MOTIVATION

Need for a system configuration that is minimally variant to domain changes

Exploit higher levels of information in larger acoustic units such
as syllables, words and phrases

Syllable unit is more stable to artifacts such as insertions and deletions of phonetic units

Pronunciation variability in speech modeled implicitly through syllable or word units to a much greater degree than phones (example of OUR, AND)

Success of the basic syllable system at WS'97