Effects of the vocal tract shape in the spectral tilt of the glottal pulse waveform

摘要

Practical models of the glottal pulse can help enhance the overall naturalness of synthetic speech. A method for capturing the first-order effects is developed based on an approximation of the spectral envelope for a family of vowel-dependent glottal pulses. A linear relationship between the vocal tract and the glottal synthesis network parameters is obtained using least-squares minimization on the derived data. The method is practical and does not use excessive computational or memory resources.