JP2017059956A - Sound source extraction system and sound source extraction method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a sound source extraction system capable of extracting the sound propagating from a target sound source reliably, regardless of the target sound source and an interfering sound source existing in the extraction region and the directivity.SOLUTION: A sound source extraction system calculates a matrix including a plurality of transfer functions of sound propagation, from each unit region associated previously with the position of each sound source (30, 31a, 31b), on the basis of the output signals from a plurality of microphones (1a, 1b) for collecting the sound propagating from an extraction region being divided into a plurality of unit regions, determines one inverse matrix obtained therefrom, and extracts the sound propagating from the target sound source (30) by using this inverse matrix.SELECTED DRAWING: Figure 6

Description

  The present invention relates to a sound source extraction system and a sound source extraction method for extracting sound propagating from a target sound source using a plurality of microphones.

  In general, various techniques for extracting only sound propagating from a target specific sound source from a space where various sound sources exist are known (see Patent Document 1). Among these, in particular, the beam forming method uses a microphone array having a plurality of microphones, collects the sound propagated from the target sound source existing at a predetermined position, and specifies the direction in which the target sound source is located by calculation processing. It is possible to extract the target sound source separately from other sound sources. By applying such a beam forming method, it is possible to follow not only when the target sound source is stationary but also when the target sound source is moving.

JP2013-183358

  When the beam forming is applied, if a disturbing sound source exists in the same direction as the target sound source when viewed from the microphone array, a situation in which the target sound source and the disturbing sound source cannot be separated may occur. In this case, by narrowing the width of the beam from the microphone array toward the target sound source, it is possible to obtain an effect of increasing the spatial resolution of the sound by separating the target sound source from the disturbing sound source. However, narrowing the beam width by the beam forming method is limited by various parameters of the microphone and sound, and an increase in the amount of calculation for increasing the spatial resolution is inevitable. In addition, it is assumed that a large number of microphone arrays are installed to monitor the target sound source from various directions, but even in this case, each microphone array independently extracts the target sound source, It is difficult to avoid the above-mentioned problem when there is an interference sound within the range of the width.

  The present invention has been made to solve these problems, and the sound source extraction that can ensure better performance than the case where the beam forming method is applied by the arithmetic processing when extracting the sound from the target sound source. The purpose is to provide a system.

  In order to solve the above problems, a sound source extraction system of the present invention is a sound source extraction system that extracts sound propagating from a target sound source, and is distributed and arranged outside a predetermined extraction region divided into a plurality of unit regions. A plurality of microphones collecting sound propagating from one or more sound sources including the target sound source, and the output signals of the plurality of microphones are associated with the positions of the one or more sound sources in advance. A matrix including a plurality of transfer functions of sound propagation from each of the plurality of unit regions to each of the plurality of microphones as an element is calculated, and one inverse matrix is obtained from the matrix, and the inverse matrix is used. And an arithmetic means for extracting a sound generated from the target sound source.

  According to the sound source extraction system of the present invention, the extraction region where the target sound source and the disturbing sound source are present is divided into a plurality of unit regions, and sound propagation is performed in advance for each position of each unit region based on each output signal of the plurality of microphones. The sound that propagates from the target sound source is extracted by obtaining a transfer function and using one inverse matrix that gives its inverse characteristics. Therefore, even in a situation where the target sound source and the disturbing sound source exist in the same direction from the position of each microphone, it is distributed over the entire extraction region without performing complicated calculations such as sound source separation in the conventional beam forming method. The target sound source can be reliably separated and extracted from the disturbing sound source based on the output signal of the microphone.

  In the present invention, a plurality of microphone arrays each having a predetermined number of the microphones included in the plurality of microphones and arranged at different positions outside the extraction region can be further provided. In this case, the plurality of microphone arrays in the sound source extraction system constitute a so-called array of arrays. For example, when each of the L microphone arrays has M microphones, L × M microphones are installed in total. Even with such an arrangement, since one inverse matrix is generated in the entire system, the effect of separating the target sound source and the disturbing sound source existing in the same direction as described above can be obtained, and in addition, it can be easily installed. Highly useful in terms. Note that it is desirable that the plurality of microphone arrays be dispersedly arranged in the vicinity of the outer edge portion without being biasedly arranged outside the extraction region.

  In the present invention, a spherical microphone array in which the predetermined number of microphones are arranged on the surface of a spherical baffle can be used as the plurality of microphone arrays. The spherical micron array is advantageous in that it can be made compact, and it is possible to perform the operation for generating the inverse matrix relatively easily. Moreover, it becomes robust as a system.

  In order to solve the above-described problem, the sound source extraction method of the present invention is a sound source extraction method for extracting sound propagating from a target sound source, and is distributed outside a predetermined extraction area divided into a plurality of unit areas. A sound collecting step of collecting sounds propagating from one or more sound sources including the target sound source using the plurality of microphones, and the one or more sound sources in advance based on respective output signals of the plurality of microphones A matrix including a plurality of transfer functions of sound propagation from each of the plurality of unit regions associated with the position to each of the plurality of microphones as an element, and one inverse matrix obtained from the matrix is calculated. And a calculation step for extracting a sound propagated from the target sound source using the inverse matrix.

  According to the sound source extraction method of the present invention, it is possible to achieve the same operational effects as those of the sound source extraction system described above. Further, a configuration in which a plurality of microphone arrays are further provided can be applied in the same manner as described above. In the sound source extraction method of the present invention, it is preferable that the target sound source is extracted using a spatial window function corresponding to the arrangement of the plurality of microphones in the calculation step.

  According to the present invention, a plurality of microphones are dispersedly arranged outside an extraction region divided into a plurality of unit regions, and propagated from a target sound source using one inverse matrix obtained based on a plurality of transfer functions of sound propagation. Since the sound is extracted, it is possible to construct a highly reliable sound source extraction system by simple arithmetic processing while avoiding the influence of the direction of the target sound source and the disturbing sound source, which is a problem in the conventional beam forming method. It becomes possible.

It is a figure which shows the structure of the spherical microphone array which is the main components used with the sound source extraction system of this embodiment. It is a figure which shows the example of arrangement | positioning of the several microphone array in the sound source extraction system of this embodiment. It is a figure which shows an example of the functional block of the sound source extraction system corresponding to the example of arrangement | positioning of FIG. It is a flowchart which shows the flow of the process mainly relevant to the calculation of an inverse matrix among the arithmetic processes performed by the arithmetic processing part. It is a figure which shows typically the case where the conventional beam forming method is applied for the comparison with the case where calculation processing is applied in the sound source extraction system of this embodiment. It is a figure which shows typically the case where a calculation process is applied in the sound source extraction system of this embodiment. It is a figure explaining the verification result of the performance in the case of using the sound source extraction system of this embodiment.

  Embodiments of a sound source extraction system to which the present invention is applied will be described below with reference to the accompanying drawings. However, the embodiments described below are examples of forms to which the technical idea of the present invention is applied, and the present invention is not limited by the contents of the present embodiments.

  FIG. 1 shows the structure of a spherical microphone array (hereinafter simply referred to as “microphone array”) 1, which is a main component used in the sound source extraction system of the present embodiment. A microphone array 1 shown in FIG. 1 includes a spherical baffle 10 made of a hard material, a plurality of microphones 11 disposed at predetermined positions on the sphere surface of the baffle 10, and electrical signals output from the plurality of microphones 11. And a wiring section 12 in which a plurality of wirings for transmitting are stored.

As shown in the lower part of FIG. 1, the position in the space including the sound source extraction system of this embodiment is displayed in polar coordinates by converting the X, Y, and Z coordinates, and the azimuth angle θ, the elevation angle φ, and the distance r. . For example, the position of an arbitrary microphone 11 on the polar coordinates can be expressed as (θ _m , φ _m , r _m ). If the center of the spherical baffle 10 is assumed to be the origin, the microphone 11 is attached to one microphone array 1. are all microphone 11 will be set at equal distances r _m from each other.

  Each position of the plurality of microphones 11 included in one microphone array 1 is not limited, but a configuration similar to a general beam forming method can be employed. Further, if the number of the plurality of microphones 11 included in one microphone array 1 is too small, the accuracy is lowered. If the number is too large, the amount of computation necessary for the later-described computation increases. For example, 64 microphones 11 are attached to one microphone array 1.

ただし、ｋ：波数（＝２πｆ／ｃ）
ｒ^→ _ｍ：マイクロホンの位置べクトル
ｒ^→ _ｓ：音源の位置べクトル
ｐ_ｓ：音源の音圧
ｈ_ｎ：球ハンケル関数
ｈ’_ｎ：ｈ_ｎを微分した関数
Ｐ_ｎ：ｎ次ルジャンドル多項式 Here, the sound pressure _{p m} of each microphone 11 is expressed by the following equation (1).

Where k: wave number (= 2πf / c)
r ^→ _m : Microphone position vector
r ^→ _s : position vector of sound source
p _s : Sound pressure of the sound source
h _n : Spherical Hankel function
h ′ _n : a function obtained by differentiating h _n
P _n : n-th order Legendre polynomial

In the conventional method, the sound pressure p _m of the above-obtained for each of the microphones 11, while the extraction of the target sound source by using a so-called beam forming method, the signal extraction system of the present embodiment, conventional It is characteristic that the target sound source is extracted by a method different from the beam forming method. Details of this point will be described later.

  In the sound source extraction system of the present embodiment, a so-called array of arrays is configured by using a plurality of microphone arrays 1 of FIG. Each microphone 11 and each microphone array 1 must be sampling-synchronized. FIG. 2 shows an arrangement example of a plurality of microphone arrays 1 in the sound source extraction system of the present embodiment. In FIG. 2, for the sake of easy understanding, the sound source extraction system arranged in the area AA represented by the XY coordinates is assumed, but the actual sound source extraction system is configured in a three-dimensional space including the Z direction. In the example of FIG. 2, among the rectangular areas AA including the extraction area A, four microphone arrays 1 (a), 1 (b), which are symmetrically arranged at four corners outside the extraction area A, 1 (c) and 1 (d) are shown. In this case, when each microphone array 1 has N microphones 11, there are 4N microphones 11 in total. In addition, although the installation positions of the respective microphone arrays 1 can be freely determined, it is desirable to disperse them in the vicinity of the outer edge of the extraction region A so that the positions are not biased as much as possible.

  The extraction area A constitutes a number of grids G (unit areas of the present invention) by a group of straight lines arranged at equal intervals along the X and Y directions. Then, it is assumed that each of one or more sound sources including the target sound source to be extracted is arranged as a point sound source in any grid G of the extraction region A, and based on the position of the grid G, Calculation processing of sound source extraction by a transfer function reaching each microphone 11 is performed. Specific calculation processing will be described later. Here, the size and number of grids G in the extraction region A are not particularly limited, but are appropriately set according to the calculation amount and the spatial resolution. That is, in the extraction area A, if the grid G is too small, the amount of computation increases. If the grid G is too large, the spatial resolution is insufficient and separation of sound sources becomes difficult, so the grid G is set to an appropriate size. There is a need.

  FIG. 3 shows an example of functional blocks of a sound source extraction system corresponding to the arrangement example of FIG. In the sound source extraction system shown in FIG. 3, in addition to the four microphone arrays 1 (a), 1 (b), 1 (c), and 1 (d) shown in FIG. And an output unit 22. Among these, the AD conversion unit 20, the arithmetic processing unit 21, and the output unit 22 can be integrally configured by, for example, a personal computer that can be connected to the four wiring units 12 of the microphone array 1 described above.

  The 4N microphones 11 included in the four microphone arrays 1 collect the sounds propagated from the respective sound sources, convert them into analog signals Sa, and transmit them to the AD conversion unit 20 via the corresponding wiring units 12. . The AD converter 20 samples the 4N analog signals Sa output from the 4N microphones 11 at a predetermined sampling frequency, and converts them into 4N digital signals Sd. That is, in the AD conversion unit 20, a plurality of AD converters corresponding to at least the number of microphones 11 are arranged in parallel. The arithmetic processing unit 21 uses the digital signals Sd obtained by the AD conversion unit 20 to perform later-described arithmetic processing necessary for extracting the target sound source, and generates a signal S corresponding to the calculation result. The output unit 22 outputs the signal S output from the arithmetic processing unit 21 to a device outside the system or a storage unit or display unit inside the system.

Next, an outline of arithmetic processing in the sound source extraction system of the present embodiment will be described. FIG. 4 is a flowchart showing a flow of processing for calculating an inverse matrix in advance. Here, a reference sound source that generates a known output sound is arranged in a grid G existing in a predetermined extraction area A. As shown in FIG. 4, on the basis of the output signals of the microphones 11 (corresponding to the digital signal Sd in FIG. 3), the sound pressure p _m corresponding to the output signals of all the microphones 11 having the N pieces of the microphone array 1 Obtain (step S1). For example, if there are L number of microphones 11 in total, the L of the sound pressure p _m corresponding to each are obtained.

ただし、ｒ^→ _ｍ：マイクロホンの位置べクトル
ｒ^→ _ａ：マイクロホンアレイの位置べクトル
ｒ^→：音源の位置べクトル
Ｒ_ａ：バッフルの半径
ｈ_ｎ：球ハンケル関数
ｈ’_ｎ：ｈ_ｎを微分した関数
Ｐ_ｎ：ｎ次ルジャンドル多項式 Here, the sound pressure p _m in each of the microphones 11 in the case of using a microphone array 1 of spherical, as already described, is represented by the above formula (1). The sound from the reference sound source existing in the grid G is input to each microphone 11 through various paths. Therefore, with respect to each microphone 11 included in each microphone array 1, a transfer function of sound propagation from the position of the reference sound source (corresponding to the grid G in FIG. 2) can be obtained. Are sequentially calculated (step S2). In step S2, the transfer function H _am of each microphone 11 included in the predetermined microphone array 1 can be expressed by, for example, the following equation (2) in relation to the equation (1).

Where r ^→ _m : microphone position vector
r ^→ _a : Microphone array position vector
r ^→ : Sound source position vector
R _a : radius of baffle
h _n : Spherical Hankel function
h ′ _n : a function obtained by differentiating h _n
P _n : n-th order Legendre polynomial

ただし、ψ：各位置の音源の分布 Here, the total sound pressure p _am of each microphone 11 included in the predetermined microphone array 1 is actually represented by volume integration in the extraction region A as represented by the following equation (3).

Where ψ: Distribution of sound sources at each position

ただし、Ｈ：全ての伝達関数Ｈ_ａｍからなる行列
Λ：全てのグリッド点に音源があると仮定したときにそれぞれの音源が発生している音の大きさ（分布）
（４）式において、分布Λの要素は、抽出領域Ａ内の任意のグリッドＧにおける音エネルギーの合計を表している。 On the other hand, since the extraction area A of the present embodiment is divided into the grid G as described above, the number of calculations of the expression (3) increases or decreases according to a predetermined spatial resolution, and the calculation process is simplified by setting the grid G. can do. First, when an output vector S having elements of outputs of all microphones 11 included in all microphone arrays 1 is used, the following expression (4) is established.

Where H: matrix consisting of all transfer functions H _am
Λ: Sound volume (distribution) generated by each sound source when it is assumed that there are sound sources at all grid points
In the equation (4), the element of the distribution Λ represents the sum of sound energy in an arbitrary grid G in the extraction area A.

In the sound source extraction system of this embodiment, in order to obtain the above distribution Λ, a matrix H having all transfer functions H _am as elements is obtained from the equation (4), and an inverse matrix H ⁻¹ is obtained in advance. It is characteristic that an operation using this one inverse matrix H ⁻¹ is performed. Therefore, the above-described inverse matrix H ⁻¹ is generated based on all the transfer functions H _am obtained by the expression (2) (step S3). On the other hand, when obtaining the sound pressure p _t of the target sound source in the extraction region A, based on the output signals of all the microphones 11 to first all the microphone array 1 has, determines the output vector S.

ただし、Ｗ_ｔ：空間的窓関数 Then calculated based on the sound pressure p _t of the target source in the following equation (5).

Where W _t : spatial window function

Note that the spatial window function W _t used in the equation (5) is set as appropriate depending on the spatial resolution of the gridded extraction region A. As described above, the target sound source can be extracted in the extraction area A as a result of the calculation processing of the sound source extraction system of the present embodiment, and the target sound source can be reliably detected even when there is a disturbing sound source as will be described later. Separable.

  Here, in the sound source extraction system of the present embodiment, the effect when the above-described arithmetic processing is applied will be described with reference to FIGS. 5 and 6. FIG. 5 schematically shows a case where the conventional beam forming method is applied in a situation where there are two disturbing sound sources 31a and 31b in the same direction when the target sound source 30 is extracted by the two microphone arrays 1a and 1b. FIG. 6 schematically shows a case where the method according to the present invention is applied in the same situation as FIG. In any case, the target sound source 30 and the one disturbing sound source 31a are arranged in the direction of the beam Ba from the one microphone array 1a, and the target sound source 30 and the other disturbing sound source 31b in the direction of the beam Bb as seen from the other microphone array 1b. Are arranged, and both beams Ba and Bb are in a positional relationship orthogonal to each other.

  First, when the conventional method is applied as shown in FIG. 5, both the target sound source 30 and the disturbing sound source 31a exist in the range of the beam Ba by one microphone array 1a, and the beam by the other microphone array 1b. Both the target sound source 30 and the disturbing sound source 31b exist in the range of Bb. In the conventional method, each of the two microphone arrays 1a and 1b independently calculates the inverse characteristic of acoustic diffusion and extracts the target sound source 30, so that the directivity (beam width) of each beam Ba and Bb is determined. Due to the restrictions, it becomes difficult to separate the target sound source 30 from the disturbing sound sources 31a and 31b. In this case, it is not realistic to apply the complicated sound source separation algorithm to separate the target sound source 30 and the disturbing sound sources 31a and 31b because the amount of calculation increases.

  On the other hand, in FIG. 6, when applying the method according to the present invention, the target sound source is obtained using the above-described one inverse matrix calculated based on the outputs of all the microphones 11 of the respective microphone arrays 1a and 1b. 30 is extracted. Therefore, for example, when viewed from one microphone array 1a, the target sound source 30 and the disturbing sound sources 31a and 31b having different directions can be separated as shown in a virtual beam Bc in FIG. The same applies to the relationship between the target sound source 30 and the disturbing sound sources 31a and 31b when viewed from the other microphone array 1b. Therefore, as long as the target sound source 30 and many other disturbing sound sources are located on different grids G in the entire sound source extraction system, the target sound source 30 can be easily extracted without being affected by each disturbing sound source. It becomes possible.

  Next, performance simulation verification results when the sound source extraction system of this embodiment is used will be described with reference to FIG. For comparison with the present invention, FIG. 7A shows a case where a method related to sound source extraction is not applied, FIG. 7B shows a case where a conventional beam forming method is applied, and FIG. 7C shows a method according to the present invention. The verification result by each experiment when applying is shown. In any figure, the extraction performance of the target sound source 30 was compared in the situation where both the target sound source 30 and the disturbing sound source 31 exist. The time range on the horizontal axis was 1 second, and the vertical axis was the sound pressure normalized in the range of −1 to +1. As the sound source extraction system, two microphone arrays 1 separated from each other by 3 m from the center of the extraction region that is the target of sound source extraction were installed, and 64 microphones 11 were attached to each. The target sound source 30 was set to repeat a 25 ms white noise burst signal output period and a 25 ms signal stop period. The target sound source 30 and the disturbing sound source 31 are independent.

  As shown in FIG. 7A, based on the results when the above methods are not applied, the target sound source 30 and the disturbing sound source 31 were set to have the same sound pressure level peaks. Therefore, in FIG. 7A, the sound pressure level of the target sound source 30 is buried in the disturbing sound source 31. On the other hand, in FIG. 7B to which the conventional beam forming method is applied, the target sound source 30 can be separated from the disturbing sound source 31, but the SN ratio thereof is about 10.4 dB. On the other hand, in FIG. 7C to which the method according to the present invention is applied, the target sound source 30 can be separated from the disturbing sound source 31, and the SN ratio is about 18.6 dB, which is clearly compared with FIG. 7B. The improvement was confirmed.

  As described above, by adopting the sound source extraction system (sound source extraction method) to which the present invention is applied, the matrix of the transfer function from each sound source in the grid extraction region A to all the microphones 11 is inverted 1 By generating two inverse matrices, the target sound source can be extracted with good performance without being affected by the directionality of the disturbing sound source. In this case, the complicated sound source separation algorithm used in the conventional beam forming method is unnecessary, and the target sound source can be extracted by a simple arithmetic processing. In addition, the sound source extraction system to which the present invention is applied can extract not only when the target sound source is stationary but also when the target sound source is moving. Note that the present invention can be applied to a plurality of microphones 11 in the extraction region A even when the microphone array 1 is not configured. However, by using a predetermined number of microphone arrays 1, the corners of the extraction region A can be used. The effect that it can be installed easily is obtained.

  The contents of the present invention have been specifically described above based on the present embodiment, but the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the scope of the present invention. The main components (FIGS. 1, 2, and 3) and the calculation processing procedure (FIG. 4) of the above embodiment are not limited to the contents disclosed in the above embodiment, and the operation of the present invention. As long as the effect is obtained, it can be appropriately changed.

DESCRIPTION OF SYMBOLS 1 ... Microphone array 10 ... Baffle 11 ... Microphone 12 ... Wiring part 20 ... AD conversion part 21 ... Arithmetic processing part 22 ... Output part 30 ... Target sound source 31 ... Interference sound source A ... Extraction area G ... Grid

Claims (7)

A sound source extraction system that extracts sound propagating from a target sound source,
A plurality of microphones that are distributed outside a predetermined extraction region divided into a plurality of unit regions and that collect sound propagating from one or more sound sources including the target sound source;
A plurality of transfer functions of sound propagation from each of the plurality of unit regions previously associated with the positions of the one or more sound sources to each of the plurality of microphones based on output signals of the plurality of microphones Calculating a matrix including the element, obtaining one inverse matrix from the matrix, and using the inverse matrix to extract sound propagating from the target sound source,
A sound source extraction system comprising:   2. The sound source extraction system according to claim 1, further comprising a plurality of microphone arrays respectively provided with a predetermined number of the microphones included in the plurality of microphones and disposed at different positions outside the extraction region.   The sound source extraction system according to claim 2, wherein the plurality of microphone arrays are distributedly arranged in the vicinity of an outer edge portion of the extraction region.   The sound source extraction system according to claim 2 or 3, wherein the plurality of microphone arrays are spherical microphone arrays in which the predetermined number of microphones are arranged on a surface of a spherical baffle. A sound source extraction method for extracting sound propagating from a target sound source,
A sound collecting step of collecting sound propagating from one or more sound sources including the target sound source by using a plurality of microphones distributed outside a predetermined extraction region divided into a plurality of unit regions;
A plurality of transfer functions of sound propagation from each of the plurality of unit regions previously associated with the positions of the one or more sound sources to each of the plurality of microphones based on output signals of the plurality of microphones Calculating a matrix including the element, obtaining one inverse matrix obtained from the matrix, and extracting the sound propagating from the target sound source using the inverse matrix;
A sound source extraction method comprising:   The sound source extraction method according to claim 5, further comprising a plurality of microphone arrays respectively provided with a predetermined number of the microphones included in the plurality of microphones and disposed at different positions outside the extraction region. The sound source extraction method according to claim 6, wherein in the calculation step, the target sound source is extracted using a spatial window function corresponding to the arrangement of the plurality of microphones.

Images