JP5550019B2 - Sound field sharing system and optimization method - Google Patents

Description

  The present invention relates to a sound field sharing system and an optimization method, and more particularly to a sound field sharing system and an optimization method using a sound field control reproduction system that records and reproduces an original sound field physically and faithfully.

  An example of this type of conventional sound field sharing system is disclosed in Non-Patent Document 1. In the three-dimensional sound field communication system disclosed in Non-Patent Document 1, a sound field control (Bond Surface Control: BoSC) reproduction system that reproduces sound data recorded by a microphone array of 70 ch (channel) with a 62 ch loudspeaker is provided. It is possible for users in remote locations to have a conversation while sharing an acoustic space. Specifically, 62ch sound field data recorded in advance and convoluted with an inverse filter is stored in the server. In this server, two client machines (PCs) are arranged at different locations via a network such as the Internet and a LAN. A three-dimensional sound field reproduction system is connected to each client machine. The server simultaneously transmits the reproduction sound field selected by the user to both sound field reproduction systems. Voice data corresponding to the voice of the user of each sound field reproduction system is transmitted to the other client machine via the network. In each client machine, voice data (1ch) corresponding to the voice of the other user is convoluted in real time and then superimposed on the sound field data (62ch) and output. Therefore, users existing in different places can share the sound field data output from the server and have a conversation.

"1. Numerical analysis technology and visualization / audibility 1.7 Three-dimensional sound field communication system" Seigo Enomoto Acoustic Technology No.148 / Dec.2009 pp37-42

  However, in the three-dimensional sound field communication system of Non-Patent Document 1, the sound field data recorded by the 70ch microphone array is reproduced by the 62ch speaker array (sound field reproduction system). It is enormous. In addition, since the number of channels is large, the load of convolution processing is heavy. For this reason, sound field data that have been recorded in advance and subjected to convolution processing in advance are transmitted to each client machine. Therefore, it is difficult to share sound field data recorded in real time.

  Therefore, the main object of the present invention is to provide a novel sound field sharing system and optimization method.

  Another object of the present invention is to provide a sound field sharing system and an optimization method capable of sharing sound field data recorded in real time by users existing in different places.

  The present invention employs the following configuration in order to solve the above problems. The reference numerals in parentheses, supplementary explanations, and the like indicate correspondence relationships with embodiments described later to help understanding of the present invention, and do not limit the present invention in any way.

  According to a first aspect of the present invention, a sound field sharing system is arranged in a certain sound field, includes a microphone array (14) having a first predetermined number of microphones, sound field data detected by the microphone array, and records the sound field. A server (12) for transmitting data to a plurality of reproduction systems, and a reproduction system (22, 26) for reproducing sound field data from the server by a speaker array having a second predetermined number of loudspeakers. The sound field sharing system includes an initial speaker selection unit, a first evaluation value calculation unit, a reference speaker selection unit, a first execution unit, an initial microphone selection unit, a second evaluation value calculation unit, a reference microphone selection unit, and a second execution unit. Is provided. For example, these means are realized by a computer (12, 18, 20, etc.).

  The initial speaker selection means selects one loudspeaker in the speaker array as the first reference loudspeaker. The first evaluation value calculation means calculates Gram Schmidt orthogonalization evaluation values between the selected reference loudspeaker and each of the evaluation target loudspeakers other than the reference loudspeaker in the speaker array. The reference speaker selection means selects the evaluation target loudspeaker having the highest Gram Schmidt orthogonalization evaluation value calculated by the first evaluation value calculation means as the reference loudspeaker. The first executing means repeatedly executes the first evaluation value calculating means and the reference speaker selecting means until the number of reference loudspeakers becomes a third predetermined number smaller than the second predetermined number as a result of selection by the reference speaker selecting means. Let

  The initial microphone selection means selects one microphone of the microphone array as the first reference microphone. The second evaluation value calculation means calculates a Gramschmitt orthogonalization evaluation value between the selected reference microphone and all the evaluation target microphones other than the reference microphone in the microphone array. The reference microphone selection unit selects the evaluation target microphone having the highest Gram Schmitt orthogonalization evaluation value calculated by the second evaluation value calculation unit as the reference microphone. The second execution means repeatedly executes the second evaluation value calculation means and the reference microphone selection means until the number of reference microphones becomes a fourth predetermined number smaller than the first predetermined number as a result of selection by the reference microphone selection means. . Then, the server transmits the sound field data detected by the fourth predetermined number of reference microphones to the plurality of reproduction systems. Accordingly, each of the plurality of reproduction systems reproduces the sound field data transmitted from the server using the third predetermined number of reference loudspeakers.

  According to the first aspect, the second predetermined number of loudspeakers is reduced to the third predetermined number and the first predetermined number of microphones is reduced to the fourth predetermined number, so that the load and data amount of the convolution process are reduced. can do. Therefore, convolution processing and data transmission can be performed in real time, and a sound field can be shared.

  A second invention is dependent on the first invention, and the sound field sharing system further includes an initial speaker changing unit, a first group storage unit, and a first group selection unit. These means are also realized by a computer (12, 18, 20, etc.). The initial speaker changing means sequentially changes the first reference loudspeaker selected by the initial speaker selecting means. Therefore, by repeatedly executing the first evaluation value calculating means and the reference speaker selecting means by the first execution means for each reference loudspeaker selected first, a plurality of sets of the third predetermined number of reference loudspeakers ( Total number of loudspeakers). The first set storage means stores a plurality of sets of the selected third predetermined number of reference loudspeakers for each case when the initial reference loudspeaker is sequentially changed by the initial speaker changing means. For example, the plurality of sets are stored in a computer memory (hard disk or RAM). The first set selection unit is a set of third predetermined groups in which the Gram Schmitt orthogonalization evaluation value calculated by the first evaluation value calculation unit satisfies a predetermined condition among the plurality of groups stored by the first group storage unit. Select a number of reference loudspeakers. Specifically, a set having the maximum average value of evaluation indices by the Gram-Schmidt orthogonalization method is selected. However, when the minimum value of the evaluation index for the group having the maximum average value of the evaluation index is extremely low, since there is a frequency with low linear independence, even if the average value of the evaluation index is maximum, It is not appropriate to choose. This is because it is considered that the sound field cannot be reproduced correctly. In such a case, the group with the next highest average value of the evaluation index is selected. However, if the minimum value of the evaluation index for the group having the next highest average value of the evaluation index is extremely low, the group having the next highest average value of the evaluation index is selected. The same applies thereafter. In this way, the third predetermined number of reference loudspeakers in the set considered optimal are selected. Therefore, each of the plurality of reproduction systems reproduces the sound field data transmitted from the server by using a set of the third predetermined number of reference loudspeakers selected by the first set selection unit.

  According to the second aspect, the loudspeaker considered to be optimal can be selected, so that the sound field can be correctly reproduced.

  A third invention is dependent on the second invention, and the sound field sharing system further includes an initial microphone changing means, a second set storage means, and a second set selection means. These means are also realized by a computer (12, 18, 20, etc.). The initial microphone changing means sequentially changes the first reference microphone selected by the initial microphone selecting means. The second set storage means stores a plurality of sets of the selected fourth predetermined number of reference microphones in each case when the initial reference microphone is sequentially changed by the initial microphone changing means. Then, the second set selecting unit selects a set of a fourth predetermined number of reference microphones, out of the plurality of sets stored by the second set storing unit, whose Gram Schmitt orthogonalization evaluation value satisfies a predetermined condition. Therefore, as in the case of the loudspeaker, a fourth predetermined number of microphones considered to be optimal are selected. Then, the server transmits the sound field data detected by the set of the fourth predetermined number of microphones selected by the second set selection means to the plurality of reproduction systems.

  According to the third aspect, since the optimum microphone is selected, the sound field can be correctly reproduced as in the second aspect.

  The fourth invention is dependent on the first to third inventions, and the fourth predetermined number is determined according to the third predetermined number. Specifically, the total number of elements of the matrix of the inverse system has been determined. Therefore, when the loudspeaker is determined to be the fourth predetermined number, the third predetermined number is obtained by dividing the total number of elements by the fourth predetermined number. It is determined.

  According to the fourth aspect, since the third predetermined number is determined according to the fourth predetermined number, the third predetermined number can be easily determined.

  The fifth invention is dependent on the first to fourth inventions, and the third predetermined number and the fourth predetermined number are determined according to at least the processing capability of the server and the reproduction system. In other words, the total number of elements in the inverse system matrix is determined by the processing capabilities of the server and the reproduction system.

  According to the fifth invention, since the third predetermined number and the fourth predetermined number are determined according to the processing capability of the server and the reproduction system, the convolution processing, transmission, and reproduction of the sound field data are surely performed in real time. Can be executed.

  A sixth invention is dependent on the first to fifth inventions, the second predetermined number is 62, and the third predetermined number is a value not exceeding 24. That is, a maximum of 24 loudspeakers are selected.

  According to the sixth aspect, since 62 loudspeakers can be reduced to 24, the convolution process and the data amount can be reduced.

  A seventh invention is dependent on the sixth invention, wherein the first predetermined number is 70 and the third predetermined number is a value not exceeding 8.

  According to the seventh invention, for example, when the number of elements of the inverse matrix is set to 192 and 24 loudspeakers are used, a maximum of 8 microphones can be selected.

  In an eighth aspect, a microphone array having a first predetermined number of microphones arranged in a certain sound field, sound field data detected by the microphone array is recorded, and the sound field data is transmitted to a plurality of reproduction systems. Optimization method for optimizing the number and arrangement of microphone arrays and speaker arrays in a sound field sharing system, including a server and a reproduction system that reproduces sound field data from the server by a speaker array having a second predetermined number of loudspeakers (A) selecting one loudspeaker of the speaker array as the first reference loudspeaker, and (b) selecting all of the speaker arrays other than the selected reference loudspeaker and the reference loudspeaker. Calculate the Gram Schmidt orthogonalization evaluation value with each of the target loudspeakers (C) selecting the evaluation target loudspeaker having the highest Gram Schmidt orthogonalization evaluation value calculated in step (b) as the reference loudspeaker; (d) selecting the reference loudspeaker as a result of the selection in step (c); Steps (b) and (c) are repeatedly executed until the number reaches a third predetermined number less than the second predetermined number, and (e) one microphone of the microphone array is selected as the first reference microphone (F) calculating a Gram-Schmidt orthogonalization evaluation value between the selected reference microphone and each of all the evaluation target microphones other than the reference microphone in the microphone array, and (g) by step (f) The evaluation target microphone with the highest calculated Gramschmitt orthogonalization evaluation value is the reference microphone. (H) Step (f) and step (g) are repeatedly executed until the number of reference microphones becomes a fourth predetermined number smaller than the first predetermined number as a result of selection in step (g). This is an optimization method.

  According to the eighth aspect of the invention, it is possible to provide a sound field sharing system capable of transmitting convolution processing and data in real time by reducing the number of loudspeakers and the number of microphones.

  According to the present invention, the second predetermined number of loudspeakers is reduced to the third predetermined number, and the first predetermined number of microphones is reduced to the fourth predetermined number, so that the load of convolution processing and the amount of data are reduced. Can do. Therefore, convolution processing and data transmission can be performed in real time, and a sound field can be shared.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

FIG. 1 is an illustrative view showing one example of a sound field sharing system of the present invention. FIG. 2 is an illustrative view showing an example of the microphone array shown in FIG. FIG. 3 is an illustrative view showing an example of the speaker array system shown in FIG. FIG. 4 is an illustrative view for explaining the principle of sound field reproduction. FIG. 5 is an illustrative view for explaining the Gramschmitt orthogonalization method. FIG. 6 is a graph showing changes in the average value and the minimum value of the evaluation index when 24 loudspeakers are selected for 62 microphones when each loudspeaker is first selected. FIG. 7 is a graph showing changes in the average value and the minimum value of the evaluation index according to the selection process when the 60th loudspeaker is first selected. FIG. 8 is a graph showing changes in the average value and the minimum value of the evaluation index when eight microphones are selected for 24 loudspeakers when each microphone is first selected. FIG. 9 is a graph showing changes in the value of the evaluation index when the 65th microphone is first selected and the number of selection processes is increased. FIG. 10 is an illustrative view showing a distribution of 24 selected speaker positions when the 60th loudspeaker is first selected. FIG. 11 is an illustrative view showing a distribution of positions of eight selected microphones when the 65th microphone is first selected. FIG. 12 is an illustrative view showing a table of experimental conditions. FIG. 13 is a graph showing the RMS value for the difference in angle recognized by the subject for the presented sound source localization and the accuracy rate of the angle recognized by the subject for the presented sound source localization. FIG. 14 is a table showing the results of Tukey's multiple comparison test for the difference RMS value and the table showing the results of Tukey's multiple comparison test for the correct answer rate.

  Referring to FIG. 1, the sound field sharing system 10 of this embodiment includes a server 12, and a microphone array 14 is connected to the server 12. The server 12 is a general-purpose server, and is connected to the computer 18 and the computer 20 via a network 16 such as the Internet and / or a LAN. The computers 18 and 20 are general-purpose PCs or workstations. A speaker array system 22 and a microphone 24 are connected to the computer 18. Similarly, a speaker array system 26 and a microphone 28 are connected to the computer 20.

  The sound field sharing system 10 shown in FIG. 1 includes two sound field control (BoSC) reproduction systems 10a and 10b. The BoSC playback system 10 a is configured by a server 12, a microphone array 14, a network 16, a computer 18, a speaker array system 22 and a microphone 24 so as to be surrounded by a dotted line frame in FIG. 1. In addition, the BoSC playback system 10b includes a server 12, a microphone array 14, a network 16, a computer 20, a speaker array system 26, and a microphone 28 so as to be surrounded by a one-dot chain line frame in FIG.

As shown in FIG. 2, the microphone array 14 includes a skeleton 14a having a nearly spherical shape and a stand 14b that supports the skeleton 14a. Skeleton _14a, based on the structure of the _{C 80} fullerene (Fullerene), has 70 vertices taken ten vertices of the bottom. Although not shown, it is the surface (outer surface) of the skeleton 14a, and one omnidirectional microphone is attached to each of the 70 apexes. For example, DPA 4060-BM can be used as the microphone. The stand 14b is constituted by a support shaft 140 and a tripod 142, and the support shaft 140 supports the ceiling of the skeleton 14a from the inside through the bottom portion cut out of the skeleton 14a.

  The skeleton 14a can be seen from the front even if it is on the back side except for the part that overlaps the front side, but for the sake of easy understanding, the part corresponding to the back side is shown by dotted lines in FIG.

  As shown in FIG. 3, the speaker array system 22, 26 includes an elliptical dome portion 220 and four pillar portions 222 that support the dome portion 220. The elliptical dome portion 220 is constituted by, for example, wooden four-layer mounts 220a, 220b, 220c, and 220d. However, in FIG. 3, the inside of the dome portion 220 is viewed from an obliquely lower side, and a part of the gantry 220 d and the column portion 222 is shown. Although illustration is omitted, the inside of the dome part 220 and the pillar part 222 is made hollow, and the gantry (220a-220d) itself serves as a closed-chamber enclosure.

  In addition, 70 loudspeakers 230 are installed in each of the speaker array systems 22 and 26. Specifically, six full-range units (Fostex FE83E), that is, loudspeakers 230 are installed on the gantry 220a, 16 loudspeakers 230 are installed on the gantry 220b, and 24 loudspeakers 230 are installed on the gantry 220c. And 16 loudspeakers 230 are installed on the frame 220d. Further, in each of the four pillars 222, two subwoofer units (Fostex FW108N), that is, loudspeakers 230 are installed to compensate for the low frequency range.

  Such speaker array systems 22 and 26 are each installed in a sound field reproduction room (not shown). The sound field reproduction room is a 1.5-cm soundproof room, and a YAMAHA woody box (sound insulation performance Dr-30) is used. In addition, a chair with a lift (not shown) is provided in the sound field reproduction room. This is in the dome portion 220 of the speaker array system 22, 26, in order to set the position (height) of the ear of the user sitting on the chair to the height of the mount 220 c where the number of loudspeakers 230 is maximum. It is.

  For the sound field reproduction room (sound field reproduction system) including the microphone array 14 and the computers (18, 20) and the speaker array system (26, 28), refer to “1. Numerical analysis technology and visualization / audibility 1. .7 Three-Dimensional Sound Field Communication System ”Seigo Enomoto Acoustic Technology No.148 / Dec.2009 pp37-42, and will not be described in further detail.

  For example, in the sound field sharing system 10 shown in FIG. 1, the microphone array 14 is arranged in a sound field such as an orchestra performance hall. The server 12 converts a sound signal (sound field signal) input from the microphone array 14 via an amplifier (not shown) into digital sound data (sound field data), and performs an inverse system on the sound field data. The convolution process is executed. The server 12 transmits the sound field data subjected to the convolution process to the computers 18 and 20 via the network 16.

  The computers 18 and 20 convert the sound field data from the server 12 into analog sound field signals and output them to the speaker array systems 22 and 26, respectively. Therefore, in the speaker array systems 22 and 26, the above-described sound field is reproduced. For this reason, each user (not shown) who uses the speaker array systems 22 and 26 is recorded at the performance hall, for example, via the speaker array systems 22 and 26 even when they are located in a remote place. You can enjoy a raw orchestra.

  Each user can input voice through the microphones 24 and 28. The audio signal detected by the microphone 24 is converted into digital audio data by the computer 18 and transmitted to the computer 20 via the network 16. The computer 20 performs a convolution operation on the received audio data and the audio filter, superimposes the audio data on the audio field data, and outputs the result to the speaker array system 26. Therefore, the sound field is reproduced and the voices of other users are reproduced. Similarly, the audio signal detected by the microphone 28 is converted into digital audio data by the computer 20 and transmitted to the computer 18 via the network 16. The computer 18 performs a convolution operation on the received sound data and the sound filter, superimposes the sound data on the sound field data, and outputs it to the speaker array system 24.

  Therefore, the user of the speaker array system 22 and the user of the speaker array system 26 can share a sound field and have a conversation.

  In addition, although detailed description is abbreviate | omitted, the microphone 24,28 can use a headset microphone, for example.

  Here, the principle of boundary sound field control (BoSC) and a sound field reproduction system using BoSC will be briefly described. In the boundary sound field control, based on the Kirchhoff-Helmholtz integral equation (KHIE), the sound field in the region V in the original sound field shown on the left side of FIG. 4 is reproduced in the region V ′ in the current sound field shown on the right side of FIG. Is done. However, the relative positions of the recording point r on the boundary S surrounding the region V and the control point r ′ on the boundary S ′ surrounding the region V ′ are equal. That is, it is assumed that Equation 1 holds. However, the point s and the point s ′ are arbitrary points inside each region.

[Equation 1]
| R−s | = | r′−s ′ |, s∈V, s′∈V ′
At this time, the sound pressures p (s) and p (s ′) in the region that does not include the sound source are expressed by Equations 2 and 3 from KHIE.

Where ω is the angular frequency, ρ ₀ is the density of the medium, p (r) and v _n (r) are the sound pressure at the point r on the boundary and the particle velocity in the direction of the normal n, respectively. G (r | s) is a free space Green's function.

  Here, from Equation 1, the relationship shown in Equation 4 is established. Further, according to Equation 4, Equation 5 is established.

  From Equation 5, if a signal is output from the secondary sound source so that the sound pressure on the boundary surface S collected by the original sound source and the particle velocity are equal in the reproduced sound field, the sound field in the region V will be the region. It can be seen that it is reproduced at V ′.

  However, the output of the secondary sound source is determined by convolving the inverse filter that cancels the transfer characteristics from all secondary sound sources to all control points and the signal observed at the recording point. Therefore, in order to realize a BoSC sound field reproduction system as shown in FIG. 4, it is important to design a stable and robust inverse filter (pinv (H)).

  The inverse filter design method is described in the literature (S. Enomoto et al., “Three-dimensional sound field reproduction and recording systems based on boundary surface control principle”, Proc. Of 14th ICAD, Presentation o 16, 2008 Jun.). Will be described briefly here.

  A method for designing a multichannel-multipoint control inverse system (hereinafter simply referred to as “inverse system”) having the number M of secondary sound sources and the number N of control points as shown in FIG. 4 will be briefly described. However, the inverse system is a general term for M × N inverse filter groups.

  When the transfer function from the secondary sound source i to the control point j is Hji (ω), the input signal is Xj (ω), and the observation signal is Pj (ω), these relations are expressed by Equation 6. Can do. Where i is the secondary sound source number (1, 2,..., M), j is the control point number (1, 2,..., N), and W (ω) is the inverse system.

[Equation 6]

  At this time, in order to satisfy P (ω) = X (ω), Equation 7 must be satisfied. However, + means a pseudo inverse matrix. Thus, [W (ω)] is defined as the inverse system of [H (ω)].

[Equation 7]
[W (ω)] = [H (ω)] ⁺
Here, it is well known that the regularization method is a rational method for solving the inverse problem. This has already been applied to sound reproduction systems (TOKUNO et al., “Inverse Filter of Sound Reproduction Systems Using Regularization” EIEIC TRANS. FUNDAMENTALS, Vol.E80-A, NO.5 MAY 1997, etc.). By using the regularization method, the calculated inverse matrix [W ^ (ω)] for rank ([H (ω)]) = N (“^” is shown next to W for convenience of description. Is actually written on W as shown in Equation 8. The same applies hereinafter.) Is given by Equation 8. In Equation 8, # means conjugate transpose, β (ω) is a regularization parameter, and _IM is an M × M unit matrix. The same applies hereinafter.

[Equation 8]

On the other hand, the inverse matrix [H (ω)] ⁺ for rank ([H (ω)]) = M shown on the right side of Equation 7 is derived as Equation 9.

[Equation 9]

Equations 8 and 9 are interpreted as a least square solution and a minimum norm solution (norm minimum general inverse matrix), respectively. However, rank ([H (ω)]) = N = M, [H (ω)] is not a singular matrix (non-regular matrix), and [W (ω)] = [H (ω)] ⁻ Given by ¹ . Moreover, -1 means an inverse matrix. Finally, the time domain inverse filter coefficients are obtained from the inverse discrete Fourier transform of [W ^ (ω)].

  In the BoSC playback system, the arrangement of the loudspeaker 230 and the microphone affects spatial sampling.

  In the equations (8) and (9), the instability of the inverse system can be reduced (removed) by selecting an appropriate regularization parameter β (ω). In this embodiment, the regularization parameter β (ω) is heuristically defined in the frequency band of each object. Furthermore, the inverse filter was calculated by using the impulse response measured in advance between each loudspeaker 230 and microphone pair in a soundproof room. Because the measured impulse response was used, it did not follow the fluctuations caused by environmental changes. However, in an actual environment that fluctuates, an adaptive inverse filter of MIMO (Multiple-Input Multiple-Output) can be applied to the BoSC reproduction system.

  Here, when the microphone array 14 and the speaker array systems 22 and 26 shown in FIGS. 1 to 3 are used as they are, the processing load on the server 12 is considerably large. Specifically, since the microphone array 14 is 70 ch and the speaker array system 22 is 62 ch, the server 12 performs the convolution process of the audio signal (sound field data) of the microphone 70 ch and the inverse system 62 × 70 times. It is necessary to perform the convolution process for each round, and it is necessary to execute the number of taps of the inverse system (inverse filter) (4096 in this embodiment).

  Further, since the amount of sound field data to be transmitted (data amount) is enormous, each client (computer 18, 20) requires a bandwidth of about 45 Mbps.

  Furthermore, even when the computer 18 or 20 performs convolution calculation of the audio data corresponding to the user's voice and the audio filter, the processing load becomes relatively large when 70 ch is fully used.

  Therefore, it is difficult to transmit the sound field data from the server 12 to the computers 18 and 20 in real time, and naturally, it is difficult for the user using the speaker array systems 22 and 26 to enjoy the orchestra and the like in real time. is there. That is, the sound field cannot be shared in real time.

  In order to avoid this, for example, by reducing the number of microphones of the microphone array 14 and the number of loudspeakers 230 of the speaker array systems 22 and 26, it is conceivable to reduce the processing load of the convolution process and the amount of data to be transmitted. . However, it is not just that the number of microphones and loudspeakers 230 is reduced, and it is necessary not to impair the realism of the reproduced sound field.

  In this embodiment, therefore, the number of microphones and loudspeakers 230 is reduced and the appropriate number of microphones and loudspeakers 230 is determined without impairing the sense of reality.

  In this embodiment, first, the loudspeaker 230 used in the speaker array system 22 is extracted (selected) when the 70ch microphone array 14 is used, using the Gramschmitt orthogonalization method. When the selected loudspeaker 230 is used, microphones used in the microphone array 14 are extracted (selected) using the Gramschmitt orthogonalization method.

  Although detailed description is omitted, extraction (selection) of the loudspeaker 230 and the microphone to be used can be performed using the server 12, the computers 18, 20 or another computer (not shown).

  Here, a basic algorithm in the case of selecting the loudspeaker 230 by using the Gram Schmidt orthogonalization method for a single frequency will be described. If the linear independence from the N-dimensional vertical vector contained in N × M is low, the determinant is said to be in a bad state. The degradation of linear independence in [H (ω)] causes instability of the BoSC playback system. Here, [H (ω)] shown in Equation 6 can be written as in Equation 10.

[Equation 10]
P (ω) = [H (ω)] Y (ω)
= {h ₁ (ω),…, h _M (ω)} Y (ω)
However, Y (ω) = [W (ω)] X (ω) and h _i (ω) are N-dimensional vertical vectors included in [H (ω)]. This vertical vector h (ω) is a transfer function between a certain loudspeaker 230 and each microphone at a frequency ω. Therefore, the selection of the loudspeaker 230 using the Gramschmitt orthogonalization means selecting a set of longitudinal vectors h (ω) having high linear independence from [H (ω)]. Hereinafter, the algorithm of the Gramschmitt orthogonalization method will be briefly described.

In the n-th step of selecting the loudspeakers 230, n-1 loudspeakers 230 have already been selected. A set of vertical vectors included in [H] is represented by τ = {h ₁ ,..., H _M }. S _n−1 indicates a subset of vectors selected up to the (n−1) th step, and τ _n−1 indicates a subset of unused vectors until the (n−1) th step. v _n−1 = {v ₁ ,..., v _n−1 } represents an orthonormal basis of a plane stretched by the subset S _n−1 .

For example, in the first step, one of the loudspeakers 230 is selected as the reference loudspeaker 230, and all the loudspeakers 230 other than the reference loudspeaker 230 are evaluated. Selected as loudspeaker 230). As will be described later, one evaluation target loudspeaker 230 is selected from the plurality of evaluation target loudspeakers 230 in relation to the reference loudspeaker 230 by the Gram Schmidt orthogonalization method. In the next step, the remaining plurality of evaluation target loudspeakers 230 in relation to the reference loudspeaker 230 initially selected and the evaluation target loudspeaker 230 selected in the previous step, also using the Gram Schmidt orthogonalization method. To 1 of the evaluation target loudspeakers 230 is selected. That is, in this step, the evaluation target loudspeaker 230 selected in the previous step can be said to be the reference loudspeaker 230. This is repeated.
However, the eight loudspeakers 230 that compensate for the low frequency band are outside the scope of the reference loudspeaker 230 and the evaluation target loudspeaker 230.

FIG. 5 is an example of a plane spanned by the subset S _n−1 . In the n-th step, h _n ^ for the plane stretched by the subset S _n−1 (in fact, “^” is written on h as shown in Equation 11. The same applies hereinafter). H _n ^ is selected so that the vertical component is maximized. A vertical component r _i of an arbitrary vector h _i included in the subset τ _n−1 is expressed by Equation 11.

[Equation 11]
r _i = z _i -p
Here, p represents a projection (projection) on a plane stretched by the subset S _n−1 . The n-th loudspeaker 230 is determined so that the norm of the vertical component r _i shown in, for example, Equation 12 is maximized.

[Equation 12]

However, J (h _i ), which is the value of the evaluation index, is defined by Equation 13.

[Equation 13]
J (h _i ) = || r _i ||
If the vertical component of h _i ^ is indicated as r _n ^ (actually, the symbol “^” is written on r. The same applies hereinafter), the nth orthonormal vector v _n is a number. 14 is determined.

[Formula 14]

The evaluation index value J _n ^ maximized in the n-th step (actually, the symbol “^” is written on J. The same applies hereinafter) is expressed by Equation 15.

[Equation 15]

Such processing according to Equation 11 to Equation 15 is repeated until the evaluation index value J _n ^ becomes smaller than a preset threshold value J _thr ^. However, for the frequency band [ω _l , ω _h ], two evaluation index values are obtained according to Equation 16.

[Equation 16]

However, h _i ￣ = {h _i (ω _l ),..., H _i (ω _h )} (in practice, “￣” is written on h as shown in Equation 16). , K is the number of discrete frequencies ω _k and a _k is an arbitrary weighting factor for the discrete frequency ω _k . The vertical component r _i (ω _k ) and the orthonormal vector v _i (ω _k ) are obtained separately for each discrete frequency as in the case of a single frequency. In the optimization process, the evaluation index value J _avg is maximized. On the other hand, the evaluation index value J _min is used to determine the end of the optimization process. That is, selection of the loudspeaker 230 ends when J _min ^ <J _thr ^.

  However, optimization processing is disclosed in the literature (Asano, Suzuki, and Swanson "Optimization of control source configuration in active control systems using Gram-Schmidt orthogonalization", Speech and Audio Processing, IEEE Transactions on, Mar. 1999). Yes.

In this document, when the value of the evaluation index is equal to or greater than the threshold (J _min ^ ≧ J _thr ^), the selection of the loudspeaker 230 is continued. However, a method for determining an appropriate threshold has not been confirmed. Therefore, in this embodiment, the maximum number of loudspeakers 230 and the maximum number of microphones that can share a sound field in real time in the sound field sharing system 10 are verified. And the number (arrangement position) of the loudspeakers 230 up to the maximum number was determined by using the Gramschmitt orthogonalization method.

  Here, as described above, in the Gram Schmidt orthogonalization method, the speaker position is determined based on the speaker position previously selected, and therefore, the selection result is the first selected speaker position. Has a strong influence.

For example, the case where the number of the loudspeakers 230 to be used is reduced to about half (32), about one third (24), or about one fourth (16) was examined. FIG. 6 shows changes in the evaluation index values J _avg and J _min when 24 loudspeakers 230 are selected (a selection process of 24 steps is executed). In FIG. 6, the horizontal axis indicates the speaker position (see FIG. 10) of the first selected loudspeaker 230 (reference loudspeaker 230), and the vertical axis indicates the evaluation value (dB). However, of the two solid lines, the thin solid line indicates the value J _avg of the evaluation index, and the thin solid line indicates the change in the value J _min of the evaluation index.

Although the detailed description is omitted, for example, the reference loudspeaker 230 selected first is sequentially changed (2, 3,..., 62) from “1” (see FIG. 10), and is selected for each case. In addition, a set of 24 speaker positions (numbers of loudspeakers 230) is selected, and evaluation index values J _avg and J _min are calculated for each set. However, the set of 24 selected speaker positions (numbers of the loudspeakers 230) and the evaluation index values J _avg and J _min calculated for each set are the memory of the computer (not shown). Stored in a hard disk or RAM). As will be described later, one set of evaluation index values J _avg and J _min satisfying a predetermined condition is selected from the plurality of sets. Therefore, the sound field is reproduced using the selected set of 24 loudspeakers 230.

  The free space Green function was also used to obtain the transfer function between each loudspeaker 230 and the microphone. The upper limit frequency for stimulation described below was not limited here. However, the configuration (setting) of the loudspeaker 230 was determined in the range from 20 Hz to 1 kHz at a frequency of 20 Hz. Although illustration is omitted, when the upper limit frequency is not limited, many loudspeakers 230 arranged on the upper layer (the gantry 220a and the gantry 220b) are selected. It is difficult to integrate a wavefront coming from a direction where there is no loudspeaker 230 in a three-dimensional sound reproduction system. Accordingly, the loudspeaker 230 should be positioned in every possible direction surrounded by the microphone array.

As described above, FIG. 6 is a graph showing the evaluation index values J _avg and J _min in a broken line when the selection process of 24 steps (times) is performed for the loudspeaker 230. As can be seen from FIG. 6, the evaluation index value J _avg when the loudspeaker 230 whose speaker position is “60” (see FIG. 10) is first selected and all 24 loudspeakers 230 are selected. , J _min is the maximum.

In this embodiment, among a plurality of sets (62 sets in this embodiment), a set of 24 loudspeakers 230 in which evaluation index values J _avg and J _min satisfy a predetermined condition are selected. . Specifically, the pair having the maximum evaluation index value J _avg is selected. However, when the evaluation index value J _min for the pair having the maximum evaluation index value J _avg is extremely low, a frequency with low linear independence exists, and therefore the evaluation index value J _avg is the maximum. However, it is not appropriate to choose. This is because it is considered that the sound field cannot be reproduced correctly. In such a case, a group having the next largest evaluation index value J _avg is selected. However, if the next value J _min of metrics for the value J _avg large set of metrics is extremely low, the set value J _avg metric the next larger is selected. The same applies thereafter. For example, the computer determines whether or not the value J _{min of the} evaluation index is extremely low based on a preset threshold value. This threshold is a value set by the developer or user of the sound field sharing system 10. However, as shown in FIG. 7 to be described later, as the number of loudspeakers 230 to be selected increases, the evaluation index values J _avg and J _min gradually decrease, so that the threshold value also depends on the number of loudspeakers 230 to be selected. Must be set variably.

FIG. 7 is a graph showing changes in the evaluation index values J _avg and J _min when the loudspeaker 230 whose speaker position is “60” is first selected and then the selection process is repeated. As can be seen from FIG. 7, the evaluation index values J _avg and J _min gradually decrease.

Although not shown for simplicity, as described above, even when the number of loudspeakers 230 is reduced to 16 or 32, changes in evaluation index values J _avg and J _min as shown in FIG. showed that. However, as described later, the maximum number of loudspeakers 230 was determined to be 24 based on the result of the sound source localization test.

  As a result of the preliminary test, the number of elements in [W (ω)] is within M × N = 192 and the number of loudspeakers 230 due to the performance of the server 12 and the computers 18 and 20 and the communication speed limitation including the network 16. It was shown that (M) and the number of microphones (N) should be determined. However, in this embodiment, the CPU (not shown) of the server 12 and the computers 18 and 20 is Xeon (registered trademark) QuadCore × 2, and the memory (not shown) is 4 GB. The server 12 employs Windows (registered trademark) XP 64 bits as an operating system. Further, as the network 16 connecting the server 12 and the computers 18 and 20, an ultrahigh-speed, high-function R & D test bed network (JGN2 plus: 1 Gbps) and a LAN (100 Mbps) were used.

  In addition, although illustration is abbreviate | omitted, in a preliminary experiment, the server 12 and the computer 18 are connected using the above-mentioned LAN, and the server 12 and the computer 20 are connected using the above-mentioned JGN2plus and LAN.

Therefore, as described above, since the number (M) of the loudspeakers 230 is determined to be “24”, the number (N) of the selected microphones is “8” at the maximum. FIG. 8 shows changes in the evaluation index values J _avg and J _min when an 8-step selection process is executed for the microphone. As can be seen from FIG. 8, when a microphone having a microphone position of “65” (reference microphone) is first selected, the evaluation index values J _avg , J when a total of eight microphones are selected. _min is the maximum.

FIG. 9 shows changes in the evaluation values J _n and J _min when the microphone whose microphone position is “65” is first selected and then the selection process is repeated. As shown in FIG. 9, the evaluation index values J _n and J _min gradually decrease when the selection process is repeated, and the evaluation index is obtained when the number of repetitions is “25”, that is, when 25 microphones are selected. The values J _n and J _min are significantly reduced. Therefore, it is considered desirable to determine the maximum number of microphones within 24. As described above, since eight microphones are selected here, it can be said that this requirement is satisfied.

  In FIG. 10, as described above, the distribution of the speaker positions of the 24 loudspeakers 230 when the loudspeaker 230 whose speaker position is “60” is selected first and 24 loudspeakers 230 are selected in total. Is shown. However, although omitted in FIG. 10, the value in the height direction (Z direction) increases as the speaker position goes toward the center. Accordingly, the loudspeaker positions of the loudspeakers 230 provided on the mount 220a are “1”-“6”. The loudspeaker positions of the loudspeakers 230 provided on the gantry 220b are “7”-“22”. Furthermore, the loudspeaker position of the loudspeaker 230 provided on the gantry 220c is “23”-“46”. And the speaker position of the loudspeaker 230 provided in the mount 220d is “47”-“62”.

  In order to compensate for the low frequency, the eight loudspeakers 230 provided on the four pillars 222 are not selected and are not shown in FIG.

  In FIG. 10, a shaded pattern is added to a circle indicating the speaker position of the first selected loudspeaker 230 (a circle having “60” written therein). Further, following this, a circle indicating the speaker position of the loudspeaker 230 selected as a result of the repetition based on the Gramschmitt orthogonalization method (here, “1”-“6”, “7”, “9”). ”,“ 11 ”,“ 13 ”,“ 15 ”,“ 17 ”,“ 19 ”,“ 21 ”,“ 23 ”,“ 31 ”,“ 35 ”,“ 48 ”,“ 51 ”,“ 54 ”, A hatched pattern is attached to the circles “56”, “58”, and “62”. Further, a circle without a pattern indicates the speaker position of the loudspeaker 230 that has not been selected. From FIG. 10, the loudspeakers 230 distributed in each direction and height are regularly observed. When viewed in a plan view as shown in FIG. 10, it can be seen that the selected loudspeakers 230 are distributed substantially symmetrically in the top, bottom, left and right.

  In addition, the microphones were selected by applying the Gramschmitt orthogonalization method described above by switching the configuration of the loudspeaker 230 and the microphones. However, since the selection method using the Gramschmitt orthogonalization method has already been described, redundant description will be omitted.

  FIG. 11 shows an array of eight microphones relative to the array of 24 loudspeakers 230 shown in FIG. Although illustration is omitted, numbers are assigned to the positions of the microphones in the same manner as the speaker positions of the loudspeaker 230. Although it is a little difficult to understand in FIG. 11, when the XY plane is viewed from above, the selected microphones are evenly distributed in all directions.

  In this way, the number of microphones and loudspeakers 230 was reduced by using the Gramschmitt orthogonalization method, but in order to evaluate the effect of this reduction, a sound source localization test on a horizontal plane was performed. .

In the sound source localization test, the stimulus reproduced using the BoSC playback system was generated from convolution of pink noise and impulse response. However, the impulse response was simulated from a free space Green's function. In the simulation, it was assumed that the position of the sound source was located 1 meter away from the center of the microphone array 14. In addition, the inverse filter for sound field reproduction (stereoscopic sound reproduction) in the BoSC reproduction system is calculated by using an impulse response measured in advance at a sampling frequency of 48 kHz, and the length of the inverse filter is 4096 points. It was. The sound pressure level of the stimulus was adjusted to L _{A, Fmax} = 55 dB at the center of the microphone array 14 to eliminate the level difference between each experimental condition and direction.

  The experimental conditions regarding the number of loudspeakers 230 and the number of microphones are summarized in the table shown in FIG. All loudspeakers 230 and all microphones were used in condition 5. In condition 4, the number of loudspeakers 230 was reduced to 24 for all microphones. Conditions 1, 2 and 3 have the number of loudspeakers of 24, 32 and 16, as described above, and the number of microphones so that the number of elements (192) of the inverse matrix [W (ω)] matches. Were determined to be 8, 6 and 12, respectively.

  In this sound source localization test, 13 subjects (5 men and 8 women) from the 20s to 50s answered the perceived angles after listening to the regenerated stimuli. . Stimulation was presented in 30 degree increments between 0 and 330 degrees in the horizontal plane. They lasted 2 seconds and were repeated twice for each angle. However, the presentation order was determined by using the Latin square method. However, subjects were allowed to move their heads and bodies while listening to the stimulus.

  The results of the sound source localization test are shown in FIGS. 13 (A) and 13 (B). FIG. 13A shows the average of the RMS values (square mean values) between subjects for the difference between the angle of the reproduced sound and the perceived angle under each condition. However, in FIG. 13A, error bars indicate a 95% confidence interval (CI). As can be seen from FIG. 13A, in the case of condition 5, the lowest RMS value was obtained, and as a result, the highest accuracy of sound source localization was realized. In condition 4, as mentioned above, 24 loudspeakers 230 and 70 microphones were used, and the RMS value was about 5 degrees larger than in condition 5. The RMS value in condition 1 was about 15 degrees greater than in condition 5. Moreover, although the RMS value of condition 2 and condition 3 is almost the same, those values were about 25 degree | times larger than the case of condition 5.

  FIG. 13B shows a confidence interval (CI) in which the accuracy rate and error bar of the subject are 95% under each condition. As can be seen from FIG. 13 (B), under condition 5, the highest accuracy rate was obtained as in the result shown in FIG. 13 (A). The correct answer rate under condition 5 is about 5% higher than under condition 4 and about 10% higher than under condition 1.

  Although detailed explanation is omitted, the Hartley test has confirmed that all conditions have the same change, and in order to find a system (condition) that is significantly different from other systems (conditions), multiple comparisons of Tukey The law applies. The results of this statistical test are shown in the tables shown in FIGS. 14 (A) and (B). However, in the tables shown in FIGS. 14A and 14B, “*” indicates a significant difference of 1% between the conditions, and “**” indicates a significant difference of 5% between the conditions. Is shown. In the Tukey multiple comparison method, it was confirmed that there was no significant difference between Condition 5 and Condition 4 and between Condition 5 and Condition 1. Therefore, it can be said that the number of loudspeakers 230 and microphones constituting the BoSC playback system can be reduced by using the Gramschmitt orthogonalization method. In contrast, it was confirmed that there were significant differences between Condition 5 and Condition 2 and between Condition 5 and Condition 3. Therefore, when Condition 2 and Condition 3 are used, it can be said that reproduction by the BoSC reproduction system impairs the sense of reality.

  As described above, in this embodiment, the method of reducing the number of the loudspeakers 230 and the microphones using the Gram Schmidt orthogonalization method is shown. Using the Gramschmitt orthogonalization method, a group of longitudinal vectors with high linear independence was selected from the transfer function matrix between each loudspeaker 230 and microphone. The selected vector corresponds to the configuration of the loudspeaker 230 and the microphone in the BoSC playback systems 10a and 10b. Thus, it is believed that such a system can overcome changes in the acoustic environment as compared to a system comprised of a loudspeaker 230 and a microphone selected using other evaluation criteria.

  Further, in the selection procedure, the limitation of the frequency band from 20 Hz to 1 kHz satisfied the configuration in which the loudspeakers 230 are regularly dispersed in the simulation. Similarly, the selection of microphones using Gramschmitt's orthogonalization method resulted in a reduced number of microphones that were regularly distributed in all horizontal directions. Thus, the microphones were selected in the same way as the loudspeakers 230, but the number of loudspeakers 230 was already reduced by the Gram Schmidt orthogonalization method.

  Further, a horizontal plane sound source localization test was performed in order to evaluate deterioration due to the reduction in the number of loudspeakers 230 and microphones. According to the results of the subjective evaluation, there was no statistically significant difference between the BoSC playback system consisting of 62 loudspeakers 230 and the BoSC playback system consisting of 24 loudspeakers 230. Furthermore, for 24 loudspeakers 230, there was no statistically significant difference between a system using 8 microphones and a system using 70 microphones.

  Therefore, it is considered that the configuration of 24 loudspeakers 230 may be applied. Further, in this embodiment, the number of elements (192) of the inverse matrix [W (ω)] is determined from the performance of the server 12 and the computers 18 and 20 and the restrictions of the network 16. On the other hand, it was decided to apply the configuration of 8 microphones.

  Although a detailed description is omitted, the sound field signal detected by the selected microphone is supplied from the microphone array 14 to the server 12. At this time, unselected microphones are disabled. That is, the server 12 does not detect a sound field signal from a microphone that is not selected. On the other hand, the computers 18 and 20 output sound field data and audio data only to the selected loudspeaker 230.

  According to this embodiment, the number of microphones for recording the sound of the primary sound source and the number of microphones for recording the sound of the primary sound source are reduced in accordance with the Gram Schmidt orthogonal evaluation method. The load can be reduced and the amount of data to be transmitted can be reduced. Therefore, sound field data corresponding to the sound recorded in the sound field can be transmitted in real time and reproduced on the client side. That is, the sound field can be shared in real time by a person who exists in the sound field and a user who uses the speaker system.

  In this embodiment, two client computers are shown, but three or more client computers may be connected to the network. In such a case, each client computer individually folds audio data from two or more other client computers and superimposes the audio data on the sound field data.

DESCRIPTION OF SYMBOLS 10 ... Sound field sharing system 12 ... Server 14 ... Microphone array 18, 20 ... Computer 22, 26 ... Speaker array system

Claims (8)

A microphone array disposed in a certain sound field and having a first predetermined number of microphones;
A sound field data detected by the microphone array is recorded, the sound field data is transmitted to a plurality of reproduction systems, and the sound field data from the server is reproduced by a speaker array having a second predetermined number of loudspeakers. A sound field sharing system comprising the reproduction system
Initial speaker selection means for selecting one loudspeaker of the speaker array as an initial reference loudspeaker;
A first evaluation value calculation means for calculating a Gram Schmidt orthogonalization evaluation value between the selected reference loudspeaker and each of the evaluation target loudspeakers other than the reference loudspeaker in the speaker array;
Reference speaker selection means for selecting, as the reference loudspeaker, the evaluation target loudspeaker having the highest Gram Schmidt orthogonalization evaluation value calculated by the first evaluation value calculation means;
As a result of the selection by the reference speaker selection means, the first evaluation value calculation means and the reference speaker selection means are repeatedly executed until the number of the reference loudspeakers becomes a third predetermined number smaller than the second predetermined number. First execution means,
Initial microphone selection means for selecting one microphone of the microphone array as a first reference microphone;
A second evaluation value calculating means for calculating a Gram Schmitt orthogonalization evaluation value between the selected reference microphone and each of all the evaluation target microphones other than the reference microphone in the microphone array;
Reference microphone selection means for selecting the evaluation target microphone having the highest Gram Schmidt orthogonalization evaluation value calculated by the second evaluation value calculation means as the reference microphone, and the result of selection by the reference microphone selection means, the reference microphone Second execution means for repeatedly executing the second evaluation value calculation means and the reference microphone selection means until the number of the first evaluation number becomes a fourth predetermined number smaller than the first predetermined number,
The server transmits sound field data detected by the fourth predetermined number of the reference microphones to the plurality of reproduction systems;
Each of the plurality of reproduction systems reproduces the sound field data transmitted from the server using the third predetermined number of reference loudspeakers. Initial speaker changing means for sequentially changing the first reference loudspeaker selected by the initial speaker selecting means;
First set storage means for storing a plurality of selected sets of the third predetermined number of reference loudspeakers for each case when the initial reference loudspeaker is sequentially changed by the initial speaker changing means; And among the plurality of sets stored by the first set storage means, the third predetermined number of the set of the third predetermined number of sets in which the Gram Schmitt orthogonalization evaluation value calculated by the first evaluation value calculation means satisfies a predetermined condition Further comprising first set selection means for selecting a reference loudspeaker;
Each of the plurality of reproduction systems reproduces the sound field data transmitted from the server using a set of the third predetermined number of the reference loudspeakers selected by the first set selection unit. The sound field sharing system according to claim 1. Initial microphone changing means for sequentially changing the first reference microphone selected by the initial microphone selecting means;
Second set storage means for storing a plurality of selected sets of the fourth predetermined number of the reference microphones for each case when the initial reference microphones are sequentially changed by the initial microphone changing means; and Second set selection means for selecting one set of the fourth predetermined number of the reference microphones among the plurality of sets stored by the second set storage means, wherein the Gramschmitt orthogonalization evaluation value satisfies a predetermined condition. Prepared,
The sound field sharing system according to claim 2, wherein the server transmits sound field data detected by the set of the fourth predetermined number of the microphones selected by the second set selection unit to the plurality of reproduction systems. .   4. The sound field sharing system according to claim 1, wherein the fourth predetermined number is determined according to the third predetermined number. 5.   5. The sound field sharing system according to claim 1, wherein the third predetermined number and the fourth predetermined number are determined according to at least the processing capability of the server and the reproduction system.   The sound field sharing system according to any one of claims 1 to 5, wherein the second predetermined number is 62 and the third predetermined number is a value not exceeding 24.   The sound field sharing system according to claim 6, wherein the first predetermined number is 70 and the fourth predetermined number is a value not exceeding eight. A microphone array disposed in a certain sound field and having a first predetermined number of microphones;
A sound field data detected by the microphone array is recorded, the sound field data is transmitted to a plurality of reproduction systems, and the sound field data from the server is reproduced by a speaker array having a second predetermined number of loudspeakers. An optimization method for optimizing the number and arrangement of the microphone array and the speaker array of a sound field sharing system comprising the reproduction system comprising:
(A) selecting one loudspeaker from the speaker array as a first reference loudspeaker;
(B) calculating a Gramschmitt orthogonalization evaluation value between the selected reference loudspeaker and each of the evaluation target loudspeakers other than the reference loudspeaker in the speaker array;
(C) selecting the evaluation target loudspeaker having the highest Gram Schmidt orthogonalization evaluation value calculated in step (b) as the reference loudspeaker;
(D) Steps (b) and (c) are repeatedly executed until the number of the reference loudspeakers becomes a third predetermined number smaller than the second predetermined number as a result of the selection in step (c). Let
(E) selecting one microphone of the microphone array as the first reference microphone;
(F) calculating a Gram Schmitt orthogonalization evaluation value between the selected reference microphone and each of all evaluation target microphones other than the reference microphone in the microphone array;
(G) selecting the evaluation target microphone having the highest Gram Schmitt orthogonalization evaluation value calculated in step (f) as the reference microphone; and (h) selecting the reference microphone as a result of the selection in step (g). The step (f) and the step (g) are repeatedly executed until the number reaches a fourth predetermined number smaller than the first predetermined number.

Images