This paper describes a Japanese spoken document retrieval system that uses voice input queries. We prepare two types of spoken queries: isolately spoken keywords and spontaneously spoken queries. To solve a mis-recognition problem of spoken queries, N-best hypotheses of transcripts of queries are used, and keyword candidates are selected from them by mutual information between recognized words. Using both an effective keyword selecting algorithm and spontaneously spoken queries instead of isolately spoken keywords, a re- trieval performance in F-measure improves to 73.4 from72.9, and using N-best hypotheses by recognizing queries, the performance also improves to 73.4 from 68.7.
Cite as: Nishizaki, H., Nakagawa, S. (2002) Comparing isolately spoken keywords with spontaneously spoken queries for Japanese spoken document retrieval. Proc. 7th International Conference on Spoken Language Processing (ICSLP 2002), 1505-1508, doi: 10.21437/ICSLP.2002-453
@inproceedings{nishizaki02_icslp,
author={Hiromitsu Nishizaki and Seiichi Nakagawa},
title={{Comparing isolately spoken keywords with spontaneously spoken queries for Japanese spoken document retrieval}},
year=2002,
booktitle={Proc. 7th International Conference on Spoken Language Processing (ICSLP 2002)},
pages={1505--1508},
doi={10.21437/ICSLP.2002-453}
}