Discontinuities in fundamental frequency: When do they really matter in synthetic speech?

Abstract

The paper reports a perceptual experiment which aimed at relaxing the criteria for concatenation cost in the domain of fundamental frequency (f0) when concatenating diphones pertaining to voiced consonants. The results suggest that f0 discontinuities only matter in sonorants and only when they exceed 1 semitone.

Most importantly, the direction of f0 change should be taken into account.

Keywords

• voicing
• fundamental frequency
• concatenative synthesis
• concatenation cost