Third International Conference on Spoken Language Processing (ICSLP 94)
In word spotting on continuous speech, powerful discrimination of known words from unknown words is required. In this paper, two word spotting techniques are proposed which discriminate known words using phoneme information obtained through Viterbi decoding. One is to select known words at every frame and extract a sequence of known words and phonemes by backtrace. The other is to locally extract known words without backtrace.
Bibliographic reference. Ariki, Y. / Kawamura, T. (1994): "Simultaneous spotting of phonemes and words in continuous speech", In ICSLP-1994, 2191-2194.