Presentation | 1999/8/6 Word Recognition using Partly-Hidden Markov Model Junko Furuyama, Tetsunori Kobayashi, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | HMM is the most popular probabilistic model in current speech recognition systems. However it can deal with only piecewise stationary process. We solved this problem by introducing the modified second order Markov Model, Partly-Hidden Markovmodel, in which the first state is hidden and the second one is observable. In HMM, state transition is dependent only on the previous state and observation is dependent only current state. While, in PHMM, state is defined by the previous state and previous observation, and observation is also dependent on current state and preious observation. Word recognition test show that the error rate was reduced by 69.O%(without delta) or 62.5%(with delta) compared with HMM. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | acoustic model / HMM / PHMM / word recognition |
Paper # | SP99-64 |
Date of Issue |
Conference Information | |
Committee | SP |
---|---|
Conference Date | 1999/8/6(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) | Word Recognition using Partly-Hidden Markov Model |
Sub Title (in English) | |
Keyword(1) | acoustic model |
Keyword(2) | HMM |
Keyword(3) | PHMM |
Keyword(4) | word recognition |
1st Author's Name | Junko Furuyama |
1st Author's Affiliation | School of Science and Engineering,Waseda University() |
2nd Author's Name | Tetsunori Kobayashi |
2nd Author's Affiliation | School of Science and Engineering,Waseda University |
Date | 1999/8/6 |
Paper # | SP99-64 |
Volume (vol) | vol.99 |
Number (no) | 256 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |