A technique for re-synthesis of speech from its low-dimensional PLP representation is presented. The technique is based on a speaker-dependent linear mapping from the vector of cepstral coefficients of the low-order PLP model to a vector of resonance peaks of the synthesis model. The linear mapping can be constructed from a relatively small amount of training data.
Bibliographic reference. Hermansky, Hynek / Anthony Cox Jr., Louis (1991): "Perceptual linear predictive (PLP) analysis-resynthesis technique", In EUROSPEECH-1991, 329-332.