Study on syllable durations in Standard Chinese

Zhang Sen (Waseda University - Tokyo)
K. Shirai (Waseda University - Tokyo)

Presently, most of Chinese Speech Synthesis Systems use a set of rules
to represent syllable duration features. Such rules are not precise
for every syllable and the naturalness of speech output of systems
based on such rules can not meet the requirements of most practical
applications. In this contribution we compared the difference of
Chinese syllables in different contexts and discussed the weakness of
rule-based approachs for modelling the durations of syllables, then we
provided a data-driven approach for accurately determining the
syllable durations. For the easy access to duration parameters we
constructed a speech database including syllable durations in typical
contexts. The synthesis quality of Chinese Speech Synthesis Systems
can be improved significantly based on this speech database.