TY - GEN
T1 - A regularized discriminative training method of acoustic models derived by minimum relative entropy discrimination
AU - Kubo, Yotaro
AU - Watanabe, Shinji
AU - Nakamura, Atsushi
AU - Kobayashi, Tetsunori
N1 - Funding Information:
Acknowledgement The authors would like to thank the valuable discussions in the Lehrstuhl für Informatik 6, RWTH Aachen University. This study was partially supported by a Grant-in-Aid for JSPS Fellows (21·04190) from the Ministry of Education, Culture, Sports, Science and Technology, Japan.
Funding Information:
The authors would like to thank the valuable discussions in the Lehrstuhl für Informatik 6, RWTH Aachen University. This study was partially supported by a Grant-in-Aid for JSPS Fellows (21·04190) from the Ministry of Education, Culture, Sports, Science and Technology, Japan.
PY - 2010
Y1 - 2010
N2 - We present a realization method of the principle of minimum relative entropy discrimination (MRED) in order to derive a regularized discriminative training method. MRED is advantageous since it provides a Bayesian interpretations of the conventional discriminative training methods and regularization techniques. In order to realize MRED for speech recognition, we proposed an approximation method of MRED that strictly preserves the constraints used in MRED. Further, in order to practically perform MRED, an optimization method based on convex optimization and its solver based on the cutting plane algorithm are also proposed. The proposed methods were evaluated on continuous phoneme recognition tasks. We confirmed that the MRED-based training system outperformed conventional discriminative training methods in the experiments.
AB - We present a realization method of the principle of minimum relative entropy discrimination (MRED) in order to derive a regularized discriminative training method. MRED is advantageous since it provides a Bayesian interpretations of the conventional discriminative training methods and regularization techniques. In order to realize MRED for speech recognition, we proposed an approximation method of MRED that strictly preserves the constraints used in MRED. Further, in order to practically perform MRED, an optimization method based on convex optimization and its solver based on the cutting plane algorithm are also proposed. The proposed methods were evaluated on continuous phoneme recognition tasks. We confirmed that the MRED-based training system outperformed conventional discriminative training methods in the experiments.
KW - Discriminative training
KW - Optimization
KW - Speech recognition
UR - http://www.scopus.com/inward/record.url?scp=79959828521&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=79959828521&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:79959828521
T3 - Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010
SP - 2954
EP - 2957
BT - Proceedings of the 11th Annual Conference of the International Speech Communication Association, INTERSPEECH 2010
PB - International Speech Communication Association
ER -