San Duanmu will present in Phondi today. The title and abstract of his presentation are given below.

Automatic Transcription of Tones

In this talk, I describe a project of ours which aims to develop a system to automatically transcribe phonetic tones. I first explain the importance of the project. Then I describe some well-known difficulties, such as the discrepancy between visual contours of F0 and perceived tone types. Next I describe our approach, which distinguishes three levels of transcription: (i) the underlying tones, (ii) the perceived phonetic tones (‘allo-tones’), and (iii) measured F0 contours. Our main task is to relate (iii) to (ii) by making appropriate adjustments, based on the context of a tone (e.g. types of surrounding consonants, stress level, and local F0 level). The relation between (i) and (ii) will reveal allo-tonic rules of a language, which will be richer than previously described.