| Presentation | 2007/12/13 Acoustic model adaptation based on a macroscopic time evolution system : A study on the number of adaptation utterances in an incremental update of an acoustic model Shinji WATANABE, Atsushi NAKAMURA, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Recently, we proposed an incremental adaptation method based on a macroscopic time evolution system, which models the dynamics of posterior distributions of acoustic model parameters to adjust time-variant acoustic characteristics during utterances. This paper examines recognition performance of the proposed incremental adaptation in terms of the length of incremental adaptation unit, which is critical for practical use in speech recognition. A preliminary experiment with the corpus of spontaneous Japanese reveals that acoustic characteristics are temporally changing even along the same lecture. Based on this experiment, we compare the relationship between batch and incremental adaptation methods, which can respectively mitigate time-invariant and time-variant mismatches of acoustic characteristics, in terms of the recognition performance of the lectures Finally, we discuss the practical trade-off between online property and estimation accuracy in terms of the length of incremental adaptation unit experimentally. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | speech recognition / acoustic model / incremental adaptation / macroscopic time evolution system / number of utterances in incremetal update |
| Paper # | NLC2007-67,SP2007-130 |
| Date of Issue |
| Conference Information | |
| Committee | SP |
|---|---|
| Conference Date | 2007/12/13(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) | Acoustic model adaptation based on a macroscopic time evolution system : A study on the number of adaptation utterances in an incremental update of an acoustic model |
| Sub Title (in English) | |
| Keyword(1) | speech recognition |
| Keyword(2) | acoustic model |
| Keyword(3) | incremental adaptation |
| Keyword(4) | macroscopic time evolution system |
| Keyword(5) | number of utterances in incremetal update |
| 1st Author's Name | Shinji WATANABE |
| 1st Author's Affiliation | NTT Communication Science Laboratory, NTT Coporation() |
| 2nd Author's Name | Atsushi NAKAMURA |
| 2nd Author's Affiliation | NTT Communication Science Laboratory, NTT Coporation |
| Date | 2007/12/13 |
| Paper # | NLC2007-67,SP2007-130 |
| Volume (vol) | vol.107 |
| Number (no) | 406 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |