Presentation | 2002/12/12 Automatic Construction of HMM Structures by the MDL Criterion on the SSS Algorithm Takatoshi JITSUHJRO, Tomoko MATSUI, Satoshi NAKAMURA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We propose a new Successive State Splitting (SSS) algorithm based on the Minimum Description Length(MDL) criterion to design HMM topologies automatically. Many methods have been proposed to create tied-state structures of HMMs but they usually need empirical values as stop criteria, for example, the number of states. The SSS algorithm is a mechanism for creating both temporal and contextual variations based on the Maximum Likelihood (ML) criterion. However, it also needs to empirically predetermine the control parameters, including the total number of states as stop criteria. We introduce the MDL criterion to the ML-SSS algorithm so that it can automatically create proper topologies without any heuristic stop criteria. Experimental results using the ATR travel arrangement dialogs show that our extended algorithm can automatically stop splitting and obtain more appropriate HMM topologies than the original one. We also try to extend the MDL-SSS algorithm by using the phonetic decision tree clustering for contextual splitting. The method using a combination of the phonetic decision tree clustering and the data-driven clustering could automatically obtain almost equivalent performances. The optimal two coefficients of the criterion were dependent on each method. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | speech recognition / acoustic model / topology training / SSS algorithm / MDL criterion |
Paper # | NLC2002-50 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 2002/12/12(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 | Natural Language Understanding and Models of Communication (NLC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Automatic Construction of HMM Structures by the MDL Criterion on the SSS Algorithm |
Sub Title (in English) | |
Keyword(1) | speech recognition |
Keyword(2) | acoustic model |
Keyword(3) | topology training |
Keyword(4) | SSS algorithm |
Keyword(5) | MDL criterion |
1st Author's Name | Takatoshi JITSUHJRO |
1st Author's Affiliation | ATR Spoken Language Translation Research Laboratories() |
2nd Author's Name | Tomoko MATSUI |
2nd Author's Affiliation | ATR Spoken Language Translation Research Laboratories |
3rd Author's Name | Satoshi NAKAMURA |
3rd Author's Affiliation | ATR Spoken Language Translation Research Laboratories |
Date | 2002/12/12 |
Paper # | NLC2002-50 |
Volume (vol) | vol.102 |
Number (no) | 527 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |