Analyzing Thai Tone Distribution through Functional Data Analysis

Hong Zhang


This paper reports an analysis of tonal properties of Thai using a method based on functional data analysis on a large collection of TIMIT-like corpus. Both density estimation pooled across syllable-wise F0 contours and Functional Principle Component Analysis (FPCA) were applied. The results suggest that the simple two dimensional representation of tones: pitch target height and contour slope, is not able to capture context dependent variations of tonal contour within and across tone categories. In addition, the shape and timing of pitch target are also crucial both in differentiating tonal categories and explaining variations associated with syllable structure. The third and fourth dimension of the functional basis have been shown to be able to represent these higher-order properties. Thus FPCA can provide a compact yet interpretable low dimension representation for the tonal property of Thai. These findings are also helpful for understanding tone distribution properties and coarticulation.


 DOI: 10.21437/Interspeech.2018-2115

Cite as: Zhang, H. (2018) Analyzing Thai Tone Distribution through Functional Data Analysis. Proc. Interspeech 2018, 2137-2141, DOI: 10.21437/Interspeech.2018-2115.


@inproceedings{Zhang2018,
  author={Hong Zhang},
  title={Analyzing Thai Tone Distribution through Functional Data Analysis},
  year=2018,
  booktitle={Proc. Interspeech 2018},
  pages={2137--2141},
  doi={10.21437/Interspeech.2018-2115},
  url={http://dx.doi.org/10.21437/Interspeech.2018-2115}
}