| Presentation | 2008/5/22 A Study on Robust Speech Recognition against Unknown Reverberation Using Single Speech Model Trained under Multiple Reverberant Environments Kenta NISHIKI, Shinji WATANABE, Takuya NISHIMOTO, Nobutaka ONO, Shigeki SAGAYAMA, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In reverberant environment, speech recognition accuracy is seriously degraded. An effective method is to train acoustic models using reverberant speech which are generated from clean speech data and reverberant impulse response. It is not clear, however, what kind of impulse response is suitable for environmental variations. In this research, we show the range of reverberant time that changes speech recognition accuracy drastically through evaluation with CENSREC-4. Then we point out that the range of reverberant time which is effective for model training is shorter than that of test conditions. We also show the multi-condition training is effective for speech recognition under unknown environments using speech data of the range of reverberant time we showed. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | reverberant environment / speech recognition / model construction / artificial reverberance / impulse response / reverberation time |
| Paper # | SP2008-8,WIT2008-8 |
| Date of Issue |
| Conference Information | |
| Committee | SP |
|---|---|
| Conference Date | 2008/5/22(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Speech (SP) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A Study on Robust Speech Recognition against Unknown Reverberation Using Single Speech Model Trained under Multiple Reverberant Environments |
| Sub Title (in English) | |
| Keyword(1) | reverberant environment |
| Keyword(2) | speech recognition |
| Keyword(3) | model construction |
| Keyword(4) | artificial reverberance |
| Keyword(5) | impulse response |
| Keyword(6) | reverberation time |
| 1st Author's Name | Kenta NISHIKI |
| 1st Author's Affiliation | Department of Information Physics and Computing, University of Tokyo() |
| 2nd Author's Name | Shinji WATANABE |
| 2nd Author's Affiliation | NTT Communication Science Laboratories |
| 3rd Author's Name | Takuya NISHIMOTO |
| 3rd Author's Affiliation | Department of Information Physics and Computing, University of Tokyo |
| 4th Author's Name | Nobutaka ONO |
| 4th Author's Affiliation | Department of Information Physics and Computing, University of Tokyo |
| 5th Author's Name | Shigeki SAGAYAMA |
| 5th Author's Affiliation | Department of Information Physics and Computing, University of Tokyo |
| Date | 2008/5/22 |
| Paper # | SP2008-8,WIT2008-8 |
| Volume (vol) | vol.108 |
| Number (no) | 66 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |