JP2018136509A - Signal processing apparatus, program, and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enhance a target sound at lower computational cost and with lower distortion.SOLUTION: This invention relates to a signal processing apparatus. The signal processing apparatus of this invention includes: means for conducting frequency analysis of a first input signal input from a first sound collecting device to obtain a first input spectrum; means for conducting frequency analysis of a second input signal input from a second sound collecting device to obtain a second input spectrum; means for computing a first feature quantity based on the first and second input spectrums, the first feature quantity assuming a value in a direction of the first sound collecting device larger relative to a front direction and assuming a value in a direction of the second sound collecting device smaller relative to the front direction; means for mapping the first feature quantity with a predetermined broad-sense monotonically increasing function to obtain an enhancement filter; and multiplication means for multiplying the enhancement filter obtained by the first input spectrum to obtain an enhanced spectrum.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a signal processing device, a program, and a method, for example, a communication terminal, an audio device, and a voice that want to emphasize and collect sound sources existing in a range in a specific direction in an environment where a plurality of sound sources exist. It can be applied to a recognition device or the like.

  As a technique for extracting a target sound source in an environment where there are a plurality of sound sources, a sound source separation using a plurality of microphones, and a beamformer using a microphone array in which the microphones are arranged on a straight line, a plane, a spherical surface, etc. And nullformers. In particular, when there are non-stationary sound sources other than the target sound source or when there are a plurality of sound sources, it is difficult to extract the target sound source with a noise suppressor using a single microphone, and it is essential to use two or more microphones. Become.

  The beam former using the above-described microphone array is a technique that emphasizes and collects only sound in a specific direction. The beam former is a technique for forming directivity by using a time difference between signals reaching each microphone.

  There are two types of beamformers: an addition type and a subtraction type. Compared with the addition beamformer, the subtraction beamformer has an advantage that a sharp directivity can be formed with a smaller number of microphones.

  FIG. 13 is a block diagram showing a configuration related to a subtractive beamformer when the number of microphones is two. The subtractive beamformer shown in FIG. 13 includes a first microphone M1, a second microphone M2, a first delay unit 3, a second delay unit 4, and a subtracting unit 5. The first input signal picked up by the first microphone M 1 is given to the first delay means 3, and the second input signal picked up by the second microphone M 2 is given to the second delay means 4. When the disturbing sound has arrived from the first microphone M1 side, the first delay means 3 delays the first input signal, thereby causing the disturbing sound included in the first input signal and the second input signal. Adjust the phase. On the other hand, when the disturbing sound has arrived from the second microphone M2 side, the second delay means 4 delays the second input signal, thereby matching the phase of the disturbing sound. The first delay signal obtained from the first delay means 3 and the second delay signal obtained from the second delay means are supplied to the subtraction means 5. The subtracting means 5 obtains emphasized speech by subtracting the second delayed signal from the first delayed signal. As described above, the subtractive beamformer emphasizes the target sound by matching and subtracting the phases of the interference sounds included in the first input signal and the second input signal and suppressing the interference sounds. The subtractive beamformer requires the direction-of-arrival information of disturbance sound given in advance.

  By the way, the subtractive beamformer has a problem that if the disturbing sound source is moved even a little, the suppression performance of the disturbing sound is greatly deteriorated.

  FIG. 14 is an explanatory diagram showing an example in which the voice of the driver U1 in the automobile (vehicle) A is emphasized using the conventional signal processing device Z.

  For example, as shown in FIG. 14, in a car navigation system that can be operated by voice using voice recognition, it is necessary to extract only the voice of the driver in the car.

  Therefore, when a person is in the driver's seat and the passenger seat, it is necessary to suppress the voice (interference sound) of the assistant U2 in the passenger seat. When moved, the subtractive beamformer cannot suppress the interference sound.

  A minimum dispersion beamformer (MVB), which is one of representative adaptive beamformers, is a method that can efficiently suppress interference sound by giving the arrival direction of a target sound in advance. The MVB suppresses the interference sound by minimizing the dispersion of the emphasized speech under the constraint condition that the gain is 1 with respect to the arrival direction of the target sound.

  Further, by using the spectral subtraction method, strong directivity can be formed in the direction of arrival of the target sound source. In Non-Patent Document 1, assuming that the target sound source is always in front, first, a target sound suppression signal in which the target sound arriving from the front direction is suppressed by a subtractive beamformer is obtained, and secondly, the first input By subtracting the amplitude spectrum of the target sound suppression signal from the amplitude spectrum of the signal (spectral subtraction), an amplitude spectrum of the emphasized speech in which the target sound is emphasized is obtained. Third, the amplitude spectrum of the emphasized speech and the phase of the first input signal Emphasized speech is obtained using the spectrum.

Takashi Yagami, Makoto Morito, Satoshi Yamada, Tetsuji Ogawa, “Sound Source Separation Technology Using Square Microphone Array”, Information Processing, Vol. 51, no. 11, 2010

  However, the conventional technology has the following problems.

  FIG. 15 is an explanatory diagram showing an image of a target sound and an interference sound in the automobile A.

  MVB can suppress only one disturbance sound, which is one less than the number of microphones. Therefore, when the target sound is emphasized by two microphones as shown in FIG. 14, the interference sound propagates as shown in FIG. 15B, so that MVB can suppress the direct sound of the interference sound but cannot suppress the reflected sound. Therefore, the target sound cannot be emphasized sufficiently.

  With the technique described in Non-Patent Document 1, all voices coming from other than the front direction are suppressed even if they originate from the target sound. Therefore, when the target sound is emphasized by two microphones as shown in FIG. 14, the target sound propagates as shown in FIG. 15 (a). Therefore, the technique described in Non-Patent Document 1 provides a reflected sound of the target sound. Since the sound is suppressed, the sound quality of the target sound is deteriorated.

  Therefore, it is possible to provide a signal processing apparatus, program, and method that emphasizes a target sound with less calculation cost and less distortion.

  The signal processing apparatus according to the first aspect of the present invention includes: (1) first frequency analysis means for obtaining a first input spectrum by performing frequency analysis on the first input signal input from the first sound collection device; 2) second frequency analysis means for obtaining a second input spectrum by frequency analysis of the second input signal inputted from the second sound collecting device; and (3) obtained by the first frequency analysis means. A straight line connecting the position of the first sound pickup device and the position of the second sound pickup device based on the first input spectrum and the second input spectrum obtained by the second frequency analysis means; A first feature value is calculated which takes a value in the front direction and the direction on the first sound collecting device side larger and a value in the direction on the second sound collecting device side smaller than the vertical front direction. And (4) the first feature amount calculated by the feature amount calculation unit. Filter determining means that obtains an enhancement filter by mapping with a predetermined monotonically increasing function in a broad sense; and (5) the enhancement filter obtained by the filter decision means on the first input spectrum obtained by the first frequency analysis means. And multiplication means for obtaining an enhanced spectrum by multiplying by.

  A signal processing program according to a second aspect of the present invention provides a computer, (1) first frequency analysis means for obtaining a first input spectrum by performing frequency analysis on a first input signal input from a first sound collection device. And (2) second frequency analysis means for obtaining a second input spectrum by performing frequency analysis on the second input signal input from the second sound collection device, and (3) the first frequency analysis means. Based on the first input spectrum obtained in step 2 and the second input spectrum obtained by the second frequency analysis means, the position of the first sound collecting device and the position of the second sound collecting device are connected. A first feature is that the values of the front direction and the direction of the first sound collector are larger and the value of the direction of the second sound collector is smaller than the front direction perpendicular to the straight line. A feature amount calculating means for calculating the amount; and (4) calculated by the feature amount calculating means. Filter determining means for mapping the first feature value by a predetermined monotonically increasing function in a broad sense to obtain an enhancement filter; and (5) the first input spectrum obtained by the first frequency analyzing means in the first input spectrum Multiplication means for obtaining an enhancement spectrum by multiplying the enhancement filter obtained by the filter determination means; (6) Waveform restoration means for obtaining the enhancement speech by inputting the enhancement spectrum obtained by the multiplication means and restoring the signal waveform. It is made to function.

  A signal processing method according to a third aspect of the present invention is the signal processing method, comprising: (1) first frequency analysis means, second frequency analysis means, feature amount calculation means, filter determination means, and multiplication means ( 2) The first frequency analysis means obtains a first input spectrum by performing frequency analysis on the first input signal inputted from the first sound collecting device, and (3) the second frequency analysis means. Obtains a second input spectrum by performing frequency analysis on the second input signal inputted from the second sound collecting device, and (4) the feature amount calculating means is obtained by the first frequency analyzing means. A straight line connecting the position of the first sound collecting device and the position of the second sound collecting device based on the obtained first input spectrum and the second input spectrum obtained by the second frequency analyzing means. And the first sound collecting device with respect to the front direction perpendicular to the front direction. The first feature value is calculated by taking a larger value in the direction of the second sound and taking a smaller value in the direction on the second sound collecting device side. (5) The filter determining means is calculated by the feature value calculating means. The first feature value is mapped with a predetermined broad monotonically increasing function to obtain an enhancement filter. (6) The multiplying unit adds the first input spectrum obtained by the first frequency analyzing unit to the first input spectrum. The enhancement spectrum obtained by the filter determination means is multiplied to obtain an enhancement spectrum.

  ADVANTAGE OF THE INVENTION According to this invention, the signal processing apparatus, program, and method which emphasize a target sound with less calculation cost and less distortion can be provided.

It is the block diagram shown about the functional structure of the signal processing apparatus which concerns on 1st Embodiment. It is explanatory drawing shown about the example of the usage environment of the signal processing apparatus which concerns on 1st Embodiment. The example of the feature-value _Fcenter processed with the signal processing apparatus which concerns on 1st Embodiment is shown. An example of the feature value F _side processed by the signal processing device according to the first embodiment is shown. It is the graph shown about the relationship between DOA feature-value F and the arrival direction (theta) of the sound processed with the signal processing apparatus which concerns on 1st Embodiment. It is the graph shown about the example of the broad sense monotone increase function processed with the signal processing apparatus which concerns on 1st Embodiment. It is the graph shown about the example of the emphasis filter used with the signal processor concerning a 1st embodiment. It is the graph shown about the example of the emphasis filter G obtained by the filter determination means which concerns on 1st Embodiment. It is the graph shown about the example of the emphasis filter G obtained by the filter determination means which concerns on 2nd Embodiment. It is the graph shown about the comparison with the emphasis filter G obtained by the filter determination means which concerns on 3rd Embodiment, and the emphasis filter G obtained by the filter determination means which concerns on 2nd Embodiment. It is the graph shown about the relationship between DOA feature-value F 'and the arrival direction (theta) of the sound processed with the signal processing apparatus which concerns on 4th Embodiment. It is explanatory drawing shown about the example of the emphasis filter G obtained by the filter determination means 404 which concerns on 4th Embodiment. It is a block diagram which shows the structure which concerns on the conventional subtraction type beam former in case the number of microphones is two. It is explanatory drawing shown about the example which emphasizes the audio | voice of the driver in a motor vehicle using the conventional signal processing apparatus. It is explanatory drawing shown about the image of the target sound and disturbance sound in a motor vehicle.

(A) First Embodiment A signal processing apparatus, program, and method according to a first embodiment of the present invention will be described in detail below with reference to the drawings.

(A-1) Configuration of First Embodiment FIG. 2 is an explanatory diagram showing an environment in which the signal processing apparatus 100 according to the first embodiment is used. In FIG. 2, the reference numerals in parentheses are those used in the second to fourth embodiments described later.

  The signal processing apparatus 100 according to the first embodiment is an explanatory diagram illustrating an example in which the voice of the driver U1 in the automobile A is emphasized. In the car A, the driver U1 sits in the driver's seat and the assistant U2 sits in the passenger seat. In the automobile A, the first microphone M1 and the second microphone M2 constituting the microphone array are arranged in front of the driver U1 (front of the driver's seat). When viewed from the driver U1, the first microphone M1 is disposed on the left side (the side opposite to the assistant U2), and the second microphone M2 is disposed on the right side (the assistant U2 side).

  FIG. 1 is a block diagram illustrating a functional configuration of the signal processing apparatus 100 according to the first embodiment.

  The signal processing apparatus 100 according to the first embodiment includes a first frequency analysis unit 101, a second frequency analysis unit 102, a feature amount calculation unit 103, a filter determination unit 104, a multiplication unit 105, and a waveform restoration unit 106. doing.

  A part or all of the signal processing apparatus 100 may be configured by software. For example, the signal processing apparatus 100 may be configured by installing a program (including the signal processing program according to the embodiment) in a computer having a memory and a processor.

  The first frequency analysis means 101 obtains a first input spectrum X1 by performing frequency analysis on the first input signal x1.

  The second frequency analysis means 102 analyzes the frequency of the second input signal x2 to obtain a second input spectrum X2.

  The feature amount calculation means 103 obtains a predetermined feature amount (hereinafter referred to as “DOA feature amount F”) based on the first input spectrum X1 and the second input spectrum X2. The DOA feature value F is a feature value that changes according to the direction of arrival of the target sound, and will be described in detail later.

  The feature quantity calculation means 103 can obtain the DOA feature quantity F by the formula (1) or a calculation formula obtained by transforming the formula (1).

In the formula (1), at a certain frequency at a certain time, the first input spectrum is X ₁ , the second input spectrum is X ₂ , and the complex conjugate of the second input spectrum is X ₂ ^* .

  The filter determination unit 104 obtains an enhancement filter G by mapping the DOA feature amount F with a predetermined broad-sense monotone increasing function.

  Multiplication means 105 multiplies first input spectrum X1 by enhancement filter G to obtain enhancement spectrum Y.

  The waveform restoration means 106 restores the signal waveform based on the enhancement spectrum Y to obtain the enhanced speech y.

  Next, the DOA feature quantity obtained by the feature quantity calculation unit 103 and the design concept of the enhancement filter obtained by the filter determination unit 104 will be described.

  The enhancement filter suppresses the direct sound and reflected sound of the interference sound coming from the second microphone M2 side (interference sound side), and from the first microphone M1 side (including the target sound side and the front direction). It is necessary to give a characteristic that does not suppress the direct sound and reflected sound of the incoming target sound. Therefore, it is desired that the DOA feature value takes a large value when the sound comes from the first microphone M1 side and takes a small value when the sound comes from the second microphone M2 side. However, since the first microphone M1 side includes the front direction, such a feature is not symmetric with respect to the sound arrival direction, and there is no known feature amount having the feature.

Therefore, a feature value that takes a large value with respect to the front direction and a feature value that takes a large value with respect to the second microphone M2 side are considered. At a certain frequency at a certain time, the first input spectrum is set as X ₁ , the second input spectrum is set as X ₂ , and the complex conjugate of the second input spectrum is set as X ₂ ^*. a feature amount _{F center} represented by), consider a feature amount _{F side} of the formula (1-2).

Here, the front direction (the direction perpendicular to the straight line connecting the positions of the two microphones) is 0 degree, and the direction on the second microphone M2 side (the second microphone M2 viewed from the first microphone M1) is Assuming that +90 degrees, the spectrum of the sound source is S, the angular frequency is ω, the distance between two microphones is d, the direction of sound arrival is θ (theta), and the speed of sound is c, X ₁ and X ₂ are respectively expressed by Equation (2) And (3), and substituting Equations (2) and (3) into Equations (1-1) and (1-2), respectively, (3-1) and (3-2) The formula is obtained. FIGS. 3 and 4 show the relationship between the feature amounts F _center and F _side represented by the equations (3-1) and (3-2) with respect to the sound arrival direction θ, respectively.

FIG. 3 shows an example of the feature amount F _center .

In FIG. 3, the horizontal axis is the sound source arrival direction θ, and the vertical axis is the feature amount F _center . FIG. 3 is a graph showing the relationship between the arrival direction θ and the feature amount F _center when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz.

FIG. 4 shows an example of the feature amount F _side .

In FIG. 4, the horizontal axis is the arrival direction θ of the sound source, and the vertical axis is the feature amount F _side . FIG. 4 is a graph showing the relationship between the arrival direction θ and the feature amount F _side when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz.

From FIG. 3 and FIG. 4, F _center is a large value with respect to the arrival direction of 0 degrees, and F _side is a large value with respect to the second microphone M2 side with respect to 0 degrees. It is a small value with respect to the M1 side.

As described above, the DOA feature amount F includes the F _center having characteristics that have a large value with respect to the direction of the target sound (the direction of 0 degrees; the front direction) and the second microphone M2 side (that is, the sound source of the interference sound). It can be seen that the feature amount is obtained using F _side having a characteristic that is large with respect to the assistant U2 side microphone).

  Next, the relationship between the sound arrival direction and the DOA feature value will be described.

The DOA feature value is defined by equation (3-3).

  Substituting Equations (2) and (3) into Equation (3-3) gives Equation (4), and transforming Equation gives Equation (5). FIG. 5 shows the relationship between the sound arrival direction θ expressed by the equation (5) and the DOA feature amount F.

  FIG. 5 is a graph showing the relationship between the sound arrival direction θ and the DOA feature amount F.

  In FIG. 5, the horizontal axis is the arrival direction θ of the sound source, and the vertical axis is the DOA feature amount F. FIG. 5 is a graph showing the relationship between the arrival direction θ and the DOA feature amount F when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz.

  From FIG. 5, the DOA feature amount F is always F = 1 for the front direction, and F <1 for the second microphone M2 side (interference sound side). On the other hand, for the first microphone M1 side, F> 1 at θ close to 0 degrees of the low frequency and high frequency, and F <1 at the portion close to 90 degrees of the high frequency.

As described above, the DOA feature amount F has a feature that takes a large value when the sound comes from the first microphone M1 side and takes a small value when the sound comes from the second microphone M2 side. You can see that In other words, the DOA feature amount F has a peak in the direction from the front direction to the first microphone M1 side (the direction opposite to the interference sound from the assistant U2), and the second microphone M2 side from the direction in which the peak exists. It can be seen that there is a characteristic that the value becomes smaller as the direction is inclined.

  The enhancement filter can be obtained by mapping the DOA feature value with a predetermined broad monotone increasing function.

  FIG. 6 is a graph showing an example of the broad-sense monotone increasing function fmap (F).

  In FIG. 6, the horizontal axis represents the DOA feature value F and the vertical axis represents the enhancement filter G value.

As can be seen from FIG. 15, it is desired that the enhancement filter suppresses the sound coming from the second microphone M2 side and does not suppress the sound coming from the front direction and the first microphone M1 side. Therefore, for example, the broad-sense monotone increasing function fmap (F) is defined as shown in Equation (6). FIG. 6 shows an example of fmap (F) in which the microphone interval is 3 cm, the sound speed is 332 m / s, and F ₀ = 0.9.

  When the enhancement filter is obtained as G = fmap (F), the relationship between the sound arrival direction θ and the enhancement filter G is as shown in FIG.

  FIG. 7 is a graph showing an example of the enhancement filter.

  In FIG. 7, the horizontal axis is the value of the sound arrival direction θ, and the vertical axis is the value of the enhancement filter G. FIG. 7 is a graph showing the relationship between the direction of arrival θ and the value of the enhancement filter G when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz.

That is, when the DOA feature amount F is larger than a value slightly smaller than 1, the enhancement filter G is set to 1. Otherwise, the enhancement filter G is made smaller than 1, so that the enhancement filter has a desired characteristic, that is, an interference sound. The direct sound and reflected sound of the target sound are suppressed, but the direct sound and interference sound of the target sound are not suppressed.

The enhancement filter similar to the present invention can be obtained by calculating the arrival direction θ for each frequency from the first input spectrum and the second input spectrum, for example, but the inverse tangent function (atan, arctan, tan ^-1 written as such) it takes computational cost of calculating. Therefore, the present invention is superior from the viewpoint of calculation cost.

(A-2) Operation of First Embodiment Next, the operation (signal processing method of the embodiment) of the signal processing apparatus 100 of the first embodiment having the above-described configuration will be described with reference to FIG. .

Signal processing apparatus 100, first the input signal x ₁ and the second input signal x _{2 (input} signal in the time domain), performs target sound is emphasized, the output signal of the enhanced speech y (time region including the target sound source ).

The first frequency analysis means 101 and the second frequency analysis means are connected to the first input signal x ₁ by an arbitrary frequency analysis method represented by Fourier transform or an arbitrary band dividing means represented by a filter bank. the second input signal x ₂ is divided into K bands, respectively, to obtain the first input spectrum X ₁ and a second input spectrum X _2. Hereinafter, the first input spectrum and the second input spectrum are written as X ₁ (k) and X ₂ (k) when the band number (for example, k-th) needs to be clearly indicated, and the band number is clearly indicated. When it is not necessary to do this, they are simply expressed as X ₁ and X ₂ . First frequency analysis means 101 gives the first of the input spectrum X ₁ feature calculating unit 103 and the multiplication means 105 a resulting second frequency analysis means 102, the resulting second input spectrum X ₂ is given to the feature quantity calculation means 103. The input spectrum is provided to multiplier 105 is set to the first input spectrum X _1, is not limited thereto, be applied to a second input spectrum X ₂ a multiplication means 105 may, both The same effect is produced.

The feature amount calculation unit 103 calculates the DOA feature amount F by the expression (7) based on the first input spectrum X ₁ and the second input spectrum X ₂ and gives the calculated value to the filter determination unit 104. The calculation may be performed using the equation (7) as it is, but the equation may be modified because it includes a redundant operation. Since X ₁ and X ₂ are complex numbers, when this is rewritten as shown in equation (8) and substituted into equation (7) for rearrangement, equation (9) is obtained. By using equation (9) instead of equation (7), the number of multiplications can be reduced.

The filter determination unit 104 calculates the enhancement filter G based on the DOA feature amount F using a predetermined broad-sense monotone increasing function, and supplies the enhancement filter G to the multiplication unit 105. In the first embodiment, the same broad-sense monotone increasing function is used for all frequencies. As a predetermined broad-sense monotone increasing function fmap (F), for example, a function having one threshold value F ₀ defined by the equation (10) or a function having two threshold values F ₁ and F ₂ defined by the equation (11) , A sigmoid function having a scale F ₃ and an offset F ₀ defined by the equation (12) can be used.

  FIG. 8 is a graph showing an example of the enhancement filter G obtained by the filter determination unit 104 according to the first embodiment.

  FIG. 8A, FIG. 8B, and FIG. 8C show examples of the enhancement filter G obtained by the equations (10), (11), and (12), respectively. 8A, 8B, and 8C are graphs showing the relationship between the arrival direction θ and the enhancement filter G when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz. Yes. In FIG. 8A, FIG. 8B, and FIG. 8C, the horizontal axis represents the arrival direction θ of the sound source, and the vertical axis represents the value of the enhancement filter G (value corresponding to the arrival direction θ).

Here, F ₀ = 0.8, F ₁ = 0.7, F ₂ = 0.9, and F ₃ = 12. There is not much difference between the expression (10) and the expression (11) regarding the interference noise suppression performance. On the other hand, with respect to the distortion of the emphasized speech, the enhancement filter G obtained by the equation (10) has only a value of 0 or 1, and thus tends to generate musical noise. Since the switching of non-suppression is gentle, musical noise is unlikely to occur. The expression (12) is a characteristic obtained by further smoothing the expression (11), and further musical noise reduction effect and distortion reduction effect can be expected. If a slight increase in calculation cost is allowed, it is preferable to use the expression (12).

The multiplication unit 105 multiplies the input spectrum X ₁ by an enhancement filter G (enhancement gain) for each frequency, and gives the obtained enhancement spectrum Y to the waveform restoration unit 106.

  The waveform restoration unit 106 uses the waveform restoration method corresponding to the frequency analysis method or the band division method used in the first frequency analysis unit 101 and the second frequency analysis unit 102, and the enhanced spectrum given from the multiplication unit 105. The signal waveform is reconstructed based on Y, and the obtained enhanced speech y (enhanced signal) is output.

(A-3) Effects of First Embodiment According to the first embodiment, the following effects can be achieved.

  In the signal processing apparatus 100 according to the first embodiment, sound arriving from the direction of the second microphone M2 is suppressed, and sound arriving from the front direction and the direction of the first microphone M1 is not suppressed. When emphasizing the voice (target sound) of the driver U1, the target sound can be emphasized with less distortion.

  In other words, the signal processing apparatus 100 can design the enhancement filter G that suppresses the direct sound and the reflected sound of the interference sound with low calculation cost but does not suppress the direct sound and the reflected sound of the target sound. The effect is that the sound can be emphasized.

(B) Second Embodiment Hereinafter, a second embodiment of the signal processing apparatus, program and method according to the present invention will be described in detail with reference to the drawings.

(B-1) Configuration of Second Embodiment The signal processing apparatus 200 of the second embodiment will be described as being used in an environment as shown in FIG. 2 as in the first embodiment.

  Also, the internal configuration of the signal processing device 200 of the second embodiment can be shown using FIG. 1 described above.

  Below, the difference from 1st Embodiment is demonstrated about the signal processing apparatus 200 of 2nd Embodiment.

  In the first embodiment, the filter determination unit 104 applies the same broad-sense monotone increasing function fmap (F) to all frequencies to obtain the enhancement filter G. Therefore, as shown in FIG. Was different for each frequency. In particular, a phenomenon occurs in which the range of arrival directions that are not suppressed is widened at low frequencies. Therefore, in the second embodiment, a broad monotone increasing function that differs for each frequency is applied so that the same characteristics are obtained at any frequency.

  The configuration of the signal processing device 200 according to the second embodiment is the same as the configuration of the signal processing device 100 according to the first embodiment except that the filter determination unit 104 is replaced with a filter determination unit 204 as shown in FIG. It is.

(B-2) Operation | movement of 2nd Embodiment Next, operation | movement (signal processing method of embodiment) of the signal processing apparatus 200 of 2nd Embodiment which has the above structures is demonstrated.

  The operation of the signal processing device 200 of the second embodiment is the same as the operation of the signal processing device 100 of the first embodiment, except that the operation of the filter determination unit 204 is different from the filter determination unit 104.

  In the first embodiment, as shown in FIG. 7, when the arrival direction is tilted from the front direction toward the second microphone M2, the gain of the enhancement filter G is predetermined or less (for example, 0.5 or less) depending on the frequency. ) (Hereinafter referred to as “cutoff arrival angle”). In other words, in the first embodiment, there is variation in the range of the arrival direction that is not suppressed depending on the frequency. On the other hand, the filter determination unit 204 of the second embodiment absorbs this variation by setting a broad-sense monotonically increasing function for each frequency, and cut-off arrival angles (in the direction of arrival without suppression) at a plurality of frequencies. Range) is approaching. The filter determining means 204 is not limited to a method for obtaining a broad monotone increasing function for each frequency so as to suppress variation in cutoff arrival angle for each frequency (variation in the range of arrival directions that are not suppressed). For example, what kind of processing can be applied.

The filter determination unit 204 calculates the enhancement filter G based on the DOA feature amount F using a predetermined broad-sense monotone increasing function, and supplies the enhancement filter G to the multiplication unit 105. In the second embodiment, a broad-sense monotone increasing function that differs for each frequency is used. Here, the DOA feature quantity of the kth frequency is written as F (k), and the enhancement gain of the kth frequency is written as G (k). A function for converting the direction of arrival and the frequency number into a DOA feature amount is defined by equation (13), where the k-th frequency is f _k . Then, as a broad monotone increasing function fmap _k (F (k)) of a predetermined k-th frequency, for example, one arrival direction threshold value θ ₀ defined by equation (14) is defined by a function or equation (15). A function having _two arrival direction threshold values θ ₁ and θ ₂ , a scale F _a defined by the equation (16), and a sigmoid function having an offset arrival direction θ ₀ can be used.

  FIG. 9 is a graph showing an example of the enhancement filter G obtained by the filter determination unit 204 according to the second embodiment.

  FIG. 9A, FIG. 9B, and FIG. 9C show examples of the enhancement filter G obtained by the equations (14), (15), and (16), respectively. 9A, 9B, and 9C are graphs showing the relationship between the arrival direction θ and the enhancement filter G when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz. Yes. 9A, 9B, and 9C, the horizontal axis represents the arrival direction θ of the sound source, and the vertical axis represents the value of the enhancement filter G (value corresponding to the arrival direction θ).

Here, θ ₀ = 15, θ ₁ = 20, θ ₂ = 10, and F _a = 12. In the enhancement filter G (FIG. 8) in the first embodiment, the range of arrival directions not to be suppressed changes for each frequency, but in the enhancement filter G (FIG. 9) in the second embodiment, the high-frequency Except for one microphone M1 side, the range of the arrival direction that is not suppressed does not change even if the frequency changes. Note that although the characteristic of the equation (16) changes depending on the frequency, the cutoff arrival angle at which the gain of the enhancement filter G becomes 0.5 is constant regardless of the frequency. That is, if the enhancement gain is calculated using the equations (13) to (16), the number of times to suppress the second microphone M2 side, that is, the disturbance sound side (passenger seat side), is the same for all frequencies. Can be set directly.

(B-3) Effects of Second Embodiment According to the second embodiment, the following effects can be obtained in addition to the effects of the first embodiment.

  In the signal processing device 200 according to the second embodiment, the range of the arrival direction in which the enhancement gain is not suppressed can be given in the same way at all frequencies, and the range can be set by the angle of the arrival direction itself. Adjustment is possible, and the target sound can be emphasized with less distortion.

(C) Third Embodiment Hereinafter, a third embodiment of the signal processing apparatus, program and method according to the present invention will be described in detail with reference to the drawings.

(C-1) Configuration of Third Embodiment The signal processing apparatus 200 of the third embodiment will be described as being used in an environment as shown in FIG. 2 as in the first and second embodiments. .

  Further, the internal configuration of the signal processing apparatus 300 of the third embodiment can also be shown using FIG. 1 described above.

  Hereinafter, differences from the second embodiment will be described for the signal processing device 200 of the third embodiment.

  In the second embodiment, how many times the direction of arrival on the disturbing sound side is suppressed is set for all frequencies. However, it is difficult to make the microphone interval sufficiently wide with respect to the signal wavelength at a low frequency (the wavelength of 100 Hz is about 3.3 m, but the microphone interval in an automobile is generally several centimeters). Information regarding the direction of arrival obtained by numerical calculation at a low frequency (DOA feature amount in the present invention) is generally ambiguous (estimation error increases in the sense of direction of arrival estimation). Therefore, in the third embodiment, the range of the arrival direction in which the enhancement gain is not suppressed is expanded at a frequency lower than a predetermined frequency (for example, a frequency band of 250 Hz or less) (expanded to the second microphone M2 side; a disturbing sound is emitted). It is designed to spread to the side of the assistant U2.

  The configuration of the signal processing device 300 according to the third embodiment is the same as the configuration of the signal processing device 100 according to the first embodiment except that the filter determination unit 104 is replaced with the filter determination unit 304.

(C-2) Operation of Third Embodiment Next, the operation (signal processing method of the embodiment) of the signal processing device 300 of the third embodiment having the above-described configuration will be described.

  The operation of the signal processing device 300 of the third embodiment is the same as the operation of the signal processing device 100 of the first embodiment, except that the operation of the filter determination unit 304 is different from the filter determination unit 104.

The filter determining unit 304 calculates the enhancement filter G based on the DOA feature amount F using a predetermined broad-sense monotone increasing function, and supplies the enhancement filter G to the multiplication unit 105. In the second embodiment, a broad-sense monotone increasing function that differs for each frequency is used. Here, the DOA feature quantity of the kth frequency is written as F (k), and the enhancement gain of the kth frequency is written as G (k). A function for converting the direction of arrival and the frequency number into a DOA feature is defined by equation (17), where the k-th frequency is f _k . A function having one arrival direction threshold value θ ₀ defined by, for example, the equation (18) can be used as the broad-sense monotone increasing function fmap _k (F (k)) of a predetermined k-th frequency.

  FIG. 10 is a graph showing a comparison between the enhancement filter G obtained by the filter determination unit 204 according to the second embodiment and the enhancement filter G obtained by the filter determination unit 304 according to the third embodiment.

  FIG. 10A shows an enhancement filter G obtained by using the above equations (13) and (14) by the filter determination means 204 in the second embodiment. FIG. 10B shows an example of the enhancement filter G obtained by the equations (17) and (18) by the filter determination unit 304 according to the third embodiment. 10A and 10B are graphs showing the relationship between the arrival direction θ and the enhancement filter G when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz. 10A and 10B, the horizontal axis is the arrival direction θ of the sound source, and the vertical axis is the value of the enhancement filter G (value corresponding to the arrival direction θ).

Here, θ ₀ = 5 and F ₀ = 0.97. From FIG. 10, it can be confirmed that by setting the upper limit value of the function Φ (Phi) for converting the arrival direction and the frequency number to the DOA feature amount as F ₀ , the range of the arrival direction in which low frequencies are not suppressed is widened. .

(C-3) Effects of the Third Embodiment According to the third embodiment, the effects of the first and second embodiments can be added.

  In the signal processing device 300 according to the third embodiment, since the range of the arrival direction that is not suppressed can be secured widely at a low frequency where the information about the arrival direction obtained by numerical calculation is ambiguous, distortion of the target sound at a low frequency is prevented. This reduces the effect of enhancing the target sound with less distortion.

(D) Fourth Embodiment Hereinafter, a fourth embodiment of the signal processing apparatus, program and method according to the present invention will be described in detail with reference to the drawings.

(D-1) Configuration of the Fourth Embodiment The signal processing apparatus 400 of the fourth embodiment will be described as being used in an environment as shown in FIG. 2 as in the first to third embodiments. .

  Further, the internal configuration of the signal processing apparatus 400 of the fourth embodiment can also be shown using FIG. 1 described above.

  Below, the difference with the 1st-3rd embodiment is demonstrated about the signal processing apparatus 400 of 4th Embodiment.

  In the first to third embodiments, assuming that two microphones M1 and M2 are set in front of the driver U1 in the automobile A, the enhancement filter G that suppresses only the passenger seat side (the assistant U2 side). Designed. On the other hand, in the fourth embodiment, an enhancement filter that enhances (does not suppress) only the front direction using the DOA feature value in the present invention is applied.

  As shown in FIG. 1, the configuration of the signal processing apparatus 400 according to the fourth embodiment is the same as that of the fourth embodiment except that the feature amount calculation unit 103 and the filter determination unit 104 are replaced with the feature amount calculation unit 403 and the filter determination unit 404, respectively. The configuration is the same as that of the signal processing apparatus 100 of the first embodiment.

(D-2) Operation of Fourth Embodiment Next, an operation (signal processing method of the embodiment) of the signal processing device 400 of the fourth embodiment having the above-described configuration will be described.

  Next, the operation of the signal processing apparatus 400 according to the fourth embodiment having the above-described configuration will be described. The operation of the signal processing apparatus 400 of the fourth embodiment is the same as that of the first embodiment, except that the operation of the feature amount calculation unit 403 and the filter determination unit 304 is different from the feature amount calculation unit 103 and the filter determination unit 104. The operation is the same as that of the signal processing apparatus 100.

The feature quantity calculation means 403 calculates two DOA feature quantities F and F ′ by the equation (19) based on the first input spectrum X ₁ and the second input spectrum X _2, and gives them to the filter determination means 404. . When the two DOA feature quantities are arranged with respect to the sound arrival direction θ, the equation (20) is obtained.

  FIG. 11 is a graph showing the relationship between the sound arrival direction θ and the DOA feature value F ′ in the equation (20).

  FIG. 11 is a graph showing the relationship between the arrival direction θ and the DOA feature amount F ′ when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz. In FIG. 11, the horizontal axis represents the sound source arrival direction θ, and the vertical axis represents the DOA feature amount F ′.

When FIG. 11 is seen, it can be confirmed that the left and right are just reversed from the DOA feature amount F (FIG. 5).

The filter determination unit 404 calculates the enhancement filter G based on the two DOA feature amounts F and F ′ using a predetermined broad-sense monotone increasing function, and supplies the enhancement filter G to the multiplication unit 105. The predetermined broad monotone increasing function includes fmap (F) according to the first embodiment, Φ (φ, k) and fmap _k (F (k)) according to the second embodiment, and the third embodiment. Any of Φ (φ, F ₀ , k) and fmap _k (F (k)) may be used, but here, as an example, a description will be given using the predetermined broad-sense monotonically increasing function of the second embodiment. . In the fourth embodiment, the enhancement filter G is calculated using equation (21).

  FIG. 12 is an explanatory diagram illustrating an example of the enhancement filter G obtained by the filter determination unit 404 according to the fourth embodiment.

12 (a), 12 (b), and 12 (c) are obtained when Equation (14), Equation (15), and Equation (16) are used as fmap _k (F (k)), respectively. An example of the enhancement filter G is shown. 12A, 12B, and 12C are graphs showing the relationship between the arrival direction θ and the enhancement filter G when the frequency of the sound source is changed to 1 kHz, 2 kHz, and 4 kHz. Yes. 12 (a), 12 (b), and 12 (c), the horizontal axis represents the arrival direction θ of the sound source, and the vertical axis represents the value of the enhancement filter G (value corresponding to the arrival direction θ).

Here, θ ₀ = 20, θ ₁ = 15, θ ₂ = 25, and F _a = 12. From FIG. 12, it can be seen that an enhancement filter that enhances only the front direction (does not suppress) is obtained.

(D-3) Effects of the Fourth Embodiment According to the fourth embodiment, the following effects can be achieved as compared with the first to third embodiments.

  In the signal processing device 400 of the fourth embodiment, it is possible to achieve a specific effect that the target sound in which the range of the arrival direction that is not suppressed by the enhancement filter G (enhancement gain) is limited to the front direction can be enhanced.

(E) Other Embodiments The present invention is not limited to the above-described embodiments, and may include modified embodiments as exemplified below.

  (E-1) In the above embodiment, the signal processing apparatus is described as restoring the waveform of the enhanced spectrum Y and outputting the enhanced speech y, but outputs the enhanced spectrum Y without restoring the waveform. Also good. Further, both the enhanced spectrum Y and the enhanced sound y may be output. In that case, the waveform restoration means 106 may be excluded.

  DESCRIPTION OF SYMBOLS 100 ... Signal processing apparatus, 101 ... 1st frequency analysis means, 102 ... 2nd frequency analysis means, 103 ... Feature-value calculation means, 104 ... Filter determination means, 105 ... Multiplication means, 106 ... Waveform restoration means, M1 ... First microphone (first sound collecting device), M2... Second microphone (second sound collecting device).

Claims (10)

First frequency analysis means for obtaining a first input spectrum by performing frequency analysis on the first input signal input from the first sound collection device;
Second frequency analysis means for obtaining a second input spectrum by performing frequency analysis on the second input signal input from the second sound collecting device;
Based on the first input spectrum obtained by the first frequency analysis means and the second input spectrum obtained by the second frequency analysis means, the position of the first sound collecting device and the second input spectrum are obtained. The front direction and the direction on the first sound collection device side are set larger than the front direction perpendicular to the straight line connecting the positions of the sound collection devices, and the value on the direction on the second sound collection device side. A feature amount calculating means for calculating a first feature amount that takes a small value;
Filter determining means for mapping the first feature quantity calculated by the feature quantity calculating means with a predetermined broad-sense monotone increasing function to obtain an enhancement filter;
A signal processing apparatus comprising: multiplication means for obtaining an enhanced spectrum by multiplying the first input spectrum obtained by the first frequency analyzing means by the enhancement filter obtained by the filter determining means.   The first feature amount is located on a side of the first sound collecting device with respect to a front direction perpendicular to a straight line connecting the position of the first sound collecting device and the position of the second sound collecting device. The signal processing apparatus according to claim 1, wherein a peak is present in a direction, and the value decreases as the direction inclines from the peak direction toward the second sound collector.   The feature quantity calculation means includes a second feature quantity that has a large value with respect to the front direction, and a third feature quantity that has a greater value with respect to the direction of the second sound collecting device than the front direction. The signal processing apparatus according to claim 1, wherein the first feature amount is calculated using a signal.   The said filter determination means maps the said 1st feature-value using the broad-sense monotone increasing function which changes for every frequency, The said emphasis filter is obtained for every frequency, The Claim 1 characterized by the above-mentioned. Signal processing equipment.   5. The signal processing apparatus according to claim 4, wherein the filter determination unit sets a broadly monotonically increasing function of each frequency so that the range of arrival directions not to be suppressed matches in the enhancement filter for each frequency. .   The filter determination means sets a broadly monotonically increasing function that widens the range of arrival directions that are not suppressed in the enhancement filter in a low frequency band below a predetermined frequency compared to a high frequency band higher than the predetermined frequency. The signal processing device according to claim 4.   The filter determination means sets a broadly monotonically increasing function that widens the range of the arrival direction that is not suppressed in the enhancement filter in the low frequency band toward the second sound collecting device side than in the high frequency band. The signal processing apparatus according to claim 6, characterized in that:   4. The signal according to claim 1, wherein the filter determination unit sets a broadly monotonically increasing function that obtains an enhancement filter that emphasizes only the front direction using the first feature amount. 5. Processing equipment. Computer
First frequency analysis means for obtaining a first input spectrum by performing frequency analysis on the first input signal input from the first sound collection device;
Second frequency analysis means for obtaining a second input spectrum by performing frequency analysis on the second input signal input from the second sound collecting device;
Based on the first input spectrum obtained by the first frequency analysis means and the second input spectrum obtained by the second frequency analysis means, the position of the first sound collecting device and the second input spectrum are obtained. The front direction and the direction on the first sound collection device side are set larger than the front direction perpendicular to the straight line connecting the positions of the sound collection devices, and the value on the direction on the second sound collection device side. A feature amount calculating means for calculating a first feature amount that takes a small value;
Filter determining means for mapping the first feature quantity calculated by the feature quantity calculating means with a predetermined broad-sense monotone increasing function to obtain an enhancement filter;
Multiplying means for multiplying the first input spectrum obtained by the first frequency analyzing means by the enhancement filter obtained by the filter determining means to obtain an enhanced spectrum;
A signal processing program that functions as a waveform restoration unit that receives an enhanced spectrum obtained by the multiplication unit and restores a signal waveform to obtain enhanced speech. In the signal processing method,
A first frequency analysis unit, a second frequency analysis unit, a feature amount calculation unit, a filter determination unit, and a multiplication unit;
The first frequency analysis means obtains a first input spectrum by performing frequency analysis on the first input signal input from the first sound collection device,
The second frequency analysis means obtains a second input spectrum by performing frequency analysis on the second input signal input from the second sound collection device,
The feature amount calculating means is based on the first input spectrum obtained by the first frequency analyzing means and the second input spectrum obtained by the second frequency analyzing means. With respect to the front direction perpendicular to the straight line connecting the position of the second sound collecting device and the position of the second sound collecting device, the values in the front direction and the direction on the first sound collecting device side are set larger. Calculating a first feature value that takes a smaller value in the direction of the sound device;
The filter determination unit maps the first feature amount calculated by the feature amount calculation unit with a predetermined broad monotone increasing function to obtain an enhancement filter,
The signal processing method, wherein the multiplication means obtains an enhanced spectrum by multiplying the first input spectrum obtained by the first frequency analysis means by the enhancement filter obtained by the filter determination means.

Images