15th International Congress of Phonetic Sciences (ICPhS-15)
This paper proposes a target cost function for F0 based on polynomial regression for use in concatenative speech synthesis. Polynomial regression is used to express the time series of F0 continuously, and remove effects of microprosody. We conducted a perceptual experiment and confirmed that the proposed function provides a higher correlation with perceptual scores than does the conventionally used cost function.
Bibliographic reference. Fujii, Kei / Kashioka, Hideki / Campbell, Nick (2003): "Target cost of F0 based on polynomial regression in concatenative speech synthesis", In ICPhS-15, 2577-2580.