Speech Synthesis Method for Flexible Voice Quality Control

点击翻译按钮获取中文摘要

Flexible voice quality control in speech synthesis includes not only that of such qualities as whisper, breathy and tense but also that of talker individuality resulting from physiological differences in the speech organ. Major aim of this research project is to establish a base to realize such control in speech synthesis. In this year we have paid much attention to synthetic strategy to generate speech of whisper, breathy, harsh and tense quality as well as various talker individualities, using ARX (auto-regressive with exogenous input) speech analysis-synthesis method that was developed last year. As for whisper voice, we have investigated acoustic mechanism to interpret the formant structure specific to whisper voice and found new theory to explain frequency shift of lower formants based on MRI (magnetic resonance imaging) measurements of the larynx and computer simulation of acoustic resonance of the vocal tract. In order to produce breathy voice, we have proposed a method to control voicing source parameters and amount of laryngeal noise. Regarding harsh voice, we have first developed a sophisticated analysis-conversion-synthesis system that allows us to manipulate characteristics of jitter, shimmer, spectral fluctuation and laryngeal noise and then studied contributions of these parameters to the perception of harsh voice. From the experiments we have found that cross effects exist among these parameters to generate harsh voice quality. Tense voice has been successfully generated by controlling open quotient and spectral tilt of a voicing source waveform. Talker individuality has been found largely related to the static nature of formant trajectories and less to the dynamics.