1987 Fiscal Year Final Research Report Summary
Analysis and Recognition of Speech by Optical Signal Processing Techniques
| Project/Area Number |
61580018
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
Informatics
|
|---|
| Research Institution | University of Tokyo |
|---|
Principal Investigator |
HIKICHI Takuo Faculty of Engineering, University of Tokyo, 工学部, 助手 (20010928)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
MORIKAWA Hiroyoshi Faculty of Engineering, University of Tokyo, 工学部, 助手 (40011217)
FUJISAKI Hirova Faculty of Engineering, University of Tokyo, 工学部, 教授 (80010776)
|
|---|
| Project Period (FY) |
1986 – 1987
|
| Keywords | Optical Signal Processing / Speech Signals / Spectrum Analysis / Template Matching / Speech Recognition / Windowing Plate / Non-linear Time Warping / 液晶パネル |
|---|
| Research Abstract |
The optical singal processing techniques have advantages over conventional techniques using a digital computer that a time-varying pattern of the speech spectrum can be obtained without shifting a window along the time axis, and the template matching between the spectral patterns can be conducted at the rate of the speed of light propagation. From this point of view, this project was planned with the goal of establishing the optical signal processing techniques for the speech signals. The main results obtained are as follows. The head investigator of the second fiscal year of this project was switched to Takuo Hikichi from Keikichi Hirose who was the head investigator of the first fiscal year, because Keikichi Hirose was on an long term official trip to abroad in 1987.
1. An optical signal processor was constructed for the spectrum analysis and for the template matching of speech. The spectrum analysis was counducted for speech samples of vowels and CV syllables. The coincidence between
… More
the results obtained by the optical method and those obtained by using the digital computer confirms the optical processor operates as expected. The template matching was also conducted for speech samples of connected vowels. It was shown that the degree of similarity between two templates could be obtained as a function of time.
2. Speech recognition is possible by the optical template matching between the spectral pattern of the unknown input speech and the standard template plates which are constructed optically by the spectrum analysis of the standard speech. Experiment on speech recognition was conducted for speech samples of five vowels uttered by a male speaker. The result indicates the validity of the recognition method based on the optical template matching.
3. It was experimentally confirmed that the time-varying pattern of the power spectrum can be nonlinearly time-warped by controlling the shape of the transparent strip of the windowing plate.
4. Non-linear time warping was performed in real time by replacing the windowing plate of the photographic film into that of a liquid crystal panel whose transparent strip shape can be controlled electrically using a personal computer. Spectrum analysis was conducted for the synthesized signals and the vowel speech in order to check the problem in the present performance of the liquid crystal panel.
As shown above, the initial goal fo this project was nearly accomplished. Further investigations are planned to develop a system which can analyze and recognize the speech continuously and in real time by replacing the photographic films of the speech signal plate and others into re-writable devices. Less
|
