荒川直哉
発話状況の情報を音声認識に用いるための
予備実験
Abstract:This report presents preliminary experiments for the application of
contextual information to speech recognition. Contextual information
here includes speaker roles, speech act types and particular expressions
in utterances. The author reports on the statistical identification of 1)
speech act types, 2) speech act type cue expressions, and 3) content
words (non-functional words). The ATR corpus of travel conversation
(SLDB, the Japanese data) was used for the experiments.