This paper describes a unified architecture for integrating sub-lexical models with speech recognition, and a layered framework for contextdependent probabilistic hierarchical sub-lexical modelling using finite state transducers. Our major motivation for designing a unified architecture is to provide a framework such that probabilistic sublexical linguistic components can be integrated with other speech recognition components without sacrificing the flexibilities of their independent developments and configurations. We are also able to obtain a tightly coupled interface between recognizers and sub-lexical components. We present a view of using layered probabilistic models to augment contextfree grammars (CFGs). It captures context-dependent probabilistic information beyond the standard CFG formalism, and provides the flexibility of developing suitable probabilistic models independently for each sub-lexical layer. Experimental results show that such an approach can achieve comparable performance to pronunciation network approaches on in-vocabulary utterances, while being able to substantially reduce errors on utterances with previously unseen words.
Cite as: Mou, X., Seneff, S., Zue, V. (2001) Context-dependent probabilistic hierarchical sublexical modelling using finite state transducers. Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001), 451-454, doi: 10.21437/Eurospeech.2001-120
@inproceedings{mou01_eurospeech,
author={Xiaolong Mou and Stephanie Seneff and Victor Zue},
title={{Context-dependent probabilistic hierarchical sublexical modelling using finite state transducers}},
year=2001,
booktitle={Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001)},
pages={451--454},
doi={10.21437/Eurospeech.2001-120}
}