Voice Quality: Functions, Analysis and Synthesis August 27-29, 2003 Geneva, Switzerland

Principal Components of Glottal Waveforms: Towards Parameterisation and Manipulation of Laryngeal Voice Quality

Parham Mokhtari (1), Hartmut R. Pfitzinger (1,2), Carlos Toshinori Ishi (1)

(1) JST/CREST at ATR, Kyoto, Japan
(2) Department of Phonetics and Speech Communication, University of Munich, Germany

We propose a more holistic approach to modelling of the glottal volume-velocita waveform and its derivative. Principal components analysis (PCA) is applied to single-cycle glottal waveforms measured in Laver's (1980) recorded speech data representing several phonation types. The first four principal component lend themselves to qualitative interpretations, and account for over 90% of the total variance. Decision trees generated using those components yield around 64% accuracy in classifying the 13 voice qualities analysed. This approach is capable of modelling a wide variety of voice qualities, and should therefore be useful in speech synthesis.

Full Paper

Bibliographic reference.  Mokhtari, Parham / Pfitzinger, Hartmut R. / Ishi, Carlos Toshinori (2003): "Principal components of glottal waveforms: towards parameterisation and manipulation of laryngeal voice quality", In VOQUAL'03, 133-138.