We report in detail the decoding strategy that we used for the past two Darpa Rich Transcription evaluations (RT'03 and RT'04) which is based on finite state automata (FSA). We discuss the format of the static decoding graphs, the particulars of our Viterbi implementation, the lattice generation and the likelihood evaluation. This paper is intended to familiarize the reader with some of the design issues encountered when building an FSA decoder. Experimental results are given on the EARS database (English conversational telephone speech) with emphasis on our faster than real-time system.
Cite as: Saon, G., Povey, D., Zweig, G. (2005) Anatomy of an extremely fast LVCSR decoder. Proc. Interspeech 2005, 549-552, doi: 10.21437/Interspeech.2005-338
@inproceedings{saon05_interspeech,
author={George Saon and Daniel Povey and Geoffrey Zweig},
title={{Anatomy of an extremely fast LVCSR decoder}},
year=2005,
booktitle={Proc. Interspeech 2005},
pages={549--552},
doi={10.21437/Interspeech.2005-338}
}