In this paper we introduce a Phone Vector Discrete HMM (PVDHMM) that decodes a phone recognizer's output. The proposed PVDHMM treats a phone recognizer as a vector quantizer whose codebook size is equal to the size of its phone set. To examine the proposed method we perform two experiments. First, the output of a phone recognizer is recognized by the PVDHMM, and its results are compared with those of a continuous speech recognizer (CSR). Second, to investigate its potential application in the field of open-vocabulary spoken document retrieval, a retrieval experiment through word spotting is carried out on the output of a phone recognizer, and its results are compared with those of retrieval through the phone-based vector space model.
Cite as: Kim, B.-W., Choi, D.-L., Um, Y., Lee, Y.-J. (2006) Phone vector DHMM to decode a phone recognizer's output. Proc. Interspeech 2006, paper 1903-Wed1BuP.11, doi: 10.21437/Interspeech.2006-445
@inproceedings{kim06d_interspeech,
author={Bong-Wan Kim and Dae-Lim Choi and Yongnam Um and Yong-Ju Lee},
title={{Phone vector DHMM to decode a phone recognizer's output}},
year=2006,
booktitle={Proc. Interspeech 2006},
pages={paper 1903-Wed1BuP.11},
doi={10.21437/Interspeech.2006-445}
}