This paper presents the integration between the quantitative target approximation (qTA) model and the unsupervised clustering technique to study Thai tones. The qTA model simulates F0 production on the basis of articulation process. Parameters extracted from the F0 of Thai speech by analysis-and-synthesis method were further analyzed by K-means clustering. The number and form of pitch target were then identified by assessing the quality of clusters using silhouette scoring. The combination of qTA model and computational intelligent techniques make it possible to study pitch targets of tone function without prior hypothesis.
Bibliographic reference. Prom-on, Santitham (2008): "Pitch target analysis of Thai tones using quantitative target approximation model and unsupervised clustering", In INTERSPEECH-2008, 1116-1119.