Demo of the F0-Contour Generation Program

The Theory in a Nutshell

The intonation of a phrase is mainly achieved by variations of the pitch (the perceptional correlate of the fundamental frequency F0 ). Within a speech utterance the range in which the pitch varies can be determined by two lines, the topline and the baseline. Both lines have a slight decrease over the full span of a phrase.

Certain words in a speech utterance are accented by a rise or a fall in the F0-curve. In the model implemented in our F0-contour generation program (the Tone-Sequence-Model by J. Pierrhumbert / C. Féry) the notation is H*L for a fall and L*H for a rise. Hence, the fall is realized by a high tone (H*) on a word-stressed syllable and a low tone (L ) on the following syllable. Phrases end either on a high (notation: H%), a low (L%) or on a default pitch tone (%).

For a more detailed description of the theory and the program see [1]. The Tone Sequence Model for German is described here. The F0 generation program, however, is not restricted to a particular intonation model. It could be easily used with any other model.

The Demo

For the German phrase "Mein Bruder mag moderne Musik" (engl. "My brother likes modern music.") you can apply a rising (L*H) or a falling (H*L) pitch accent for two specific words. The first possible accent is on "Bruder" and the second on "moderne". The phrase will end either on a high (H%) or on a default boundary tone (%). The result will be a the appropriate F0-contour and a sound file in which we imposed this F0-contour on a recording (PSOLA-resynthesis).
Mein Bruder mag moderne Musik .

Reference:
[1] Gregor Moehler & Grzegorz Dogil. Test Environment for the Two Level Model of Germanic Prominence. Proceedings EUROSPEECH, Session WEam2C.4, Madrid 1995.


27. September 1996
IMS Stuttgart / WWW@IMS.Uni-Stuttgart.DE / Fri Jun 19 17:40:27 1998 (gregor)