Perceptual Continuity and Naturalness of Expressive Strength in Singing Voice based on Speech Morphing

This paper experimentally shows the importance of perceptual continuity of the expressive strength in vocal timbre for natural change in vocal expression. In order to synthesize various and continuous expressive strengths with vocal timbre, we investigated gradually changing expressions by applying the STRAIGHT speech-morphing algorithm to singing voices. Here, a singing voice without expression is used as the base of morphing, and singing voices with three different expressions are used as the target. Through statistical analyses of perceptual evaluations, we confirmed that the proposed morphing algorithm provides perceptual continuity of vocal timbre. Our results showed the following: i) gradual strengths in absolute evaluations, and ii) a perceptually linear strength provided by the calculation of corrected intervals of the morph ratio by the inverse (reciprocal) function of an equation that approximates the perceptual strength. Finally, we concluded that applying continuity was highly effective for achieving perceptual naturalness, judging from the results showing that iii) our gradual transformation method can perform well for perceived naturalness. ∗1 ATR Intelligent Robotics and Communication Laboratories ∗2 Nagoya University ∗3 ATR Media Information Science Laboratories † presently also affiliated with ATR Media Information Science Laboratories ‡ presently with National Institute of Information and Communication Technology/ATR Cognitive Information Science Laboratories ‡‡ presently with ATR Knowledge Science Laboratories