ISCA Archive ICSLP 2002
ISCA Archive ICSLP 2002

On F0 trajectory optimization for very high-quality speech manipulation

Hideki Kawahara, Parham Zolfaghari, Alain de Cheveigné

An optimized fundamental frequency (F0) trajectory extraction method, which alleviates systematic F0 glitches at vowel-nasal boundaries and in the vicinity of consonants, is introduced. The proposed method employs minimum phase group delay compensation for apparent F0 modulations due to variations in their corresponding vocal tract transfer functions. This method can also be considered as an implementation of a generalized version of analysis by synthesis. Evaluation using EGG reference signals revealed that the proposed method reduces the systematic biases by 50%.