LPC-based voice transformation from adult to child

Abstract

An effective transformation of sentence level voiced speech from young to aged, or male to female, requires the parametrization and subsequent modification of various components in speech. A glottal excited linear predictive model (GELP), similar to that developed by Childers and Hu [D. G. Childers and H. T. Hu, J. Acoust. Soc. Am. 96, 2026–2036 (1994)] was used as a basic platform for experiments in gender, age, and vocal quality transformation. This model permits a parametrization initially into excitation and articulation components, which were then separately modified. This study describes the changes in pitch, prosody, glottal pulse shape, and formant structure necessary to evoke a perceptual transformation from an adult male to a prepubescent child. The transformation was performed at the sentence level. A time-domain glottal pulse model which is sensitive to gender/age was used, along with a male-to-child vocal-tract pseudoarea warping. [Work supported by NIH Grant DCO2532-01.]