Signal Separation for Multiple Sound Sources and Its Application to Speech Recognition System for Multiple Speakers
| Project/Area Number |
09450167
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
計測・制御工学
|
|---|
| Research Institution | Tokyo University of Agriculture and Technology |
|---|
Principal Investigator |
KOBATAKE Hidenori Tokyo Univ. of Agri. & Tech. Graduate School of Bio-Applications and system Engineering, Professor, 大学院・生物システム応用科学研究科, 教授 (80013720)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
HAHIHARA Yoshihiro Tokyo Univ. of Agri. & Tech. Faculty of Technology, Assistant, Professor, 工学部, 助手 (80293009)
|
|---|
| Project Period (FY) |
1997 – 1999
|
| Project Status |
Completed (Fiscal Year 1999)
|
| Budget Amount *help |
¥8,300,000 (Direct Cost: ¥8,300,000)
Fiscal Year 1999: ¥1,600,000 (Direct Cost: ¥1,600,000)
Fiscal Year 1998: ¥2,500,000 (Direct Cost: ¥2,500,000)
Fiscal Year 1997: ¥4,200,000 (Direct Cost: ¥4,200,000)
|
|---|
| Keywords | Signal separation / location of sound sources / Least mean-square method / inverse filter / generalized inverse matrix / multi-microphone system / 逐次推定 / モデルマッチング / 逆フィルター / マルケマイクロホンシステム |
|---|
| Research Abstract |
Blind source separation (BSS) is a useful method to separate signals radiated from multiple signal sources. It assumes, however, that each signal is mutually independent. In this research, we have developed a fundamental method to separate acoustic signals radiated from acoustic signal sources, in which each signal is not required to be mutually independent. The number of microphones necessary for this purpose must be larger than the number of signal sources. Typical number of microphones is N+1, where N is the number of signal sources.
The fundamental method is based on the least-mean squared error method. A new criterion function is proposed, by which the optimal solution can be stably obtained.
Relatively accurate initial estimates for signal source locations must be given for stable convergence of the least-mean squared error method. The initial estimator for each signal source location has been developed, which is based on the genetic algorithm. The delay time of the direct wave to each microphone is another important factor for the stable convergence of the least-mean squared method. We have developed a method to estimate the delay time of the direct wave. Experimental results showed that the proposed method works well if the power level of echoes is less than 30% of that of the direct wave.
The results of computer simulation showed that the proposed method works well under echo environment.
|
Report
(4 results)
Research Products
(12 results)
