Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

A Quantitative Description of German Prosody Offering Symbolic Labels as a By-Product

Hansjörg Mixdorff (1), Hiroya Fujisaki (2)

(1) Dresden University of Technology, Germany
(2) Science University of Tokyo, Japan

The prosodic quality of a text-to-speech system is important for the intellegibility and perceived naturalness of synthetic speech. In earlier works the author developed a linguistically motivated model of German intonation based on the quantitative Fujisaki model of the production process of F0. The current paper compares results yielded by automatic Fujisaki modeling with a GToBI-style anotation. On the accent level, a good correlation between tone labels and accent commands can be observed. On the phrase level, most level 3 and 4 break index boundaries are aligned with phrase commands whereas lower level boundaries are presumably marked with durational cues. Subsequently a regression model of syllable duration is introduced which permits to decompose the measured duration contour into an extrinsic and an intrinsic component.


Full Paper

Bibliographic reference.  Mixdorff, Hansjörg / Fujisaki, Hiroya (2000): "A quantitative description of German prosody offering symbolic labels as a by-product", In ICSLP-2000, vol.2, 98-101.