Acoustical F0 Analysis of Continuous Cantonese Speech

This paper presents a preliminary study on acoustical analysis of fundamental frequency (F0) in continuous Cantonese speech. By understanding how the surface F0 contour is determined by many co-functioning and inter-playing linguistic or non-linguistic factors, our ultimate goal is to facilitate automatic F0 prediction for highly natural text-to-speech synthesis. A novel method of F0 normalization is proposed to effectively reduce the undesirable fluctuation of the speaker’s F0 range. Statistical analysis is performed on the normalized F0 contours. Specifically, our investigation is focused on: (1) F0 movement over intonation phrases; (2) tone contours in continuous speech; (3) effect of tonal context; and (4) co-articulated tone contours in disyllabic words.