EUROSPEECH 2001 Scandinavia
7th European Conference on Speech Communication and Technology

Aalborg, Denmark
September 3-7, 2001


Automatic Prosody Generation - a Model for Hungarian

Gábor Olaszy (1), Géza Németh (2), Péter Olaszi (2)

(1) Hungarian Academy of Sciences, Hungary
(2) Budapest University of Technology and Economics, Hungary

In our model a complex function set is described for the three prosody components of read speech. Each of them is modelled separately by a three-step procedure. A new method, based on indirect determination of specific sound durations was developed. Final duration values are calculated from the specific durations in two further steps. F0 changes are also modelled by three levels, starting with rules on sentence level, followed by the word and syllable level, and completed by the micro intonation level. Another three level model serves the intensity structure, i.e. rules applied on sounds, on words and on the complete sentence. The three component models have influence on each other during prosody generation. Cross effects among them are also mentioned. The model can be applied in speech research and in applications (synthesis and recognition). It was tested for Hungarian. Keywords: prosody generation, three-level model, specific sound durations, word-level duration map

Full Paper

Bibliographic reference.  Olaszy, Gábor / Németh, Géza / Olaszi, Péter (2001): "Automatic prosody generation - a model for hungarian", In EUROSPEECH-2001, 525-528.