PARAMETERIZATION AND SPEECH ANALYSIS USING INFORMATION OF VOCAL-TRACT SHAPE
| Project/Area Number |
60550271
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
電子機器工学
|
|---|
| Research Institution | HOKKAIDO UNIVERSITY |
|---|
Principal Investigator |
NOBUHIRO MIKI HOKKAIDO UNIVERSITY, RESEARCH INSTITUTE OF APPLIED ELECTRICITY, Associate Professor, 応用電気研究所, 助教授 (30002314)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
YOSHIKAZU MIYANAGA HOKKAIDO UNIVERSITY, RESEARCH INSTITUTE OF APPLIED ELECTRICITY, Research Associa, 応用電気研究所, 助手 (20166185)
|
|---|
| Project Period (FY) |
1985 – 1986
|
| Project Status |
Completed (Fiscal Year 1986)
|
| Budget Amount *help |
¥1,900,000 (Direct Cost: ¥1,900,000)
Fiscal Year 1986: ¥500,000 (Direct Cost: ¥500,000)
Fiscal Year 1985: ¥1,400,000 (Direct Cost: ¥1,400,000)
|
|---|
| Keywords | SPEECH PRODUCTION SYSTEM / VOCAL-TRACT MODEL / ARMA ANALYSIS / ARTICULATION ESTIMATION / 音声合成 |
|---|
| Research Abstract |
In this project we aimed at estimating the vocal-tract parameters for speech synthesis based on a physical model of speech production, and researched a estimation method of vocaltract area functions of vowels from the ARMA analysis model of speech waveforms.
Using a new evaluation of wide range frequency region and the model with the boundary condition at lips, we can estimate smooth area functions of vocal-tracts composed by multi-sections. The boundary condition at lips is a very important condition for speech synthesis from the digital model of speech production and estimation of the area function from speech analysis. We showed the results of measure of the lip impedance using a manikin and the acoustic pipe which simulates the lip shape in utterance, and obtained the characteristics of reflection at lips.
An accurate speech analysis method is needed for estimation of vocal-tract shapes from measured speech signals only. In this project we developed a new recursive ARMA analysis method, and showed that it is a accurate analysis method. In this method a non-stationaly analysis method is introduced, the tracking ability of spectral estimation is excellent for the speech with sudden spectral change such as some consonants.
When the spectral analysis is performed for real speechs, the estimation error is caused by the LPF for A/D conversion. In this project we developed the ARMA analysis method which can compensate the LPF characteristics. In our experiment for analysis of real consonants, accurate analysis is shown.
|
Report
(1 results)
Research Products
(11 results)
