Title :
Discriminative Training Based on the Criterion of Least Phone Competing Tokens for Large Vocabulary Speech Recognition
Author :
Liu, Bo ; Jiang, Hui ; Zhou, Jian-lai ; Wang, Ren-Hua
Author_Institution :
Microsoft Res. Asia, Beijing, China
fDate :
March 18-23, 2005
Keywords :
gradient methods; hidden Markov models; learning (artificial intelligence); speech recognition; GPD algorithm; HMM; acoustic models; competing token set; discriminative training method; gradient normalization; large vocabulary speech recognition; least phone competing tokens; merging mechanism; off-line token collection procedure; relative error rate reduction; sigmoid-based objective function minimization; word lattices; Asia; Automatic speech recognition; Hidden Markov models; Lattices; Management training; Maximum likelihood decoding; Maximum likelihood estimation; Merging; Speech recognition; Vocabulary;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 2005. Proceedings. (ICASSP '05). IEEE International Conference on
Print_ISBN :
0-7803-8874-7
DOI :
10.1109/ICASSP.2005.1415064
