Presentation | 2010-12-20 Robust Acoustic Modeling Using MLLR Transformation-based Speech Feature Generation Arata ITOH, Sunao HARA, Norihide KITAOKA, Kazuya TAKEDA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We propose a novel acoustic model training method based on the new acoustic feature generation. Recently, the speaker adaptation method, such as MLLR and MAP, are widely used. However, all speaker adaptation methods need adaptation data. On the contrary, our method makes speaker-independent acoustic models that cover not only known but also unknown speakers. We focused on MLLR transformation matrix. Our method is a kind of generative training which generates new acoustic features by inverse transformation of MLLR transformation matrix and uses generated features to train acoustic models. We obtain MLLR transformation matrices from a limited number of existing speakers. Then we extract the bases of the MLLR transformation matrices using PCA and express MLLR transformation matrix by linear combination of bases. The probability distribution of the weight parameters to express the MLLR transformation matrices for the existing speakers are estimated. Finally we generate pseudo-speaker MLLR transformation by sampling the weight parameters from the distribution and apply the inverse of the transformation to the normalized existing speaker features to generate the pseudo-speakers' features. Using these features, we train the acoustic models. Evaluation results show that the acoustic models trained by our method are robust for unknown speakers. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Speech Recognition / Acoustic Model / MLLR Transformation Matrix / Generative Training |
Paper # | NLC2010-19,SP2010-92 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 2010/12/13(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 | 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 Modeling Using MLLR Transformation-based Speech Feature Generation |
Sub Title (in English) | |
Keyword(1) | Speech Recognition |
Keyword(2) | Acoustic Model |
Keyword(3) | MLLR Transformation Matrix |
Keyword(4) | Generative Training |
1st Author's Name | Arata ITOH |
1st Author's Affiliation | Graduate School of Information Science, Nagoya University() |
2nd Author's Name | Sunao HARA |
2nd Author's Affiliation | Graduate School of Information Science, Nagoya University |
3rd Author's Name | Norihide KITAOKA |
3rd Author's Affiliation | Graduate School of Information Science, Nagoya University |
4th Author's Name | Kazuya TAKEDA |
4th Author's Affiliation | Graduate School of Information Science, Nagoya University |
Date | 2010-12-20 |
Paper # | NLC2010-19,SP2010-92 |
Volume (vol) | vol.110 |
Number (no) | 356 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |