JP2003084793A - Method, device, and program for analyzing independent component and recording medium with this program recorded thereon - Google Patents
Method, device, and program for analyzing independent component and recording medium with this program recorded thereonInfo
- Publication number
- JP2003084793A JP2003084793A JP2001279058A JP2001279058A JP2003084793A JP 2003084793 A JP2003084793 A JP 2003084793A JP 2001279058 A JP2001279058 A JP 2001279058A JP 2001279058 A JP2001279058 A JP 2001279058A JP 2003084793 A JP2003084793 A JP 2003084793A
- Authority
- JP
- Japan
- Prior art keywords
- separation
- independence
- signal
- matrix
- separation matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 15
- 239000011159 matrix material Substances 0.000 claims abstract description 52
- 230000004913 activation Effects 0.000 claims abstract description 31
- 238000012937 correction Methods 0.000 claims abstract description 22
- 238000000926 separation method Methods 0.000 claims description 60
- 238000004364 calculation method Methods 0.000 claims description 25
- 238000012880 independent component analysis Methods 0.000 claims description 14
- 238000012986 modification Methods 0.000 claims 1
- 230000004048 modification Effects 0.000 claims 1
- 230000006870 function Effects 0.000 description 31
- 238000010586 diagram Methods 0.000 description 12
- 230000005236 sound signal Effects 0.000 description 3
- 210000004556 brain Anatomy 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
Description
【0001】[0001]
【発明の属する技術分野】本発明は信号処理の技術分野
に属し、観測したい元の信号は直接観測はできないが、
いくつかの信号が混合されたもののみが観測できる状況
において、元の信号を推定する技術に関する。本技術に
より、様々な妨害信号が発生する実環境において、目的
の信号を精度良く取り出すことが可能となる。音信号に
対する応用例としては、話者とマイクが離れた位置にあ
りマイクが話者の音声以外の音を拾ってしまうような状
況でも、認識率の高い音声認識装置を構成できる。ま
た、脳の仕組みを明らかにする研究においては、1つ1
つの脳波を直接観測することはできず、複数の混合され
た脳波を脳の外部において観測することになるが、本技
術により1つ1つの脳波を精度良く推定できる。BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention belongs to the technical field of signal processing, and the original signal to be observed cannot be directly observed.
The present invention relates to a technique for estimating an original signal in a situation where only a mixture of some signals can be observed. According to the present technology, it is possible to accurately extract a target signal in a real environment where various interference signals are generated. As an application example of a sound signal, a voice recognition device having a high recognition rate can be configured even in a situation where the microphone is picked up by a microphone other than the speaker because the microphone is far from the speaker. In addition, in research to clarify the mechanism of the brain,
Although it is not possible to directly observe one electroencephalogram, a plurality of mixed electroencephalograms are to be observed outside the brain, but with the present technology, each electroencephalogram can be accurately estimated.
【0002】[0002]
【従来の技術】複数の線形混合された信号を独立性に基
づいて分離する技術は、独立成分分析(ICA:Independen
t Component Analysis)と呼ばれる。その中でも、信
号が複素数の系列である場合(すなわち複素信号である
場合)、複素信号に対する独立成分分析が用いられる例
としては、実環境において残響を含めて混合された音信
号を分離する際、音信号をフーリエ変換して周波数領域
で表現する事例が代表的である。ここでは、まず、実信
号に対する独立成分分析の方法を説明し、その後、従来
技術による複素信号への拡張を説明する。2. Description of the Related Art A technique for separating a plurality of linearly mixed signals based on independence is known as Independent Component Analysis (ICA).
t Component Analysis). Among them, when the signal is a sequence of complex numbers (that is, when the signal is a complex signal), an example of using the independent component analysis for the complex signal is, when separating a mixed sound signal including reverberation in a real environment, A typical example is a case where a sound signal is Fourier transformed and expressed in the frequency domain. Here, first, a method of independent component analysis for a real signal will be described, and then, an extension to a complex signal according to the related art will be described.
【0003】[独立成分分析]互いにN個の源信号s(t)=
[s1(t),・・・,sN(t)]TがM×N行列Aにより線形混合x(t)
=As(t)されて、M個のセンサによりx(t)=[x1(t),・・
・,xM(t)]Tが観測されたとする。ICAの目的は、混合系A
や源信号s(t)を知らずに、x(t)を互いに独立なN個の信
号y(t)=[y1(t),・・・,yN(t)]T=Wx(t)に分離するN×M
行列Wを求めることである。図1にN=M=2の場合を示
す。[Independent Component Analysis] N source signals s (t) =
[s 1 (t), ・ ・ ・, s N (t)] T is a linear mixture x (t) by M × N matrix A
= As (t), and M (sensors) x (t) = [x 1 (t), ...
・, X M (t)] T is observed. The purpose of ICA is mixed system A
Or n (t) = [y 1 (t), ..., y N (t)] T = Wx (t) without knowing the source signal s (t) ) N × M
To find the matrix W. FIG. 1 shows the case where N = M = 2.
【0004】[独立成分分析の方法]分離行列Wは、y(t)
の各要素間の相互情報量の最小化を目指して、学習則W
=W+ΔWにより徐々に改良される。ΔWは、自然勾配法と
呼ばれるΔW=μ[I−<φ[y(t)]y(t)T>]Wの式に従って
計算される。ここでIは単位行列、μは学習の速度を制
御する小さな定数値、<φ[y(t)]y(t)T>はφ[y(t)]y
(t)Tの時間平均(tに関する平均)を表す。<φ[y(t)]y
(t)T>は、N×N行列であることに注意されたい。また、
φ[・]は活性化関数と呼ばれるものであり、一般にφ[y
(t)]=tanh[η・y(t)]が非線形の活性化関数として広く
用いられている。ηは非線形性の強さを制御するパラメ
ータである。以下では簡単のため、時間tを省略してy、
φ[y]と記載する。[Method of Independent Component Analysis] Separation matrix W is y (t)
Aiming to minimize the mutual information between each element of
= W + ΔW gradually improves. ΔW is calculated according to the equation of ΔW = μ [I− <φ [y (t)] y (t) T >] W called the natural gradient method. Where I is an identity matrix, μ is a small constant value that controls the learning speed, and <φ [y (t)] y (t) T > is φ [y (t)] y.
(t) represents the time average of T (average with respect to t). <Φ [y (t)] y
Note that (t) T > is an N × N matrix. Also,
φ [・] is called an activation function, and φ [y] is generally
(t)] = tanh [η · y (t)] is widely used as a nonlinear activation function. η is a parameter that controls the strength of nonlinearity. For simplicity, the time t is omitted and y,
Describe as φ [y].
【0005】[複素信号への拡張]以上が独立成分分析の
方法であるが、複素数を扱うためには、ΔWの計算を複
素数に拡張する必要がある。これまでには、以下の拡張
が提案されている。
ΔW=μ[I−<φ[y]yH>]W (1)
Φ[y]=φ[re(y)]+j・φ[im(y)] (2)
ここで、yHはyの共役転置(複素数の共役を取り、転置
を行う)、re(y)とim(y)はそれぞれyの実部と虚部であ
る。なお、Φ[・]は実関数φ[・]の複素数への拡張であ
る。[Extension to Complex Signal] The above is the method of independent component analysis, but in order to handle a complex number, it is necessary to extend the calculation of ΔW to a complex number. The following extensions have been proposed so far. ΔW = μ [I- <φ [y] y H >] W (1) Φ [y] = φ [re (y)] + j · φ [im (y)] (2) where y H is y The conjugate transpose of (taking the conjugate of a complex number and transposing), re (y) and im (y) are the real and imaginary parts of y, respectively. Note that Φ [•] is an extension of the real function Φ [•] to a complex number.
【0006】[収束点での状況]さて、学習則W=W+ΔWに
従ってΔWは0に収束することから、式(1)によるとW
は
<Φ[yp]yq *>=0(p≠q ) (3)
<Φ[yp]yq *>=1(p=q ) (4)
を満たす点に収束する。ここで、yq *はyqの複素共役で
ある。制約(3)は、ypとyqが互いに独立である場合に
満たされる。従って、式(1)が持つこの制約により、
ypとyqの独立性が高まる。一方式(4)では、p=qの場
合を扱っているが、これによりypの振幅の平均値がある
値に近づくことになる。[Situation at Convergence Point] Since ΔW converges to 0 according to the learning rule W = W + ΔW, according to the equation (1), W
Converges to a point that satisfies <Φ [y p ] y q * > = 0 (p ≠ q) (3) <Φ [y p ] y q * > = 1 (p = q) (4). Where y q * is the complex conjugate of y q . Constraint (3) is satisfied when y p and y q are independent of each other. Therefore, due to this constraint of equation (1),
The independence of y p and y q increases. On the other hand, in the equation (4), the case of p = q is dealt with, but by this, the average value of the amplitudes of y p approaches a certain value.
【0007】[余分な制約]しかし、上記の方法では、余
分な制約が発生して収束を阻むことがある。すなわち、
式(4)を実部と虚部に分解すると、
<φ[re(yp)]re(yp)+φ[im(yp)]im(yp)>=1 (5)
<φ[im(yp)]re(yp)−φ[re(yp)]im(yp)>=0 (6)
となる。ここで式(6)が余分な制約を課していること
がわかる。例えば re(yp)とim(yp)が互いに独立であれ
ばこの制約を満たすが、一般には満たさない。そうする
と、式(1)に基づくΔWの計算で、ΔWがいつまでも0
に収束しないことがある。[Extra Constraints] However, in the above method, extra constraints may occur to prevent convergence. That is,
Decomposing equation (4) into a real part and an imaginary part, <φ [re (y p )] re (y p ) + φ [im (y p )] im (y p )> = 1 (5) <φ [ im (y p )] re (y p ) −φ [re (y p )] im (y p )> = 0 (6). It can be seen that equation (6) imposes an extra constraint. For example, if re (y p ) and im (y p ) are independent of each other, this constraint is satisfied, but generally it is not. Then, in the calculation of ΔW based on equation (1), ΔW is always 0.
May not converge to.
【0008】[0008]
【発明が解決しようとする課題】従来の技術では、活性
化関数の複素数への拡張として式(2)が提案されてい
るが、上記の余分な制約(6)が発生して収束を阻むこ
とがある。そこで本発明の目的は、上記の様な余分な制
約が発生しない新たな活性化関数を提供することにあ
る。In the prior art, equation (2) is proposed as an extension of the activation function to a complex number, but the above-mentioned extra constraint (6) occurs to prevent convergence. There is. Therefore, an object of the present invention is to provide a new activation function in which the above-mentioned extra constraint does not occur.
【0009】[0009]
【課題を解決するための手段】上記目的を達成するため
に、本発明では、複素数の絶対値のみを変更する活性化
関数を用いて独立性に相当する値を計算する手段を備え
る。一般に複素数yは、絶対値|y|と偏角θ=angle(y)
を用いてy=|y|・exp(jθ)と表現できる。絶対値のみ
を変更する活性化関数は、偏角θを変更しないため、複
素数を入力として実数を出力する任意の関数α(・)とす
ると、Φ[y]=α(y)・exp(jθ)と表記できる。本発明で
は、複素数の絶対値のみを変更する活性化関数を用い
て、独立性に相当する値を計算する。これにより、式
(4)において、式(6)のような余分な制約は発生し
ない。なぜなら、yp *がypの複素共役であることから、
θ=angle(y p)とすると、Φ[yp]yp *=α(yp)・exp(jθ)
・|yp|・exp(−jθ)=α(yp)|yp|となり、虚部が常
に0になるからである。[Means for Solving the Problems] To achieve the above object
In the present invention, activation that changes only the absolute value of a complex number
Equipped with means to calculate the value corresponding to independence using a function
It In general, complex number y is absolute value | y | and argument θ = angle (y)
Can be expressed as y = | y | · exp (jθ). Absolute value only
The activation function that changes
Let it be an arbitrary function α (・) that takes a prime number and outputs a real number
Then, it can be expressed as Φ [y] = α (y) · exp (jθ). In the present invention
Uses an activation function that changes only the absolute value of a complex number
And calculate the value corresponding to independence. This gives the formula
In (4), an extra constraint like Equation (6) occurs.
Absent. Because yp *Is ypIs a complex conjugate of
θ = angle (y p), Φ [yp] yp *= Α (yp) ・ Exp (jθ)
・ | yp| ・ exp (−jθ) = α (yp) | yp| And the imaginary part is always
Because it becomes 0.
【0010】[0010]
【発明の実施の形態】図2は、本発明の独立成分分析装
置の構成を示すブロック図である。分離信号計算部1
は、分離行列保持部に分離行列Wを保持し、混合信号x
(t)=[x 1(t),・・・,xM(t)]Tから分離信号y(t)=[y1(t),
・・・,yN(t)]T=Wx(t)を計算する。また、分離信号y(t)
の独立性が高まるように、分離行列の修正値ΔWと学習
則W=W+ΔWに従って、分離行列を徐々に修正する。分離
行列修正値計算部2は、現状のW(分離行列保持部に保
持している分離行列W)と分離信号y(t)と活性化関数の
値Φ[y]から、分離行列の修正値ΔWを計算する。活性化
関数の値Φ[y]は、活性化関数計算部3にて、現状の分
離信号y(t)から計算される。DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 2 shows the independent component analysis device of the present invention.
It is a block diagram which shows the structure of a device. Separated signal calculator 1
Holds the separation matrix W in the separation matrix holding unit, and the mixed signal x
(t) = [x 1(t), ..., xM(t)]TSeparated signal from y (t) = [y1(t),
..., yN(t)]T= Wx (t) is calculated. Also, the separated signal y (t)
So that the independence of
The separation matrix is gradually modified according to the rule W = W + ΔW. Separation
The matrix correction value calculation unit 2 stores in the current W (separation matrix holding unit).
Of separation matrix W), separation signal y (t) and activation function
From the value Φ [y], the correction value ΔW of the separation matrix is calculated. activation
The function value Φ [y] is the current value in the activation function calculation unit 3.
Calculated from the separation signal y (t).
【0011】[分離行列修正値計算部]図3は、分離行列
修正値計算部2の構成を示すブロック図である。ベクト
ル積計算部2−1では、分離信号y(t)と活性化関数の値
Φ[y(t)]から、ベクトル積Φ[y(t)]y(t)Hを計算する。
平均値計算部2−2では、その結果を総和してサンプル
数で割ることにより、平均値<Φ[y(t)]y(t)H>を求め
る。その後、修正値計算部2−3では、その平均値と分
離行列WからΔW=μ[I−<Φ[y(t)]y(t)H>]Wを計算す
る。[Separation Matrix Correction Value Calculation Unit] FIG. 3 is a block diagram showing the configuration of the separation matrix correction value calculation unit 2. The vector product calculator 2-1 calculates the vector product Φ [y (t)] y (t) H from the separated signal y (t) and the activation function value Φ [y (t)].
The average value calculator 2-2 sums the results and divides by the number of samples to obtain the average value <Φ [y (t)] y (t) H >. Then, the correction value calculation unit 2-3 calculates ΔW = μ [I− <Φ [y (t)] y (t) H >] W from the average value and the separation matrix W.
【0012】[活性化関数計算部]本実施例では、複素数
の絶対値のみを変更する活性化関数として、
Φ[y]=φ[|y|]・exp(jθ)、θ=angle(y) (7)
を用いる。これは、実数に対する活性化関数φ[・]の自
然な拡張であり、実数に対しては双方とも同じ値を出力
する。[Activation Function Calculation Unit] In this embodiment, as activation functions that change only the absolute value of a complex number, Φ [y] = φ [| y |] exp (jθ), θ = angle (y ) (7) is used. This is a natural extension of the activation function φ [·] for real numbers, and both output the same value for real numbers.
【0013】図4は活性化関数計算部3の第1の実施形
態を示すブロック図である。簡単のため、時間tの表記
は省略している。偏角計算部3−1ではθ=angle(y)を
計算し、その後、指数関数計算部3−2でexp(jθ)を計
算する。また、絶対値計算部3−3では |y|を計算
し、その後、非線形関数計算部3−4にてφ[|y|]を計
算する。双方の計算を終えると、乗算部3−5において
2つの結果を掛け合わせてΦ[y]=φ[|y|]・exp(jθ)を
得る。上記の活性化関数は別の方法でも計算できる。す
なわち、exp(j・angle(y))=y/|y|であることから、Φ
[y]=φ[|y|]・y/|y|と計算できる。これにより、上記
での偏角計算部と指数関数計算部が不要となる。図5
は、活性化関数計算部3の第2の実施形態を示すブロッ
ク図である。絶対値計算部3−10で分離信号の絶対値
|y|を計算し、非線形関数計算部3−11にてφ[|y|]
を計算するのは、上記の実施形態と同様である。乗除算
部3−12では、計算し終わったφ[|y|]と|y|、およ
び元のyから、Φ[y]=φ[|y|]・y/|y|を計算する。FIG. 4 is a block diagram showing a first embodiment of the activation function calculator 3. For simplicity, the notation of time t is omitted. The argument calculating unit 3-1 calculates θ = angle (y), and then the exponential function calculating unit 3-2 calculates exp (jθ). The absolute value calculation unit 3-3 calculates | y |, and then the nonlinear function calculation unit 3-4 calculates φ [| y |]. After completing both calculations, the multiplication unit 3-5 multiplies the two results to obtain Φ [y] = φ [| y |] · exp (jθ). The activation function above can be calculated in other ways. That is, since exp (j ・ angle (y)) = y / | y |
[y] = φ [| y |] ・ y / | y | This eliminates the need for the argument calculation unit and the exponential function calculation unit described above. Figure 5
FIG. 7 is a block diagram showing a second embodiment of the activation function calculation unit 3. The absolute value calculation unit 3-10 calculates the absolute value | y | of the separated signal, and the nonlinear function calculation unit 3-11 calculates φ [| y |].
Is calculated in the same manner as in the above embodiment. The multiplication / division unit 3-12 calculates Φ [y] = φ [| y |] · y / | y | from the calculated φ [| y |] and | y | and the original y.
【0014】図6を参照して分離行列Wの修正手順を説
明する。分離信号yより活性化関数の値Φ[y]を計算する
(s-1)。分離信号yと活性化関数の値Φ[y]からベクトル
積Φ[y]yHを計算し、その結果を総和してサンプル数で
割ることにより平均値<Φ[y]yH>を求める。その平均
値と分離行列Wから分離行列修正値ΔW=μ[I−<Φ[y]yH
>]Wを計算する(s-2)。保持している分離行列Wと修正値
ΔWよりW=W+ΔWを計算する(s-3)。次にこの修正され
た分離行列Wを用いてy=Wxを計算する(s-4)。手順(s-1)
〜(s-4)を繰り返し独立性が十分に高まるまで分離行列W
を修正する。A procedure for correcting the separation matrix W will be described with reference to FIG. The value Φ [y] of the activation function is calculated from the separation signal y (s-1). The vector product Φ [y] y H is calculated from the separation signal y and the activation function value Φ [y], and the results are summed and divided by the number of samples to obtain the average value <Φ [y] y H >. . From the average value and the separation matrix W, the separation matrix correction value ΔW = μ [I− <Φ [y] y H
>] Calculate W (s-2). W = W + ΔW is calculated from the separation matrix W held and the correction value ΔW (s-3). Next, y = Wx is calculated using this modified separation matrix W (s-4). Step (s-1)
~ (S-4) is repeated until the independence is sufficiently increased.
To fix.
【0015】本発明の独立成分分析装置は、CPUやメモ
リ等を有するコンピュータと、アクセス主体となるユー
ザが利用する利用者端末と、記録媒体とから構成するこ
とができる。記録媒体はCD-ROM、磁気ディスク装置、半
導体メモリ等の機械読み取り可能な記録媒体であり、こ
こに記録されたアクセス制御用プログラムは、コンピュ
ータに読み取られ、コンピュータの動作を制御し、コン
ピュータ上に前述した実施形態の構成要素、すなわち、
分離信号計算部、分離行列修正値計算部、活性化関数計
算部等を実現する。The independent component analyzer of the present invention can be composed of a computer having a CPU, a memory, etc., a user terminal used by a user who is an access subject, and a recording medium. The recording medium is a machine-readable recording medium such as a CD-ROM, a magnetic disk device, and a semiconductor memory. The access control program recorded here is read by a computer, controls the operation of the computer, and is stored on the computer. The components of the above-described embodiment, namely,
A separation signal calculation unit, a separation matrix correction value calculation unit, an activation function calculation unit, etc. are realized.
【0016】[0016]
【発明の効果】本発明によれば、複素信号を対象とする
独立成分分析において、従来技術で発生していた余分な
制約(6)が発生しない。従って、従来技術では収束が
阻まれることがあったのに対し、本発明では滑らかな収
束が可能となる。この事実を示すものとして、図7と図
8にそれぞれ、従来技術の活性化関数(2)と本発明の
活性化関数(7)を用いた場合の収束の様子を示す。具
体的には、残響を含めて混合された音声をフーリエ変換
した後、複素信号に対する独立成分分析を行った際の、
[I−<Φ[y]yH>]の各要素の絶対値(Absolute Value)
を示す。横軸は、学習則を適用した繰り返しの回数(Ite
ration)である。明らかに、従来技術では収束が阻まれ
ているが、本発明では滑らかに収束している。これによ
り、非対角成分[1,2]と[2,1]によって示されているyの
相互情報量(小さいほど独立性が高い)は十分に小さく
なっている。According to the present invention, in the independent component analysis targeting a complex signal, the extra constraint (6) generated in the prior art does not occur. Therefore, while the conventional technique may prevent convergence, the present invention enables smooth convergence. As an indication of this fact, FIGS. 7 and 8 show the state of convergence when the activation function (2) of the prior art and the activation function (7) of the present invention are used, respectively. Specifically, after performing a Fourier transform on the speech mixed with reverberation, when performing independent component analysis on the complex signal,
The absolute value of each element of [I- <Φ [y] y H>] (Absolute Value)
Indicates. The horizontal axis is the number of iterations (Ite
ration). Obviously, the convergence is blocked in the prior art, but the convergence is smooth in the present invention. As a result, the mutual information amount of y (the smaller the higher the independence) indicated by the non-diagonal components [1,2] and [2,1] is sufficiently small.
【図1】独立成分分析を説明するための図。FIG. 1 is a diagram for explaining independent component analysis.
【図2】本発明の独立成分分析装置のブロック図。FIG. 2 is a block diagram of an independent component analyzer according to the present invention.
【図3】分離行列修正値計算部のブロック図。FIG. 3 is a block diagram of a separation matrix correction value calculation unit.
【図4】活性化関数計算部(第1の実施形態)のブロッ
ク図。FIG. 4 is a block diagram of an activation function calculation unit (first embodiment).
【図5】活性化関数計算部(第2の実施形態)のブロッ
ク図。FIG. 5 is a block diagram of an activation function calculation unit (second embodiment).
【図6】分離行列Wの修正手順を説明するための図。FIG. 6 is a diagram for explaining a correction procedure of a separation matrix W.
【図7】従来技術における収束の様子を示す図。FIG. 7 is a diagram showing a state of convergence in a conventional technique.
【図8】本発明における収束の様子を示す図。FIG. 8 is a diagram showing a state of convergence in the present invention.
1 分離信号計算部 2 分離行列修正値計算部 3 活性化関数計算部 1 Separated signal calculator 2 Separation matrix correction value calculator 3 Activation function calculator
フロントページの続き (72)発明者 荒木 章子 東京都千代田区大手町二丁目3番1号 日 本電信電話株式会社内 (72)発明者 牧野 昭二 東京都千代田区大手町二丁目3番1号 日 本電信電話株式会社内 Fターム(参考) 4C027 AA03 CC00 GG00 KK03 5D015 EE05 Continued front page (72) Inventor Akiko Araki 2-3-1, Otemachi, Chiyoda-ku, Tokyo Inside Telegraph and Telephone Corporation (72) Inventor Shoji Makino 2-3-1, Otemachi, Chiyoda-ku, Tokyo Inside Telegraph and Telephone Corporation F-term (reference) 4C027 AA03 CC00 GG00 KK03 5D015 EE05
Claims (4)
持している分離行列を用いて分離信号を生成する独立成
分分析方法において、 生成された分離信号に基づいて複素数の絶対値のみを変
更する活性化関数を用いて独立性に相当する値を計算す
る手順と、 独立性に相当する値と分離信号と前記分離行列から分離
行列の修正値を計算する手順と、 分離行列の修正値により分離信号の独立性が十分に高ま
るまで分離行列を修正する手順と、を有することを特徴
とする独立成分分析方法。1. An independent component analysis method for generating a separation signal from a plurality of linearly mixed complex signals using a separation matrix held in advance, wherein only the absolute value of a complex number is changed based on the generated separation signal. The procedure to calculate the value corresponding to independence using the activation function, the procedure to calculate the correction value of the separation matrix from the value corresponding to the independence, the separation signal and the separation matrix, and the correction value of the separation matrix And a step of modifying the separation matrix until the independence of the separated signals is sufficiently increased.
列保持部に保持している分離行列を用いて分離信号を生
成する分離信号計算部と、 生成された分離信号に基づいて複素数の絶対値のみを変
更する活性化関数を用いて独立性に相当する値を計算す
る活性化関数計算部と、 独立性に相当する値と分離信号と前記分離行列から分離
行列の修正値を計算する分離行列修正値計算部と、 分離行列の修正値により分離信号の独立性が十分に高ま
るまで分離行列を修正する分離行列修正手段と、を備え
たことを特徴とする独立成分分析装置。2. A separation signal calculation unit that generates a separation signal from a plurality of linearly mixed complex signals using a separation matrix held in a separation matrix holding unit, and an absolute value of a complex number based on the generated separation signals. An activation function calculation unit that calculates a value corresponding to independence using an activation function that changes only values, and a separation that calculates a correction value of the separation matrix from the value corresponding to independence, the separation signal, and the separation matrix An independent component analysis device comprising: a matrix correction value calculation unit; and a separation matrix correction unit that corrects the separation matrix until the independence of the separated signal is sufficiently increased by the correction value of the separation matrix.
持している分離行列を用いて分離信号を生成する独立成
分分析プログラムにおいて、 生成された分離信号に基づいて複素数の絶対値のみを変
更する活性化関数を用いて独立性に相当する値を計算す
るステップと、 独立性に相当する値と分離信号と前記分離行列から分離
行列の修正値を計算するステップと、 分離行列の修正値により分離信号の独立性が十分に高ま
るまで分離行列を修正するステップと、をコンピュータ
に実行させる独立成分分析プログラム。3. In an independent component analysis program for generating a separation signal from a plurality of linearly mixed complex signals using a separation matrix stored in advance, only the absolute value of a complex number is changed based on the generated separation signal. The step of calculating the value corresponding to the independence using the activation function, the step of calculating the correction value of the separation matrix from the value corresponding to the independence, the separation signal and the separation matrix, and the correction value of the separation matrix An independent component analysis program that causes a computer to modify the separation matrix until the independence of the separated signals is sufficiently enhanced.
持している分離行列を用いて分離信号を生成する独立成
分分析プログラムを記録した記録媒体において、 生成された分離信号に基づいて複素数の絶対値のみを変
更する活性化関数を用いて独立性に相当する値を計算す
るステップと、 独立性に相当する値と分離信号と前記分離行列から分離
行列の修正値を計算するステップと、 分離行列の修正値により分離信号の独立性が十分に高ま
るまで分離行列を修正するステップと、をコンピュータ
に実行させる独立成分分析プログラムを記録したコンピ
ュータ読み取り可能な記録媒体。4. A recording medium in which an independent component analysis program for generating a separation signal from a plurality of linearly mixed complex signals using a separation matrix stored in advance is recorded, and a complex number based on the generated separation signals is recorded. Calculating a value corresponding to independence using an activation function that changes only the absolute value; calculating a correction value of the separation matrix from the value corresponding to the independence, the separation signal, and the separation matrix; A computer-readable recording medium having an independent component analysis program for causing a computer to execute the step of modifying the separation matrix until the independence of the separated signal is sufficiently increased by the modification value of the matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2001279058A JP3887192B2 (en) | 2001-09-14 | 2001-09-14 | Independent component analysis method and apparatus, independent component analysis program, and recording medium recording the program |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2001279058A JP3887192B2 (en) | 2001-09-14 | 2001-09-14 | Independent component analysis method and apparatus, independent component analysis program, and recording medium recording the program |
Publications (2)
Publication Number | Publication Date |
---|---|
JP2003084793A true JP2003084793A (en) | 2003-03-19 |
JP3887192B2 JP3887192B2 (en) | 2007-02-28 |
Family
ID=19103318
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
JP2001279058A Expired - Fee Related JP3887192B2 (en) | 2001-09-14 | 2001-09-14 | Independent component analysis method and apparatus, independent component analysis program, and recording medium recording the program |
Country Status (1)
Country | Link |
---|---|
JP (1) | JP3887192B2 (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005260744A (en) * | 2004-03-12 | 2005-09-22 | Advanced Telecommunication Research Institute International | Method and apparatus for synchronizing phases of microphone reception signals in microphone array |
JP2006084974A (en) * | 2004-09-17 | 2006-03-30 | Nissan Motor Co Ltd | Sound input device |
EP1686831A2 (en) | 2005-01-26 | 2006-08-02 | Sony Corporation | Apparatus and method for separating audio signals |
JP2007044104A (en) * | 2005-08-08 | 2007-02-22 | Shimadzu Corp | Apparatus and method for signal analysis |
KR100745689B1 (en) | 2004-07-09 | 2007-08-03 | 한국전자통신연구원 | Apparatus and Method for separating audio objects from the combined audio stream |
JP2007526511A (en) * | 2004-02-26 | 2007-09-13 | ヒョン ナム,スン | Method and apparatus for blind separation of multipath multichannel mixed signals in the frequency domain |
WO2010005050A1 (en) * | 2008-07-11 | 2010-01-14 | 日本電気株式会社 | Signal analyzing device, signal control device, and method and program therefor |
JP2010117653A (en) * | 2008-11-14 | 2010-05-27 | Yamaha Corp | Signal processing device and program |
-
2001
- 2001-09-14 JP JP2001279058A patent/JP3887192B2/en not_active Expired - Fee Related
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007526511A (en) * | 2004-02-26 | 2007-09-13 | ヒョン ナム,スン | Method and apparatus for blind separation of multipath multichannel mixed signals in the frequency domain |
JP2005260744A (en) * | 2004-03-12 | 2005-09-22 | Advanced Telecommunication Research Institute International | Method and apparatus for synchronizing phases of microphone reception signals in microphone array |
KR100745689B1 (en) | 2004-07-09 | 2007-08-03 | 한국전자통신연구원 | Apparatus and Method for separating audio objects from the combined audio stream |
JP2006084974A (en) * | 2004-09-17 | 2006-03-30 | Nissan Motor Co Ltd | Sound input device |
JP4529611B2 (en) * | 2004-09-17 | 2010-08-25 | 日産自動車株式会社 | Voice input device |
EP1686831A2 (en) | 2005-01-26 | 2006-08-02 | Sony Corporation | Apparatus and method for separating audio signals |
JP2007044104A (en) * | 2005-08-08 | 2007-02-22 | Shimadzu Corp | Apparatus and method for signal analysis |
JP4561524B2 (en) * | 2005-08-08 | 2010-10-13 | 株式会社島津製作所 | Signal analysis apparatus and signal analysis method |
WO2010005050A1 (en) * | 2008-07-11 | 2010-01-14 | 日本電気株式会社 | Signal analyzing device, signal control device, and method and program therefor |
JPWO2010005050A1 (en) * | 2008-07-11 | 2012-01-05 | 日本電気株式会社 | Signal analysis apparatus, signal control apparatus and method, and program |
JP2010117653A (en) * | 2008-11-14 | 2010-05-27 | Yamaha Corp | Signal processing device and program |
Also Published As
Publication number | Publication date |
---|---|
JP3887192B2 (en) | 2007-02-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US8233636B2 (en) | Method, apparatus, and computer program for suppressing noise | |
KR102236471B1 (en) | A source localizer using a steering vector estimator based on an online complex Gaussian mixture model using recursive least squares | |
JP4403436B2 (en) | Signal separation device, signal separation method, and computer program | |
Ono | Fast stereo independent vector analysis and its implementation on mobile phone | |
WO2016100460A1 (en) | Systems and methods for source localization and separation | |
JP2012155339A (en) | Improvement in multisensor sound quality using sound state model | |
CN103180752B (en) | For resolving equipment and the method for the fuzziness arriving direction estimation | |
CN107180642A (en) | Audio signal bearing calibration, device and equipment | |
US10718742B2 (en) | Hypothesis-based estimation of source signals from mixtures | |
JP2020148909A (en) | Signal processor, signal processing method and program | |
WO2012070670A1 (en) | Signal processing device, signal processing method, and signal processing program | |
JP2003084793A (en) | Method, device, and program for analyzing independent component and recording medium with this program recorded thereon | |
Tesch et al. | Nonlinear spatial filtering in multichannel speech enhancement | |
Liao et al. | An effective low complexity binaural beamforming algorithm for hearing aids | |
JP6190373B2 (en) | Audio signal noise attenuation | |
JP6182169B2 (en) | Sound collecting apparatus, method and program thereof | |
JP5726709B2 (en) | Sound source separation device, sound source separation method and program | |
WO2020184210A1 (en) | Noise-spatial-covariance-matrix estimation device, noise-spatial-covariance-matrix estimation method, and program | |
Oyabu et al. | Linear multichannel blind source separation based on time-frequency mask obtained by harmonic/percussive sound separation | |
JP7270869B2 (en) | Information processing device, output method, and output program | |
US20240127841A1 (en) | Acoustic signal enhancement apparatus, method and program | |
CN113808606B (en) | Voice signal processing method and device | |
JP2018191255A (en) | Sound collecting device, method thereof, and program | |
JP2000097758A (en) | Sound-source signal estimating device | |
CN117953912A (en) | Voice signal processing method and related equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
A131 | Notification of reasons for refusal |
Free format text: JAPANESE INTERMEDIATE CODE: A131 Effective date: 20050621 |
|
A521 | Request for written amendment filed |
Free format text: JAPANESE INTERMEDIATE CODE: A523 Effective date: 20050822 |
|
RD03 | Notification of appointment of power of attorney |
Free format text: JAPANESE INTERMEDIATE CODE: A7423 Effective date: 20050822 |
|
A02 | Decision of refusal |
Free format text: JAPANESE INTERMEDIATE CODE: A02 Effective date: 20060627 |
|
A521 | Request for written amendment filed |
Free format text: JAPANESE INTERMEDIATE CODE: A523 Effective date: 20060824 |
|
A911 | Transfer to examiner for re-examination before appeal (zenchi) |
Free format text: JAPANESE INTERMEDIATE CODE: A911 Effective date: 20060911 |
|
TRDD | Decision of grant or rejection written | ||
A01 | Written decision to grant a patent or to grant a registration (utility model) |
Free format text: JAPANESE INTERMEDIATE CODE: A01 Effective date: 20061121 |
|
A61 | First payment of annual fees (during grant procedure) |
Free format text: JAPANESE INTERMEDIATE CODE: A61 Effective date: 20061124 |
|
FPAY | Renewal fee payment (event date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20101201 Year of fee payment: 4 |
|
FPAY | Renewal fee payment (event date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20101201 Year of fee payment: 4 |
|
FPAY | Renewal fee payment (event date is renewal date of database) |
Free format text: PAYMENT UNTIL: 20111201 Year of fee payment: 5 |
|
LAPS | Cancellation because of no payment of annual fees |