Roberto Gretter and Maurizio Omologo
A speech synthesizer is being developed which exploits a Linear Predictive Coding (LPC) technique based on the Line Spectrum Pair (LSP) representation.
Elementary speech units are defined as non uniform segments by taking into account the widely accepted view that a small set of short units (e.g. phonemes, allophones) can hardly provide high quality synthetic speech.
The aim of this paper is to describe the LSP representation and its application to the synthesis task.
Then some preliminary results, related to the impact of interpolation techniques on the speech unit coding, are presented and discussed.