International Symposium on Tonal Aspects of Languages
With Emphasis on Tone Languages

Beijing, China
March 28-31, 2004

Visual Cues in Thai Tone Recognition

Hansjörg Mixdorff (1), Patavee Charnvivit (2)

(1) Faculty of Computer Science, TFH Berlin University of Applied Sciences, Germany
(2) CRSLP, Chulalongkorn University, Bangkok, Thailand

The current paper presents preliminary experiments on the exploitation of visual cues in the perception of Thai tones. The lower half of a female speaker's face was recorded on digital video as she uttered a set of monosyllabic tokens covering the five different tones of Thai. The trajectories of 12 light points on the speaker's face were traced frame-by-frame and converted into velocity tracks which were then time-aligned with the speech signal. In parallel, a perception experiment was conducted in which the silent movies were shown to native speakers of Thai who had to decide which tone they perceived. Results, so far, are inconclusive. Whereas the identification results by the speaker herself being the subject are clearly above chance level, results from the other subjects are not and suggest that visual cues only are not sufficient for reliable identification. Since, however, subjects in Thailand were presented a compressed movie at a reduced resolution this circumstance might have also flawed their results.

Full Paper

Bibliographic reference.  Mixdorff, Hansjörg / Charnvivit, Patavee (2004): "Visual cues in Thai tone recognition", In TAL-2004, 143-146.