Top-down controlled and optimal syllable-template chaining algorithm for large vocabulary speech recognition

The vowel-consonant-vowel (VCV) template-matching is useful for large vocabulary speech recognition and continuous speech recognition from the point of view of the allophonic variation in phonemes due to co-articulation, the segmentation ambiguity, and the size of vocabulary. In this paper a new matching algorithm for a large vocabulary speech recognition is proposed, and experimental result and some discussion are given. This algorithm has the advantage of moderating an intensive and time-consuming task for template-training by reducing the number of templates to about 40% of the all VCV combinations. Further, by top-down controlled matching, this algorithm has the advantage of reducing computational requirement, which is required to avoid the effect of co-articulation and the segmentation ambiguity.