We consider the problem of predicting the surface pronunciations of a word in conversational speech, using a feature-based model of pronunciation variation. We build context-dependent decision trees for both phone-based and feature-based models, and compare their perplexities on conversational data from the Switchboard Transcription Project. We find that feature-based decision trees using featur e bundles based on articulatory phonology outperform phone-based decision trees, and are much more r obust to reductions in training data. We also analyze the usefulness of various context variables.
Cite as: Bowman, S., Livescu, K. (2010) Modeling pronunciation variation with context-dependent articulatory feature decision trees. Proc. Interspeech 2010, 326-329, doi: 10.21437/Interspeech.2010-122
@inproceedings{bowman10_interspeech,
author={Sam Bowman and Karen Livescu},
title={{Modeling pronunciation variation with context-dependent articulatory feature decision trees}},
year=2010,
booktitle={Proc. Interspeech 2010},
pages={326--329},
doi={10.21437/Interspeech.2010-122}
}