In order to solve the problem of the performance decrease when state-of-art automatic speech recognition (ASR) system facing accent speech, we propose the Pronunciation Variation Model (PVM). Two approaches are proposed to construct the PVM in this paper. 6.38% and 7.78% relative error rate reduction is achieved for Shanghai and Wuhan accent mandarin, respectively. The experiment on these two typical accent mandarin shows it is a possible way to deal with accent speech.
Cite as: Zhang, C., Wu, J., Xiao, X., Wang, Z. (2006) Pronunciation variation modeling for Mandarin with accent. Proc. Interspeech 2006, paper 1849-Mon3FoP.5, doi: 10.21437/Interspeech.2006-246
@inproceedings{zhang06d_interspeech,
author={Chi Zhang and Ji Wu and Xi Xiao and Zuoying Wang},
title={{Pronunciation variation modeling for Mandarin with accent}},
year=2006,
booktitle={Proc. Interspeech 2006},
pages={paper 1849-Mon3FoP.5},
doi={10.21437/Interspeech.2006-246}
}