frequency component whose area feature amount is smaller than the threshold value. vinegar Rukoto and features.

The third sound collecting method of the present invention includes (1) a feature amount calculating means, a feature amount integrating means, and a target area sound extracting means, and (2) the feature amount calculating means includes a plurality of microphones. For each microphone array, it changes according to the direction of arrival of sound, and takes a large value for sound arriving from the direction of the target area and a small value for sound arriving from a direction other than the direction of the target area. The arrival direction feature amount having the feature to be taken is calculated, and (3) the feature amount integration means acquires an area feature amount in which the arrival direction feature amount of each of the microphone arrays is integrated for each frequency component, and (5). ) The target area sound extracting means extracts the target area sound from the signal based on the captured signal output by the microphone array using the area feature amount , and (5) the target area sound extracting means is the area feature. The target area sound is extracted from the signal based on the captured signal according to the magnitude of the amount. (6) The target area sound extracting means holds a threshold value for each frequency component in advance, and the signal based on the captured signal. Therefore, the target area sound is extracted by attenuating a frequency component whose area feature amount is smaller than the threshold value.

According to the present invention, it is possible to provide a sound collecting device, a program and a method for emphasizing only the target area sound with less distortion.

It is a block diagram which showed the functional structure of the sound collecting apparatus which concerns on embodiment. It is explanatory drawing which showed the example of the 1st arrival direction feature quantity which concerns on embodiment. It is explanatory drawing which showed the example of the 2nd arrival direction feature quantity which concerns on embodiment. It is explanatory drawing which showed the example of the area feature amount which concerns on embodiment. It is explanatory drawing which showed the example of the determination result of the target area required by the sound collecting apparatus in an embodiment. It is explanatory drawing which showed the example of the conventional sound collection method. It is a block diagram which showed the functional structure of the conventional sound collecting apparatus.

(A) Main Embodiments Hereinafter, one embodiment of a sound collecting device, a program and a method according to the present invention will be described in detail with reference to the drawings.

(A-1) Configuration of the Embodiment FIG. 1 is a block diagram showing a functional configuration of the sound collecting device 100 of the embodiment.

The sound collecting device 100 uses the _{acoustic signals supplied from the M microphone arrays MA (MA 1 to} MA M) to perform the target area sound collecting process for collecting the target area sound from the sound source in the target area. ..

Each microphone array MA is arranged in a space where the target area exists so that the target area can be directed. Each microphone array MA is composed of _{two microphones 1 (1 1} , 1 2). In each microphone array MA, an acoustic signal based on the sound captured by the _{two microphones 1 1} and 12 is supplied to the data input unit 102.

Next, the internal configuration of the sound collecting device 100 will be described with reference to FIG.

As shown in FIG. 1, the sound collecting device 100 according to this embodiment includes a data input unit 102, a frequency domain conversion unit 103, a feature amount calculation unit 104, a feature amount integration unit 105, and a target area sound extraction unit 106. doing. Details of each component inside the sound collecting device 100 will be described later.

In the sound collecting device 100, a computer having a processor, a memory, or the like may execute a program (including a sound collecting program according to the embodiment) for the processing configuration after being converted into a digital signal. Even so, functionally, it can be represented in FIG.

(A-2) Operation of Embodiment Next, the operation of the sound collecting device 100 of this embodiment having the above configuration (sound collecting method of this embodiment) will be described.

The data input unit 102 converts _{the acoustic signal captured by the microphone arrays MA 1 to M M} from an analog signal to a digital signal (data) for each microphone 1. The data input unit 102 gives the obtained captured signal to the frequency domain conversion unit 103.

In the following, the microphone array _MA 1 to MA _M of the captured captured signal by the microphone _{1 1} each represent _{x 1,1 (t) ~x M,} 1 (t), the microphone 1 of the microphone array _MA 1 to MA _{M The} captured signals captured in _{2 are represented as x 1, 2,} (t) to _{x M, 2} (t), respectively.

The frequency domain conversion unit 103 converts the captured signals x ₁ , 1 (t) to x _{M, 1 (t), x 1 , 2,} (t) to x M, 2 (t) from the time domain to the frequency domain, respectively. To do.

In the following, the signals obtained by converting _{the captured signals x 1} , 1 (t) to x _{M, 1 (t), x 1 , 2,} (t) to x M, 2 (t) into the frequency domain are converted into X ₁ , 1 (t). It is expressed as t) to X _{M, 1} (t), X _{1, 2,} (t) to X _{M, 2} (t).

The frequency domain conversion unit 103 features the acquired frequency domain capture signals X _{1, 1} (t) to X _{M, 1} (t), X _{1, 2,} (t) to X _{M, 2} (t). It is supplied to the calculation unit 104 and the target area sound extraction unit 106.

For the conversion performed by the frequency domain transforming unit 103, a fast Fourier transform (FFT), a wavelet transform, a filter bank, or the like can be used, but the FFT is the most preferable. Here, various window functions such as a humming window may be used when performing the FFT.

The feature amount calculation unit 104 is the arrival direction for each microphone array MA from the _{captured signals X 1, 1} (t) to X _{M, 1} (t), X _{1, 2,} (t) to X _{M, 2 (t).} The feature quantities D ₁ (f) to D _M (f) are calculated. The feature amount calculation unit 104 _{supplies the obtained feature amounts D 1} (f) to D _M (f) in the arrival direction to the feature amount integration unit 105.

In the feature amount calculation unit 104, the arrival direction feature amount _{D 1 (f) ~D M (} f) is captured signal _{X 1,1 (t) ~X M,} 1 (t), X 1,2 (t) ~X _{It is calculated from M, 2} (t) for each microphone array MA by the same calculation method. I-th (i is any of 1 to M) acquired signal _X i, 1 _(f) in the microphone array MA _{i of, X i,} 2 (f) and DOA feature amount _D i (f) is described in the following ..

The arrival direction feature amount _Di (f) preferably has a feature that takes a large value with respect to the target area direction and a small value with respect to a direction other than the target area direction. Any calculation method may be used as long as it is a calculation method that allows the arrival direction feature amount _{Di (f) to have such a feature.} When the target area is located in the front direction of all the microphone arrays MA, it is preferable to use the equation (5), for example.

The captured signals X _{i, 1} (f) and X _{i, 2} (f) are signals in which the target area sound and the non-target area sound are mixed. Only area sounds and non-purpose area sounds will be included. Therefore, if the angle at which a certain sound source reaches the microphone array MA is defined as θ (theta), equation (5) can be expanded as equation (6). In equation (6), c is the speed of sound, and d is the distance between the _{two microphones 1 1} , 12 that make up the microphone array. Similarly, assuming the sparsity of voice, a calculation method for explicitly obtaining the arrival direction, such as Eq. (7), can be used as the calculation method for _{the arrival direction feature amount Di (f).} The inside of the absolute value in Eq. (7) is the value (sin θ) of the sine function in the arrival direction θ.

Next, a specific example of the arrival direction feature amount _Di (f) will be described with reference to FIGS. 2 and 3.

2 (a) and 3 (a) show the cases where the arrival direction features D ₁ (f) and D _{2 (f) corresponding to the microphone arrays MA 1} and MA ₂ are obtained by using the equation (5), respectively. Is shown in a three-dimensional (vertical, horizontal, height) graph.

In the graphs of FIGS. 2 (a) and 3 (a), _{the distance from the microphone array MA 1} is the vertical position (vertical axis of the graph), and _{the distance from the microphone array MA 2} is the horizontal position (horizontal direction of the graph). Axis), and _{the values of the feature quantities D 1} (f) and D ₂ (f) in the arrival direction are the height (the axis in the height direction (vertical direction) of the graph). The graphs of FIGS. 2 (a) and 3 (a) show the arrival direction feature amount D when the target area sound or the non-target area sound arrives from various vertical and horizontal positions when f = 3 kHz. _{The values of 1} (f) and D ₂ (f) are shown.

FIG. 2 (b) shows the value of the incoming direction feature amount _D 1 (f) at each position of P411~P416 illustrated in FIG. 2 (a). As shown in FIG. 2B, the values of the arrival direction feature amounts D ₁ (f) at the respective positions of P411 to P416 are −0.13, 1, −0.13, 0.72, 1, 0. It becomes 72.

Incidentally, FIG. 2 (a), the lateral position 1.5m microphone array MA _1, when installed in a vertical position 0 m, the arrival of sound f = 3 kHz about the microphone array MA ₁ direction feature amount _D 1 of the (f) It is a graph. As shown in FIGS. 2 (a) and 2 (b), the _{feature amount D 1} (f) in the arrival direction becomes the peak value in the front direction (when the lateral position is 1.5 m) of the _{microphone array MA 1.} You can see that there is.

_{FIG. 3 (b) shows the values of the arrival direction feature amount D 2} (f) at each position of P421 to P426 illustrated in FIG. 3 (a). As shown in FIG. 3 (b), the values of the arrival direction feature amounts D ₂ (f) at the respective positions of P421 to P426 are 0.72, −0.13, 1, −0.13, 0.72, It becomes 1.

3 (a) is, the microphone array MA ₂ lateral position 0 m, when placed in a vertical position 1.5 m, a graph of the arrival direction feature amount _D 2 (f) in f = 3 kHz about the microphone array MA ₂ is there. As shown in FIGS. 3 (a) and 3 (b), the _{feature amount D 2} (f) in the arrival direction becomes the peak value in the front direction (when the vertical position is 1.5 m) of the _{microphone array MA 2.} You can see that there is.

The feature amount integration unit 105 integrates the arrival direction feature amounts D ₁ (f) to D _M (f) for each frequency component to calculate the area feature amount E (f). The obtained area feature amount E (f) is given to the target area sound extraction unit 106.

The calculation method (integration method) of the _{area feature amount E (f) is such that when all the arrival direction feature amounts D 1} (f) to D _M (f) are large, the area feature amount E (f) is also large. Any calculation method may be used as long as it is a method (integration method). For example, as in Eq. (8), the minimum arrival direction feature amount D _{1 for all microphone arrays for each frequency component.} (f) _~D M (f) may be used as the select by area feature quantity E (f).
E (f) = min [D ₁ (f), ..., D _M (f)] ... (8)

Next, a specific example of the area feature amount E (f) will be described with reference to FIG.

FIG. 4A is a three-dimensional (vertical, horizontal, height) graph showing an example in which the area feature amount E (f) is obtained using the equation (8).

_{FIG. 4A shows the arrival direction features D 1} (f) and D ₂ (f) when the microphone arrays MA ₁ and MA ₂ (two microphones 1) are arranged as shown in FIG. 6, respectively. An example is shown in which the area feature amount E (f) is calculated by applying _{the arrival direction feature amounts D 1} (f) and D ₂ (f) calculated by the formula (5) to the formula (8). There is. That is, FIG. 4A shows the area feature amount E (8) in which the _{arrival direction feature amounts D 1} (f) and D ₂ (f) shown in FIGS. 2 (a) and 3 (a) are integrated by the equation (8). f) is shown.

In the graph of FIG. 4A, _{the distance from the microphone array MA 1} is the vertical position (vertical axis of the graph), and _{the distance from the microphone array MA 2} is the horizontal position (horizontal axis of the graph). The value of the feature amount E (f) is defined as the height (the axis in the height direction (vertical direction) of the graph). Note that FIG. 4A shows the values of the area feature amounts E (f) at various vertical and horizontal positions when f = 3 kHz.

FIG. 4B shows the value of the area feature amount E (f) at each position of P51 to 59 shown in FIG. 4A. As shown in FIG. 4B, the values of the area feature amounts E (f) at the respective positions of P51 to 59 are -0.13, 0.36, -0.13, 0.36, and -0.13. , 0.36, 0.72, 0.36, 1.

As shown in FIGS. 4 (a) and 4 (b _{), the area feature amount in the front direction of the microphone array MA 1} and the microphone array MA ₂ (around the point where both the horizontal position and the vertical position are 1.5 m). It can be seen that E (f) is a large value.

The target area sound extraction unit 106 includes captured signals X _{1, 1} (t) to X _{M, 1} (t), X _{1, 2,} (t) to X _{M, 2} (t) and an area feature amount E (f). The target area emphasis sound Y (f) is calculated based on. Then, the target area sound extraction unit 106 supplies (outputs) the obtained target area emphasis sound Y (f) to the next stage.

In the target area sound extraction unit 106, selection of the captured signal to be extracted (emphasized) of the target area sound (X _{1, 1} (t) to X _{M, 1} (t), X _{1, 2,} (t) to X _{M , 2} (t) is optional), for example, it may be the first X _{1, 1} (f), it may be the captured signal related to the microphone closest to the target area, or it may be the closest target area. A delay sum beam former may be applied to a group of captured signals of a nearby microphone array MA to provide a signal (called an integrated captured signal) in which the target area sound is slightly emphasized. Hereinafter, the selected capture signal or the integrated capture signal is referred to as an extraction target signal X'(f).

In the target area sound extraction unit 106, extraction (emphasis) of the target area sound is achieved by attenuating the frequency components other than the target area sound among the frequency components of the extraction target signal X'(f). Since the area feature amount E (f) has a larger value as it is closer to the target area, the target area sound extraction unit 106 determines the extraction target signal X'(in accordance with the magnitude of the area feature amount E (f). By attenuating f), the target area sound can be extracted (emphasized). In the target area sound extraction unit 106, for example, as in Eq. (9), a predetermined threshold value F (f) is set in advance for each frequency component, and the area feature amount E (f) is the threshold value F (f). If it is smaller, the frequency component of the extraction target signal X'(f) is attenuated (for example, set to zero), so that the target area emphasized sound Y (f) in which only the frequency component of the target area sound remains can be obtained. it can.

In the target area sound extraction unit 106, the threshold value F (f) may be a constant value regardless of the frequency component, but in that case, the range of the area extracted (emphasized) by the frequency component changes. This is because the target area becomes wider as the frequency is lower and narrower as the frequency is higher. Therefore, in the target area sound extraction unit 106, for example, by setting the threshold value F (f) for each frequency component as in Eq. (10), the target area (range in which the frequency component is not attenuated) can be set regardless of the frequency level. It can be set within a certain range. In equation (10), φ (phi) is the size (angle) of the target area as seen from each microphone array.

Similar to FIGS. 2 to 4, FIG. 5 shows a range determined to be the target area when φ = π / 10 when the _{microphone arrays MA 1} and MA _{2 are arranged as shown in FIG.} Is shown.

In FIG. 5, the area filled in black indicates the range determined not to be the target area based on the threshold value, and the other area (the area not filled in black) is determined to be the target area based on the threshold value. Is shown.

As shown in FIG. 5, it can be seen that the range of about 1 to 2 m is determined to be the target area based on the threshold value in both the vertical and horizontal directions.

(A-3) Effect of Embodiment According to this embodiment, the following effects can be achieved.

In the sound collecting device 100 of this embodiment, since spectrum subtraction is not performed, only the target area sound can be emphasized with a small amount of distortion even in a situation where the target area is surrounded by a non-target area sound source.

(B) Other Embodiments The present invention is not limited to the above embodiments, and modified embodiments as illustrated below can also be mentioned.

(B-1) In the feature amount calculation unit 104, the equations (11) and (12) can also be applied to the calculation method of the _{feature amount Di (f) in the arrival direction.}

Further, the feature quantity integration unit 105, the method of calculating the area feature quantity E (f) (a method of integrating the arrival direction feature amount _{D 1 (f) ~D M (} f)), (13) type or (14) Expressions can also be applied.

(B-2) In the target area sound extraction unit 106, a constant value may be set for a frequency component smaller than a certain frequency (for example, 250 Hz) having a threshold value F (f). For example, when F (f) at 250 Hz is used as the value of F (f) below 250 Hz, the range determined to be the target area is widened for frequency components below 250 Hz, and low frequency components are less likely to be distorted. The target area emphasis sound Y (f) with less distortion can be obtained.

Further, in the target area sound extraction unit 106, two threshold values F ₁ (f) and F ₂ (f) are prepared, the area emphasis gain G (f) is calculated, and the obtained area emphasis gain G (f) is calculated. ) May be multiplied by the extraction target signal X'(f) to calculate the target area emphasis sound Y (f). For example, assuming that the two threshold values F ₁ (f) and F ₂ (f) are calculated according to the equation (15), the area is emphasized by the equation (16) with _{φ 1} = π / 9 and φ _{2-π / 11.} The gain may be calculated. As a result, among the frequency components of the extraction target signal X'(f), the degree of attenuation of the components derived from the sound source existing near the boundary between the target area and the non-target area becomes gentle, so that the target area with less distortion. The emphasis sound Y (f) is obtained.

100 ... Sound collecting device, 102 ... Data input unit, 103 ... Frequency domain conversion unit, 104 ... Feature amount calculation unit, 105 ... Feature amount integration unit, 106 ... Target area sound extraction unit.

Claims (6)

Each of a plurality of microphone arrays consisting of two microphones changes according to the direction of arrival of sound, takes a large value for the sound arriving from the direction of the target area, and arrives from a direction other than the direction of the target area. A feature amount calculation means for calculating an arrival direction feature amount having a feature that takes a small value with respect to sound, and
For each frequency component, a feature quantity integrating means for acquiring an area feature quantity by integrating the arrival direction feature quantity of each microphone array, and a feature quantity integrating means.
It has a target area sound extraction means for extracting a target area sound from a signal based on a captured signal output by the microphone array using the area feature amount .
The target area sound extraction means extracts the target area sound from the signal based on the captured signal according to the magnitude of the area feature amount.
The target area sound extraction means holds a threshold value for each frequency component in advance, and extracts a target area sound from a signal based on the captured signal by attenuating a frequency component whose area feature amount is smaller than the threshold value. A sound collecting device characterized by that. The feature quantity integrating means takes a large value for each frequency component when all the arrival direction features are large, and takes a small value when any of the arrival direction features is small, thereby causing the area feature quantity. The sound collecting device according to claim 1, wherein the sound collecting device is obtained. The sound collecting device according to claim 2, wherein the feature amount integrating means acquires the minimum value of the arrival direction feature amount for all the microphone arrays as the area feature amount for each frequency component. The sound collecting device according to claim 1 , wherein the threshold value for each frequency component is set so as to set a range in which the frequency component is not attenuated, regardless of the frequency level. Computer,
Each of a plurality of microphone arrays consisting of two microphones changes according to the direction of arrival of sound, takes a large value for the sound arriving from the direction of the target area, and arrives from a direction other than the direction of the target area. A feature amount calculation means for calculating an arrival direction feature amount having a feature that takes a small value with respect to sound, and a feature amount calculation means.
For each frequency component, a feature quantity integrating means for acquiring an area feature quantity by integrating the arrival direction feature quantity of each microphone array, and a feature quantity integrating means.
Using the area feature amount, it is made to function as a target area sound extraction means for extracting a target area sound from a signal based on a captured signal output by the microphone array.
The target area sound extraction means extracts the target area sound from the signal based on the captured signal according to the magnitude of the area feature amount.
The target area sound extraction means holds a threshold value for each frequency component in advance, and extracts a target area sound from a signal based on the captured signal by attenuating a frequency component whose area feature amount is smaller than the threshold value. A sound collection program characterized by that. In the sound collection method
It is equipped with a feature amount calculation means, a feature amount integration means, and a target area sound extraction means.
The feature amount calculation means changes according to the direction of arrival of sound for each of a plurality of microphone arrays composed of two microphones, takes a large value for the sound arriving from the direction of the target area, and takes a large value for the target area. Calculate the arrival direction feature amount, which has a feature that takes a small value for the sound coming from a direction other than the direction.
The feature quantity integrating means acquires an area feature quantity that integrates the arrival direction feature quantity of each microphone array for each frequency component.
The target area sound extraction means extracts the target area sound from the signal based on the captured signal output by the microphone array using the area feature amount .
The target area sound extraction means extracts the target area sound from the signal based on the captured signal according to the magnitude of the area feature amount.
The target area sound extraction means holds a threshold value for each frequency component in advance, and extracts a target area sound from a signal based on the captured signal by attenuating a frequency component whose area feature amount is smaller than the threshold value. A sound collection method characterized by that.

Images