7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Information Retrieval Based on Speech Recognition Results

Masatoshi Watanabe, Masahide Sugiyama

University of Aizu, Japan

This paper proposes a new information retrieval method based on speech recognition results. Nowadays, there is huge speech data like news speech in TV and Radio broadcasts, etc. However, it is not easy to retrieve necessary information from these data because extra informations for retrieval are not attached to most of these data. To solve the problems, in this paper a Fuzzy retrieval method based on speech recognition results is proposed. As speech data is converted into the character data and character-based query is retrieved on the character data, the query can be quickly retrieved. It is necessary to consider the speech recognition error at the retrieval. This paper proposes two techniques considering the speech recognition errors; the first technique is based on the phoneme error tendency in the speech recognition results and the second technique is based on the phoneme spectral distance. This paper shows the effectiveness of Fuzzy retrieval through evaluation experiments.

Full Paper

Bibliographic reference.  Watanabe, Masatoshi / Sugiyama, Masahide (2002): "Information retrieval based on speech recognition results", In ICSLP-2002, 1473-1476.