of Phonemes, Words and Phrases
Yasuo ARIKI
Department of Electronics and Informatics, Ryukoku University
Seta, Otsu, Japan
e-mail: ariki@rins.ryukoku.ac.jp
In Dialogue speech, we can hear phonemes, words and phrases selectively according to our understanding level instead of bottom-up understanding. From this view points, our goal is to construct simultaneous and level-interactive processing for spotting phonemes, words and phrases, and for predicting unclear portion from already analyzed one. We propose, in this paper, a fast algorithm for word spotting which can discriminate short and similar words by HMMs without time duration control. The algorithm employs simultaneous spotting of phonemes and words, along with duration check and known-unknown word check using phoneme information at the phoneme segment boundaries. ATR 25 spoken sentences were used for evaluation of spotting 57 words, and 77.8\% extraction rate was obtained with 128 times false alarm by 5 best extraction.
Keywords: level-interaction, simultaneous spotting, known words, phoneme information