F0 stylization and intonation modelling for Standard Yorùbá Text-to-speech application

This technical report documents experiment into stylization of the f0 curve on Standard Yorùbá (SY ) syllables as well as a technique for intonation modelling. A number of interpolation polynomials were evaluated using root mean square error and mean opinion score techniques. The stylisation experiment resulted in the selection of a 3 degree polynomial for modelling the f0 curves on Yorùbá syllables. Thereafter, a symbolic representation scheme for abstracting the f0 curves in terms of peak and valley was developed. A modelling technique based on a relational tree (RT ) approach was then designed for intonation generation. RT is a sophisticated data structure which provides a multi-dimensional description of a waveform. The algorithm devised for generating intonation pattern is based on the intuitive notion of waveform as a sequence of peaks and valleys. The applicability of the technique is demonstrated by showing its application to modelling Yorùbá statement sentence intonation.