Presentation | 1998/12/11 Robust acoustic models by mixing distributions with various precision Masaru Takano, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this report, new acoustic models made by mixing probabilistic distributions with various precision are proposed. The proposed models can prevent phonetic confusion among precise phonetic models caused by acoustic variations in spontaneous speech, while keeping the advantage of precise model that can deal with acoustic variations caused by phonetic environment. Distributions in the proposed models are taken from both precise cotext-dependent(CD)model and another rough model, such as context-independent model. By using distributions from rough model, a model can match acoustic features which don't fit the precise model because of variations due to spontaneity. Experiments on spontaneous speech for hotel reservation task indicated that some of the proposed models can reduce error rate with CD model by around 1/10. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | spontaneous speech / context-dependent model / context-independent model / robustness / continuous mixture density / precision / speech recognition |
Paper # | NLC98-51,SP98-115 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 1998/12/11(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) | Robust acoustic models by mixing distributions with various precision |
Sub Title (in English) | |
Keyword(1) | spontaneous speech |
Keyword(2) | context-dependent model |
Keyword(3) | context-independent model |
Keyword(4) | robustness |
Keyword(5) | continuous mixture density |
Keyword(6) | precision |
Keyword(7) | speech recognition |
1st Author's Name | Masaru Takano |
1st Author's Affiliation | ATR Interpreting Telecommunications Research Laboratories() |
Date | 1998/12/11 |
Paper # | NLC98-51,SP98-115 |
Volume (vol) | vol.98 |
Number (no) | 461 |
Page | pp.pp.- |
#Pages | 8 |
Date of Issue |