15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Phrase Time Structure Modeling for Speech Synthesis Purposes

Grażyna Demenko

Adam Mickiewicz University, Poland

The paper presents a model of the intonational - rhythmic phrase structure in the Polish language as well as the premises for speech sound duration analysis to be used in text-to speech synthesis. The model of the prosodic phrase has been tested on the acoustic and perceptual analysis of the 50 syntactically and semantically diversified utterances produced by 40 native Polish speakers. The results showed that tempo changes within a phrase and the locus for a change in tempo is the focus. The statistical analysis showed the importance of the separate rhythm modeling in the prenuclear and nuclear part of the phrase. The results showed the possibility of timing modeling with the 76-87% correctness depending on the complexity of the phrase. The phrase time structure modeling will be tested in the actually build synthesizer of Polish speech.

Full Paper

Bibliographic reference.  Demenko, Grażyna (2003): "Phrase time structure modeling for speech synthesis purposes", In ICPhS-15, 2441-2444.