1992 Fiscal Year Final Research Report Summary
Study on cocktail-party effect processor
| Project/Area Number |
02650272
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
情報工学
|
|---|
| Research Institution | Kyushu Institute of Design |
|---|
Principal Investigator |
FUKUDOME Kimitoshi Kyushu Institute of Design, Faculty of Design, Associate Professor, 芸術工学部, 助教授 (90038977)
|
|---|
| Project Period (FY) |
1990 – 1992
|
| Keywords | Cocktail Party Processor / Binaural Signal Processing / Direction of Arrival Estimation / Spectral Estimation / Diffractive Information Dictionary / Narrow Band Frequency Analysis / Wiener Filter |
|---|
| Research Abstract |
Purpose of this research project is to present a solution for extracting the desired sound signal in high fidelity from an interfering sound field generated by multiple sound sources. Based on two channel input signals, it has been studied what would be the best acoustical information for realizing a cocktail-party effect processor. Results obtained are summarized as follows: 1. Two-element sphere baffled microphone is adopted to receive sounds in the sound field. 2. Diffractive information dictionary is made as a database for the binaural input signals. 3. In simulation research for the free sound field, the system can separately extract multiple sound source signals more than two, i.e. the number of sensors of the two-element sphere baffled microphone. 4. The beset effective acoustical information from the binaural input signals is an existence of pitch frequency. The system detects it and uses it as the clue for estimating the direction arrival. 5. The interferer canceling method is effective on Wiener filter tunable to the direction of arrival with the diffractive information dictionary. Effects of the Wiener filter length and the time-varing pitch frequency on the system performance have been evaluated. 6. The canceling method cancels the components of the desired source signal which correlates with interferes. To recover these, a post-processing method has been presented. 7. In simulation research for the sound field where reflections exist, a pri-processing for estimating the direction of arrival of the direct sound was investigated. It is to be constructed based on a pattern learning system with a supervisor, using the frequency range information from the narrow band analysis of binaural signal inputs.
|
Research Products
(12 results)
