| Yoshiaki Itoh, Jiro Kiyama and Ryuichi Oka TRC Information Integration Laboratory (1994) | |||||||||||||
Abstract | |||||||||||||
| Introduction Spontaneous speech includes many sentences that fall outside the task domain. Furthermore, the boundary between sentences is often unclear in spontaneous speech because of the likes of corrections, stammering or overlap with the next utterance. We previously developed a sentence spotting system that uses Vector ContinuousDynamic Programming (VCDP). This system works well for sentence spotting in spontaneous speech [1] because it is not required to consider sentence boundaries and utterances which fall outside the task domain. The previous system supported only "complete sentence " utterances. However, partial sentences that are intended to convey almost the same meaning as complete sentences, and which consist of parts of complete sentences often appear in spontaneous speech. We must be able to deal with such expressions to enable flexible recognition, even though such partial sentences are subject to a wide degree of variation. We propose a means of extending a s | |||||||||||||
Publication details | |||||||||||||
| |||||||||||||