The goal of this work is to automatically determine which syllable an accent is associated with. This has many applications in both synthesis and recognition, and can also to be used as a device for investigating cues to stress in general.5 features are used to detect the stressed syllable. These are as follows: vowel quality, length and energy of syllable nucleus, distance from mid-vowel point to F0 peak, the start and the end of the accent. These features are used to train CART trees. In order to determine the difficulty of the task, I also ran simple detection algorithms such as choosing the nearest vowel before or after the accent peak.
The CART system detects 83% of the stressed syllables correctly, this is a considerable improvement on the simple methods.
To download this paper, please return to Proceedings of the 1998 Postgraduate Conference