EUROSPEECH 2001 Scandinavia
The current study investigates the relationship between perceived syllable prominence and the F0 contour as described by a linguistically motivated model of German intonation based on the Fujisaki formula. A subcorpus of the Bonn Prosodic Database was analyzed using the F0 model, and normalized log syllable durations calculated. Analysis shows that, for accented syllables, prominences strongly correlate with the amplitude Aa of accent commands underlying the F0 movements in these syllables, whereas comparable F0 movements in unaccented syllables have little effect on prominence. The influence of Aa versus syllable duration on prominence is stronger for higher prominence levels. The fact that the F0 movement does not necessarily take place in the accented syllable proper, indicates that prominence judgment is partly guided by linguistic considerations. The results also show that F0 modeling in TTS needs to be especially accurate in accented syllables which supports the main rationale of the F0 model.
Bibliographic reference. Mixdorff, Hansjörg / Widera, Christina (2001): "Perceived prominence in terms of a linguistically motivated quantitative intonation model", In EUROSPEECH-2001, 403-406.