HLT '90 Proceedings of the workshop on Speech and Natural Language
Hidden Markov Models for Speech Recognition
Hidden Markov Models for Speech Recognition
Recent progress on the VOYAGER system
HLT '90 Proceedings of the workshop on Speech and Natural Language
New uses for the N-best sentence hypotheses within the BYBLOS speech recognition system
ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
Hi-index | 0.00 |
In most speech recognition systems, acoustic and language models are usually constructed separately, where language models are derived from a large text corpus without considering confusable acoustic data, and acoustic models are optimized without considering language model discrimination capacity. One example is that unbalanced acoustic and language models have to be combined with an ad-hoc constant language weight that is tuned from development data. This paper describes a unified stochastic engine that jointly optimizes both acoustic and language models. We develop a general modeling framework. At present, we focus only on language weight optimization, which is a special case of acoustic-driven language modeling. We report preliminary experimental results for Wall Street Journal continuous 5OOO-word speaker-independent dictation, where the error rate is reduced from 7.3% to to 6.990 with the proposed method.