Experience with a stack decoder-based HMM CSR and back-OFF N-gram language models
HLT '91 Proceedings of the workshop on Speech and Natural Language
A rapid match algorithm for continuous speech recognition
HLT '90 Proceedings of the workshop on Speech and Natural Language
1993 benchmark tests for the ARPA spoken language program
HLT '94 Proceedings of the workshop on Human Language Technology
| Hi-index | 0.00 |
The system described here is a large-vocabulary continuous-speech recognition (CSR) system with results obtained using the Wall Street Journal-based database [15]. The recognizer uses a stack decoder-based search strategy[1, 7, 14] with a left-to-right stochastic language model. This decoder has been shown to function effectively on 20K and 64K-word recognition of continuous speech. It operates left-to-right and can produce final textual output while continuing to accept additional input speech. Thus it need not wait for the end of the sentence and can be structured so that it can accept an unbounded length stream of input speech. The recognizer also features recognition-time adaptation to the user's voice. This system showed improvements of 42% for a 5K vocabulary and 35% for a 20K vocabulary compared to the November 92 evaluation test system.