ISCA Archive SpeechProsody 2010
ISCA Archive SpeechProsody 2010

Analysis of prosodic classes using voice source measurements

Ran Han, Jeung-Yoon Choi

In this study, we use voice source measurements over single-and multi-syllable regions to find acoustic cues for broad prosodic classes. Voice source measurements include duration, fundamental frequency, harmonic amplitudes, spectral tilt, and speech amplitude measurements. Using the Boston University Radio Speech Corpus, significant measurements are found from ANOVA tests, and the distributions of those measurements are observed. Finally, using all measurements, broad class prosody detection is carried out. The best detection rates for 4 broad accent classes and 5 broad boundary classes are 46.2% and 51.8%. From these results, it can be seen that voice source measurements and multi-syllable measurements may be useful in detecting detailed prosodic classes.

Cite as: Han, R., Choi, J.-Y. (2010) Analysis of prosodic classes using voice source measurements. Proc. Speech Prosody 2010, paper 399

  author={Ran Han and Jeung-Yoon Choi},
  title={{Analysis of prosodic classes using voice source measurements}},
  booktitle={Proc. Speech Prosody 2010},
  pages={paper 399}