JP2017026795A - Transmission system, encoding device, decoding device, and method and program therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a transmission system that efficiently transmits sound space information.SOLUTION: An encoding device 100 generates a matrix M having observation signals S-Sof respective microphones and values normalized by autocovariance of the observation signals (110), sorts the N eigenvalues λand eigenvectors eof a square matrix MM so that they are in descending order with respect to the eigenvalues λ(140), obtains a matrix E'=[e(1)e(2) to e(N')] consisting only of N' eigenvectors e(n'), corresponding to N' eigenvalues, respectively, on the basis of a relationship between a predetermined threshold and values corresponding to the eigenvalues after the sorting (150), calculates a variance of the observation signals and a matrix μ' that is a product of the matrix M and matrix E' at a dimension reduction unit 160, and determines a difference between the observation signals S-Sand an approximate value as an error at an error calculation unit 170. A decoding device includes a decoding unit that decodes the observation signals using the matrix μ' and the matrix E', N variances σ, and the error.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a technique for transmitting sound collected using a plurality of microphones. It should be noted that, by arranging a plurality of microphones in a three-dimensional space and collecting sound, the observation signal includes information on the sound field. Good. The sound field information is, for example, an impulse response or the like, and may be referred to as characteristics including sound space information or sound three-dimensional space information.

  Non-Patent Documents 1 to 3 are known as techniques for transmitting sound collected using a plurality of microphones. In Non-Patent Documents 1 to 3, audio information is compressed and transmitted.

M. Hans, R. Schafer, “Lossless compression of digital audio”, Signal Processing Magazine, IEEE, 2001, vol. 18, pp. 21-32. J. Herre, “Form joing stereo to spatial audio coding recent progress and standardization”, In 6th Intl. Conf. On Digital Audio Effects, 2004, p. 6. J. Breebaart, et.al, “Backgroud, Concept, and Architecture for the Recent MPEG Surround Standard on Multichannel Audio Compression”, J. Audio Eng. Soc., 2007, vol. 55 (5), pp. 331-351 .

  However, the sound compression techniques of Non-Patent Documents 1 to 3 do not consider the correlation between microphones when a plurality of microphones are arranged adjacent to each other. That is, when a plurality of microphones are arranged adjacent to each other, a highly correlated signal is input to each microphone, but these are handled as independent signals in the conventional technology.

  An object of the present invention is to provide a transmission system, an encoding device, a decoding device, and a method and a program for efficiently transmitting sound space information in consideration of the correlation between a plurality of microphones.

In order to solve the above problem, according to one aspect of the present invention, a transmission system uses N as an integer equal to or greater than 2, and an observation signal matrix S collected by N microphones from an encoding device. Transmit to the decoding device. Encoding device, n = 1, 2, ..., and N, and each of the observed signal s _n of each microphone, the average s of each observed signal s _n ^- autocovariance of the observed signal s _n a difference between _n sigma A normalization unit that generates a matrix M having elements m _n normalized by _{s_ns_n} , and N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M are arranged in descending order with respect to the eigenvalue λ _n. N '<N, n' = 1,2, ..., N ', and the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N ), The eigenvector corresponding to λ (n) is e (n), and N ′ eigenvalues λ (n ′) are respectively determined based on the magnitude relationship between the predetermined threshold and the value corresponding to the eigenvalue λ (n ′). and the corresponding n 'eigenvectors e (n') consisting only of matrix E '= [e (1) e (2) ... e (n')] truncation unit to obtain, for each of the n observed signal s _n and variance _σ s_ns_n, a 'matrix μ is the product of' the matrix M matrix E, obtains a 'approximate value matrix S of the observation signal matrix S by using the' matrix E, the observed signal matrix S Including the difference between the Nitinol matrix S 'and the error calculating section for obtaining an error R, a. The decoding device includes a decoding unit that decodes the observed signal matrix S using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n,} and the error R.

In order to solve the above-described problem, according to another aspect of the present invention, the encoding device corresponds to an observation signal matrix S collected by N microphones, where N is any integer greater than or equal to 2. Generate a code. Encoding device, n = 1,2, ..., and N, and each of the observed signal s _n, times the average s of each observed signal s _n ^- in autocovariance sigma _{S_ns_n} of each observation signal the difference between _n s _n A normalization unit that generates a matrix M having elements of normalized values m _n and an array that rearranges the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M in descending order with respect to the eigenvalues λ _n. And N ′ <N, n ′ = 1, 2,..., N ′, and the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), The eigenvector corresponding to λ (n) is e (n), and a truncation unit that obtains a matrix E ′ composed of only eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold And an approximate value matrix S ′ of the observed signal matrix S using the variance σ _{s_ns_n} of each of the N observed signals s _n , the matrix μ ′ that is the product of the matrix M and the matrix E ′, and the matrix E ′. And an error calculation unit that calculates a difference between the observed signal matrix S and the approximate value matrix S ′ as an error R.

In order to solve the above-described problem, according to another aspect of the present invention, a decoding device uses a code corresponding to an observation signal matrix S collected by N microphones, where N is any integer of 2 or more. And the observed signal matrix S is decoded. Decoding apparatus, n = 1,2, ..., and N, and each of the observed signal s _n, times the average s of each observed signal s _n ^- normal differences between _n in autocovariance sigma _{S_ns_n} of each observed signal s _n M is a matrix whose elements are generalized values m _n , and is obtained by rearranging the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M in descending order with respect to the eigenvalues λ _n Is the eigenvalue of λ (n), the rearranged eigenvalue is λ (n), λ (1)> λ (2)>...> λ (N), and the eigenvector corresponding to λ (n) is e (n) N ′ <N, n ′ = 1, 2,..., N ′, and a matrix consisting only of eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold _Let E 'be the approximate value matrix of the observed signal matrix S obtained using the variance σ _{s_ns_n} of each of the N observed signals s _n and the matrix μ' = ME 'and the matrix E', and the observed signal matrix The difference between S and the approximate value matrix S ′ is the error R, and the observed signal row is calculated using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n} and the error R. A decoding unit for decoding the sequence S is included.

In order to solve the above-described problem, according to another aspect of the present invention, a transmission method uses N as an integer equal to or greater than 2, and encodes an observation signal matrix S collected by N microphones. To the decoding device. Transmission method, n = 1, 2, ..., and N, the coding apparatus, and the observed signal s _n of each microphone, the average s of each observed signal s _n ^- a difference between _n of each observed signal s _n A normalization step for generating a matrix M whose elements are values m _n normalized by the autocovariance σ _{s_ns_n} , and an encoding device _converts the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M into eigenvalues. A rearrangement step for rearranging λ _{n in} descending order, N ′ <N, n ′ = 1, 2,..., N ′, and eigenvalues after the rearrangement are λ (n), λ (1)> λ (2)>...> The eigenvectors corresponding to λ (N) and λ (n) are set to e (n), and the encoding device has a magnitude relationship between the predetermined threshold and the value corresponding to the eigenvalue λ (n ′). Based on the matrix E ′ = [e (1) e (2)... E (N ′)] consisting only of N ′ eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′), respectively. a truncation obtaining, encoding apparatus, the variance sigma _{S_ns_n} of n each observed signal s _n, which is the product of the the matrix M matrix E 'line An error calculation step of obtaining an approximate value matrix S ′ of the observation signal matrix S using μ ′ and the matrix E ′, and obtaining a difference between the observation signal matrix S and the approximate value matrix S ′ as an error R, and a decoding device , A decoding step of decoding the observed signal matrix S using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n,} and the error R.

In order to solve the above-described problem, according to another aspect of the present invention, an encoding method corresponds to an observation signal matrix S collected by N microphones, where N is any integer greater than or equal to 2. Generate a code. Encoding method, n = 1, 2, ..., and N, normalization section, and each of the observed signal s _n, times the average s of each observed signal s _n ^- a difference between _n own respective observation signal s _n A normalization step for generating a matrix M whose elements are values m _n normalized by the covariance σ _{s_ns_n} , and a reordering unit _converts the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M into eigenvalues λ _A rearrangement step for rearranging _{n in} descending order, N ′ <N, n ′ = 1, 2,..., N ′, and eigenvalues after the rearrangement are λ (n), λ (1)> λ ( 2)>...> The eigenvector corresponding to λ (N), λ (n) is e (n), and the eigenvector e corresponding to N ′ eigenvalues λ (n ′) whose truncation part is equal to or less than a predetermined threshold value A truncation step for obtaining a matrix E ′ consisting only of (n ′), and an error calculating unit, a variance σ _{s_ns_n} of each of the N observation signals s _n , and a matrix μ ′ that is a product of the matrix M and the matrix E ′ , Find approximate matrix S 'of observed signal matrix S using matrix E' and approximate observed signal matrix S Including, an error calculation step of obtaining an error R the difference between the matrix S '.

In order to solve the above-described problem, according to another aspect of the present invention, a decoding method uses a code corresponding to an observation signal matrix S collected by N microphones, where N is any integer greater than or equal to 2. And the observed signal matrix S is decoded. Decoding method, n = 1,2, ..., and N, and each of the observed signal s _n, times the average s of each observed signal s _n ^- normal differences between _n in autocovariance sigma _{S_ns_n} of each observed signal s _n M is a matrix whose elements are generalized values m _n , and is obtained by rearranging the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M in descending order with respect to the eigenvalues λ _n Is the eigenvalue of λ (n), the rearranged eigenvalue is λ (n), λ (1)> λ (2)>...> λ (N), and the eigenvector corresponding to λ (n) is e (n) N ′ <N, n ′ = 1, 2,..., N ′, and a matrix consisting only of eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold _Let E 'be the approximate value matrix of the observation signal matrix S obtained using the variance σ _{s_ns_n} of each of the N observation signals s _n , the matrix μ' = ME ', and the matrix E', and the observation signal The difference between the matrix S and the approximate value matrix S ′ is an error R, and the decoding unit uses the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n,} and the error R. A decoding step of decoding the observed signal matrix S.

  According to the present invention, there is an effect that sound space information can be efficiently transmitted.

The figure for demonstrating the approximation matrix of the diagonal matrix which uses an eigenvalue as a diagonal element. The functional block diagram of the transmission system which concerns on 1st embodiment. The functional block diagram of the encoding apparatus which concerns on 1st embodiment. The figure which shows the example of the processing flow of the encoding apparatus which concerns on 1st embodiment. The functional block diagram of the decoding apparatus which concerns on 1st embodiment. The figure which shows the example of the processing flow of the decoding apparatus which concerns on 1st embodiment.

  Hereinafter, embodiments of the present invention will be described. In the drawings used for the following description, constituent parts having the same function and steps for performing the same process are denoted by the same reference numerals, and redundant description is omitted. In the following description, it is assumed that processing performed for each element of a vector or matrix is applied to all elements of the vector or matrix unless otherwise specified.

<Points of first embodiment>
The sound space information compression method according to the first embodiment includes an encoding step and a decoding step. First, the encoding step will be described.

  Let S (j) be a matrix (hereinafter also referred to as an observation signal matrix) composed of observation signals of frame length L input from N microphones. In this case, the observed signal matrix S (j) is expressed by equation (1)

j is the index of the frame, l is the index of the sample in frame j, n is the index of the microphone, l = 0, 1,..., L−1, n = 1, 2,. In the following description, unless otherwise specified, processing is performed for each frame j, and the index j of the frame is omitted.

Here, the observed signal s _n (l) from the average s ^- _n (the average of the observation signal of each microphone n, for example, a time _{average, s n (0), s} n (1), ..., s _n (L-1) mean value) is calculated by subtracting an, autocovariance _σ s_ns_n (However, A_B in subscript represents a _B, s_ns_n is s _n s _n represents an) in normalized value m A matrix M having _n (l) as an element is defined as in equations (2) and (3).

Here we introduce the Karhunen-Loeve transformation. This conversion decomposes highly correlated signals into components independent of each other. An N-by-N square matrix M ^T M expressed by the following equation is obtained by Karoonen-Loeve transform.

However, the element σ _{m_j, m_k} (where j = 1,2, ..., N, k = 1,2, ..., N of the square matrix M ^T M, and the subscript m_jm_k is m _j m _k The value of (represents) indicates the degree of correlation between the microphones j and k. This square matrix M ^T M is further subjected to eigenvalue decomposition according to the following equation.

Here, E is a square matrix having N eigenvectors e _n = [e _{n, 1} , e _{n, 2} ,..., E _{n, N} ] ^T (number of elements N) as elements. Λ is a diagonal matrix having N eigenvalues λ _n as diagonal elements. Furthermore, the eigenvalues lambda _n and eigenvectors e _n, rearranges to be a descending order with respect to the eigenvalue lambda _n. When the rearranged N eigenvalues λ _n are _expressed as λ (1), λ (2),..., Λ (N), λ (1)> λ (2)>. N eigenvectors e _n after sorting in descending order with respect to the eigenvalue λ _n are _expressed as e (1), e (2),..., E (N), and e (n) = [e (n, 1) , e (n, 2),..., e (n, N)] ^T.

  Next, an L × N matrix μ obtained as a product of the matrix M and the matrix E is defined by the following equation.

Since the following equation (7) holds because of the orthogonality, the normalized original matrix M can be obtained by equation (8).

The diagonal matrix Λ can be approximated by truncating eigenvalues λ (n) arranged in descending order at a certain threshold. In this case, the approximate matrix Λ ′ and the matrix E of the diagonal matrix Λ have the relationship of Equation (9).

Note that the threshold and N ′ may be set to any value as long as Expression (9) is satisfied. For example, a threshold value y ₁ (0 <y ₁ <1) may be provided for the sum x ′ from eigenvalues λ (1) to λ (n ′) ((x ′ <y ₁ of S4 in FIG. 1). )), Or a threshold value y ₂ (1 ≦ y ₂ ≦ N) may be provided for the number n ′ of eigenvalues from eigenvalues λ (1) to λ (n ′) (in S4 of FIG. 1). (equivalent to n ′ <y ₂ ?)), a threshold value y ₃ (0 <y ₃ <1) may be provided for each eigenvalue λ (n ′) ((λ (n ′) of S4 in FIG. 1). > y ₃ ?)). Furthermore, these conditions may be combined (for example, an and condition or an condition is set for at least two conditions described above). The magnitudes and conditions of the thresholds y ₁ , y ₂ , and y ₃ may be determined in advance prior to transmission through experiments and simulations. Note that if the sum x from the eigenvalues λ (1) to λ (N ') is 1 or more, the diagonal matrix Λ cannot be approximated appropriately, so N' is set so that the sum x is less than 1. (Refer to S3 and S6 in FIG. 1).

  By using the N-row N'-column matrix E 'instead of the N-row N-column matrix E in Equation (6), the L-row N'-column is reduced in dimension from the L-row N-column matrix μ by the following equation: Can be obtained (Equation (10)).

From this, a matrix M ′ of L rows and N columns, which is an approximation of the matrix M of L rows and N columns, can be obtained as in the following equation (11).

S ′, which is an approximate value matrix of the observation signal matrix S, can be obtained by Expression (12).

Here, the element s ′ _n (l) is expressed by the following equation (13).

The error between the observed signal matrix S and its approximate value matrix S 'is defined as R (Equation (14))

Here, the error R is a matrix of L rows and N columns like the observation signal matrix S, but is a difference between the observation signal matrix S and its approximate value matrix S ′, and each element of the error R is a small value. Become. Therefore, each element of the error R can be expressed with a smaller number of bits than each element of the observation signal matrix S, and the transmission amount can be reduced as compared with sending the observation signal matrix S. From the above, instead of transmitting the original observed signal matrix S, μ ′, E ′, R, and σ ² = [σ _{s_1s_1} σ _{s_2 s_2} … σ _{s_n s_n} ] are transmitted to reduce the transmission amount Is possible.

In the encoding step, μ ′, E ′, R, and σ ² are calculated using the observed signal matrix S, and μ ′, E ′, R, and σ ² are transmitted instead of the observed signal matrix S.

In the decoding step, μ ′, E ′, R, and σ ² are received, and using these values, the observed signal matrix S is decoded by the following equation (15).

However, ". *" Is an operator that represents the multiplication of matrix elements.

It is. Also,

It is. That is, Σ ² is a matrix having L σ ² as elements.

  Hereinafter, a configuration for realizing the above transmission method will be described.

<Transmission system 10 according to the first embodiment>
FIG. 2 shows a functional diagram of the transmission system 10 according to the first embodiment.

The transmission system 10 includes an encoding device 100 and a decoding device 200. The encoding device 100 and the decoding device 200 can communicate with each other via the communication line 11. Encoding apparatus 100 inputs the N number of N picked up respectively by the microphone 12-n of the observation signal s _n, code (μ ', E', R , σ 2) generates, the communication line 11 To the decoding device 200. However, n = 1, 2,..., N, and s _n = [s _n (0) s _n (1)... S _n (L−1)]. Decoding device 200 receives a code (μ ', E', R , σ 2), to decode the N pieces of observed signal s _n with the sign. For example, decoding device 200, respectively to the N speakers 13-n and outputs N observation signals s _n, N speakers 13-n plays the N pieces of observed signal s _n respectively of N The sound field where the microphone 12-n is arranged is reproduced.

<Encoder 100>
FIG. 3 is a functional block diagram of the encoding apparatus 100, and FIG. 4 shows a processing flow thereof.

  The encoding apparatus 100 includes a normalization unit 110, a variance calculation unit 120, a Karoonen-Labe conversion unit 130, a rearrangement unit 140, a truncation unit 150, a dimension reduction unit 160, and an error calculation unit 170.

<Normalization unit 110>
Normalizing unit 110 receives the N number of N picked up respectively by the microphone 12-n of the observation signal s _n, and generates an observation signal matrix S according formula based on these values (1 '), the variance calculator 120 and the error calculation unit 170.

Further, the normalization unit 110, using N observation signals s _n, formula (2), to generate a matrix M according to (3), Karhunen-Loeve transform unit 130, abort section 150, the dimension reduction unit 160 Output.

Each element m _n matrix M (l), the average s of each observed signal s _n and the observed signal s _n (l) ^- _n (eg, time average) auto-covariance of the observed signal s _n a difference between σ It is a value normalized by _{s_ns_n} .

<Distributed calculation unit 120>
Variance calculator 120 receives the observed signal matrix S, the variance sigma _{S_ns_n} of each observed signal s _n calculated (S120), variance _{σ = [σ S_1S_1 σ S_2S_2 ...} σ S_NS_N] as the output value of the encoding apparatus 100 At the same time, it is output to the error calculator 170.

<Kalunen-Labe conversion unit 130>
The Karoonen-Loeve transform unit 130 receives the matrix M, performs the Karoonen-Labe transform (S130), obtains a square matrix M ^T M expressed by Equation (4), and outputs the square matrix M ^T M to the rearrangement unit 140.

<Sort section 140>
Rearranging unit 140 receives the square matrix M ^T M, eigenvalue decomposition (Equation (5)) to give the N eigenvalues lambda _n and eigenvectors e _n square matrix M ^T M.

Further, the rearrangement unit 140 rearranges the eigenvalues λ _n so as to be in descending order (S140), and the N eigenvalues λ (1), λ (2),. Diagonal matrix Λ = diag (λ (1), λ (2), ..., λ (N)) with N as λ (1)> λ (2)>...> λ (N)) L = N matrix E = [e (1) e (2)... E (N)] consisting of eigenvectors e (1), e (2),. . The eigenvector e (n) is an eigenvector corresponding to the eigenvalue λ (n).

<Cutoff part 150>
The truncation unit 150 includes a diagonal matrix Λ whose diagonal elements are N eigenvalues λ (n) after rearrangement, and N eigenvectors e (1), e (2),. An L-by-N matrix E = [e (1) e (2) ... e (N)] consisting of (N), and a magnitude relationship between a predetermined threshold value y and a value corresponding to the eigenvalue λ (n ′) N ′ is set based on (see S6 in FIG. 1), and a matrix E ′ = [e () consisting of only N ′ eigenvectors e (n ′) respectively corresponding to N ′ eigenvalues λ (n ′). 1) e (2)... E (N ′)] is obtained (S150), and is output as an output value of the encoding apparatus 100, and is also output to the dimension reduction unit 160 and the error calculation unit 170. However, N ′ <N, n ′ = 1, 2,..., N ′. The values corresponding to the eigenvalue λ (n ′) are, for example, (i) the sum x ′ from the eigenvalue λ (1) to λ (n ′), and (ii) from the eigenvalue λ (1) to λ (n ′). The number of eigenvalues n ′, (iii) the eigenvalue λ (n ′) itself, (iv) a combination of the above (i) to (iii) (see S4 in FIG. 1). The predetermined threshold value y is, for example, y ₁ , y ₂ , y _{3 in} FIG. 1 or a combination thereof.

Note that if the sum x from the eigenvalues λ (1) to λ (N ') is 1 or more, the diagonal matrix Λ cannot be approximated appropriately, so N' is set so that the sum x is less than 1. (Refer to S3 and S6 in FIG. 1). Also, N ′ is set so that N ′ <N (see S3 and S6 in FIG. 1).

<Dimension reduction unit 160>
The dimension reduction unit 160 receives the matrix M and the matrix E ′, obtains a matrix μ ′ = ME ′, which is a product of the matrix M and the matrix E ′, according to the equation (10) (S160), and While outputting as an output value, it outputs to the error calculation part 170.

<Error calculation unit 170>
The error calculation unit 170 receives the observation signal matrix S, the matrix μ ′, the matrix E ′, and the variance σ ^2, and uses them to approximate the observation signal matrix S according to the equations (11) to (13). Ask for '.

Further, the error calculation unit 170 obtains a difference between the observed signal matrix S and the approximate value matrix S ′ as an error R according to the equation (14) (S170), and outputs it as an output value of the encoding device 100.

The encoding apparatus 100 transmits a code (μ ′, E ′, R, σ ² ) for each frame from a transmission unit (not shown).

<Decoding device 200>
FIG. 5 shows a functional block diagram of the decoding apparatus 200, and FIG. 6 shows a processing flow thereof.

  The decoding device 200 includes a decoding unit 210 and a sound signal generation unit 220. The decoding unit 210 includes an approximate matrix calculation unit 211 and an observation signal prediction unit 212.

<Decoding unit 210>
The decoding unit 210 receives the code (μ ′, E ′, R, σ ² ), and uses these values to decode the observation signal matrix S according to the following equation (S210) and outputs it to the sound signal generation unit 220. To do.

However, ". *" Is an operator that represents the multiplication of matrix elements.

It is. For example, the observation signal matrix S is decoded as follows.

  The approximate matrix calculation unit 211 of the decoding unit 210 obtains an approximate matrix M ′ of the matrix M using the matrix μ ′ and the matrix E ′ according to the equation (11) (S211), and outputs it to the observation signal prediction unit 212. .

  The observation signal prediction unit 212 of the decoding unit 210 receives the approximation matrix M ′, the error R, and the variance σ, and uses the approximation matrix M ′ and the variance σ to observe the observation signal matrix S according to the equations (12) and (13). An approximate value matrix S ′ is obtained.

  Further, the observation signal prediction unit 212 uses the approximate value matrix S ′ and the error R to obtain the observation signal matrix S according to the following equation (S212) and outputs it to the sound signal generation unit 220.

<Sound signal generation unit 220>
The sound signal generator 220 receives the observed signal matrix S, separates the observed signal (reproduced signal) s _n of the speakers 13-n, respectively and outputs to the speaker 13-n.

Each speaker 13-n receives the observed signal (reproduced signal) s _n, to play.

<Effect>
With the above configuration, sound space information can be efficiently transmitted. The error R is a matrix of L rows and N columns similarly to the observation signal matrix S, but is a difference between the observation signal matrix S and its approximate value matrix S ′, and each element of the error R has a small value. Therefore, each element of the error R can be expressed with a smaller number of bits than each element of the observation signal matrix S, and the transmission amount can be reduced as compared with sending the observation signal matrix S. From the above, it is possible to reduce the transmission amount by transmitting the code (μ ′, E ′, R, σ ² ) instead of transmitting the original observation signal matrix S.

<Other variations>
The present invention is not limited to the above-described embodiments and modifications. For example, the various processes described above are not only executed in time series according to the description, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes. In addition, it can change suitably in the range which does not deviate from the meaning of this invention.

<Program and recording medium>
In addition, various processing functions in each device described in the above embodiments and modifications may be realized by a computer. In that case, the processing contents of the functions that each device should have are described by a program. Then, by executing this program on a computer, various processing functions in each of the above devices are realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Further, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its storage unit. When executing the process, this computer reads the program stored in its own storage unit and executes the process according to the read program. As another embodiment of this program, a computer may read a program directly from a portable recording medium and execute processing according to the program. Further, each time a program is transferred from the server computer to the computer, processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program includes information provided for processing by the electronic computer and equivalent to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In addition, although each device is configured by executing a predetermined program on a computer, at least a part of these processing contents may be realized by hardware.

Claims (8)

N is any integer of 2 or more, and is a transmission system for transmitting an observation signal matrix S collected by N microphones from an encoding device to a decoding device,
The encoding device includes:
n = 1, 2, ..., and N, and each of the observed signal s _n of each microphone, the average s of each observed signal s _n ^- regular self-covariance sigma _{S_ns_n} of each observed signal the difference between _n s _n A normalization unit that generates a matrix M whose elements are generalized values m _n ,
A reordering unit that reorders the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M in descending order with respect to the eigenvalues λ _n ;
N ′ <N, n ′ = 1, 2,..., N ′, and the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), λ (n) Let e (n) be the eigenvector corresponding to, and based on the magnitude relationship between the predetermined threshold and the value corresponding to the eigenvalue λ (n ′), N ′ pieces of N ′ corresponding to N ′ eigenvalues λ (n ′) respectively A truncation unit for obtaining a matrix E ′ = [e (1) e (2)... E (N ′)] consisting only of eigenvectors e (n ′);
An approximate value matrix of the observed signal matrix S using a variance σ _{s_ns_n} of each of the N observed signals s _n , a matrix μ ′ that is a product of the matrix M and the matrix E ′, and the matrix E ′ An error calculator that calculates S ′ and calculates the difference between the observed signal matrix S and the approximate value matrix S ′ as an error R, and
The decoding device
A decoding unit for decoding the observed signal matrix S using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n} and the error R;
Transmission system. An encoding device that generates a code corresponding to an observation signal matrix S collected by N microphones, where N is any integer of 2 or more,
n = 1, 2, ..., and N, and each of the observed signal s _n, the time average s of each observed signal s _n ^- normalized by auto-covariance sigma _{S_ns_n} of the difference between _n each observed signal s _n A normalization unit that generates a matrix M having elements of value m _n ,
A reordering unit that reorders the N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M in descending order with respect to the eigenvalues λ _n ;
N ′ <N, n ′ = 1, 2,..., N ′, and the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), λ (n) An eigenvector corresponding to e (n), and a truncation unit for obtaining a matrix E ′ consisting only of eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold value;
An approximate value matrix of the observed signal matrix S using a variance σ _{s_ns_n} of each of the N observed signals s _n , a matrix μ ′ that is a product of the matrix M and the matrix E ′, and the matrix E ′ An error calculation unit that calculates S ′ and calculates a difference between the observed signal matrix S and the approximate value matrix S ′ as an error R.
Encoding device. The encoding device according to claim 2, comprising:
M ′ = μ ′ (E ′) ^T , each element of M ′ is m ′ _n , each approximate value of each observation signal s _n is s ′ _n , and the error calculation unit is
s ' _n = m' _n σ _{s_ns_n}
To obtain the approximate value matrix S ′ = [s ′ ₁ s ′ ₂ ... S ′ _N ],
Encoding device. N is any integer of 2 or more, receives a code corresponding to an observation signal matrix S collected by N microphones, and decodes the observation signal matrix S,
n = 1,2, ..., and N, and each of the observed signal s _n, times the average s of each observed signal s _n ^- _n value difference normalized with the auto-covariance sigma _{S_ns_n} of each observed signal s _n and m A matrix having _n as an element is M, and N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M are rearranged so as to be in descending order with respect to the eigenvalue λ _n. (n), the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), the eigenvector corresponding to λ (n) is e (n), and N ′ <N, n ′ = 1, 2,..., N ′, and E ′ a matrix consisting only of eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold value, An approximate value matrix of the observed signal matrix S obtained by using the variance σ _{s_ns_n} of each of the N observed signals s _n , the matrix μ ′ = ME ′, and the matrix E ′ is defined as S ′, and the observed signal matrix S and The difference from the approximate value matrix S ′ is the error R,
A decoding unit that decodes the observed signal matrix S using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n,} and the error R;
Decoding device. N is any integer of 2 or more, and is a transmission method for transmitting an observation signal matrix S collected by N microphones from an encoding device to a decoding device,
n = 1, 2, ..., and N, the coding apparatus, and the observed signal s _n of each microphone, the average s of each observed signal s _n ^- self of each observed signal the difference between _n s _n A normalization step for generating a matrix M having elements m _n normalized by the covariance σ _{s_ns_n} ;
The encoding apparatus, the N eigenvalues lambda _n and eigenvectors e _n square matrix M ^T M, and rearranging step for rearranging so that descending with respect to the eigenvalue lambda _n,
N ′ <N, n ′ = 1, 2,..., N ′, and the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), λ (n) The eigenvector corresponding to e (n) is set to e (n), and the encoding device determines each of N ′ eigenvalues λ (n ′) based on a magnitude relationship between a predetermined threshold and a value corresponding to the eigenvalue λ (n ′). A truncation step to obtain a matrix E ′ = [e (1) e (2)... E (N ′)] consisting only of the corresponding N ′ eigenvectors e (n ′);
The encoding apparatus, the variance sigma _{S_ns_n} of N each observed signal s _n, a 'matrix μ is the product of' the matrix E and the matrix M, the observed signal using said matrix E ' Obtaining an approximate value matrix S ′ of the matrix S, and calculating an error R as a difference between the observed signal matrix S and the approximate value matrix S ′;
The decoding apparatus includes a decoding step of decoding the observed signal matrix S using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n} and the error R,
Transmission method. An encoding method for generating a code corresponding to an observation signal matrix S collected by N microphones, where N is any integer of 2 or more,
n = 1, 2, ..., and N, normalization section, and each of the observed signal s _n, the time of each observation signal s _n Mean s ^- wherein a difference between _n each observed signal s _n autocovariance σ of a normalization step for generating a matrix M having elements m _n normalized by _{s_ns_n} ;
Rearranging portion, a N eigenvalues lambda _n and eigenvectors e _n square matrix M ^T M, and rearranging step for rearranging so that descending with respect to the eigenvalue lambda _n,
N ′ <N, n ′ = 1, 2,..., N ′, and the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), λ (n) The eigenvector corresponding to is e (n), and the truncation unit obtains a matrix E ′ consisting only of eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold. When,
Error calculation section, and the variance sigma _{S_ns_n} of N each observed signal s _n, a 'matrix μ is the product of' the matrix E and the matrix M, the observed signal using said matrix E 'Toto Calculating an approximate value matrix S ′ of the matrix S, and calculating an error R as a difference between the observed signal matrix S and the approximate value matrix S ′.
Encoding method. N is any integer of 2 or more, receives a code corresponding to the observed signal matrix S collected by N microphones, and decodes the observed signal matrix S,
n = 1,2, ..., and N, and each of the observed signal s _n, times the average s of each observed signal s _n ^- _n value difference normalized with the auto-covariance sigma _{S_ns_n} of each observed signal s _n and m A matrix having _n as an element is M, and N eigenvalues λ _n and eigenvectors e _n of the square matrix M ^T M are rearranged so as to be in descending order with respect to the eigenvalue λ _n. (n), the rearranged eigenvalues are λ (n), λ (1)> λ (2)>...> λ (N), the eigenvector corresponding to λ (n) is e (n), and N ′ <N, n ′ = 1, 2,..., N ′, and E ′ a matrix consisting only of eigenvectors e (n ′) corresponding to N ′ eigenvalues λ (n ′) that are equal to or less than a predetermined threshold value, An approximate value matrix of the observed signal matrix S obtained by using the variance σ _{s_ns_n} of each of the N observed signals s _n , the matrix μ ′ = ME ′, and the matrix E ′ is defined as S ′, and approximated to the observed signal matrix S The difference from the value matrix S ′ is the error R,
The decoding unit includes a decoding step of decoding the observation signal matrix S using the matrix μ ′, the matrix E ′, the N variances σ _{s_ns_n} and the error R,
Decryption method.   A program for causing a computer to function as the encoding device according to claim 2 or claim 3 or the decoding device according to claim 4.

Images