Interspeech'2005 - Eurospeech
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.
Bibliographic reference. Lin, Shiuan-Sung / Yvon, François (2005): "Discriminative training of finite state decoding graphs", In INTERSPEECH-2005, 733-736.