Second ESCA/IEEE Workshop on Speech Synthesis

September 12-15, 1994
Mohonk Mountain House, New Paltz, NY, USA

Text-to-Speech Synthesis With Dynamic Control of Source Parameters

Luis C. Oliveira

INESC/IST, Lisbon, Portugal

This paper describes the study of some characteristics of the source parameters dynamics to derive a preliminary set of rules that were integrated in text-to-speech systems. An automated procedure estimated the source parameters of 534 seconds of voiced speech from a set of 300 English sentences spoken by a single female speaker. The results showed a strong inverse correlation between the vowel midpoint value of source parameters and the vowel duration. The same parameters tend to decrease on vowel onsets and to increase on vowels offsets. This seems to indicate a prosodic nature of this parameters requiring special treatment in concatenative-based tts systems that use source modification techniques, like PSOLA and multi-pulse.

Full Paper

Bibliographic reference.  Oliveira, Luis C. (1994): "Text-to-speech synthesis with dynamic control of source parameters", In SSW2-1994, 81-84.