Automatic Speech Recognition systems integrate three main knowledge sources: acoustic models, pronunciation dictionary and language models. In contrast to common practices, where each source is optimized independently, then combined in a finite-state search space, we investigate here a training procedure which attempts to adjust (some of) the parameters after, rather than before, combination. To this end, we adapted a discriminative training procedure originally devised for language models to the more general case of arbitrary finite-state graphs. Preliminary experiments performed on a simple name recognition task demonstrate the potential of this approach and suggest possible improvements.
Cite as: Lin, S.-S., Yvon, F. (2005) Discriminative training of finite state decoding graphs. Proc. Interspeech 2005, 733-736, doi: 10.21437/Interspeech.2005-13
@inproceedings{lin05_interspeech,
author={Shiuan-Sung Lin and François Yvon},
title={{Discriminative training of finite state decoding graphs}},
year=2005,
booktitle={Proc. Interspeech 2005},
pages={733--736},
doi={10.21437/Interspeech.2005-13}
}