Perceptual Linear Predictive (PLP) Analysis-Resynthesis Technique

Abstract

A common wisdom in speech re-synthesis is that while the vocal tract excitation can be modified to represent the message prosody, the accurate preservation of the formants is needed in order to ensure that both the linguistic message and the speaker-dependent information is well represented in the synthesized speech. Formants are speaker-dependent. A further decomposition of the formant-based speech representation into its message-bearing and the speaker-dependent parts and the inverse problem of combining those two sources of speech information is of interest. The current paper addresses this issues.