Psychoacoustical Approach to Sinusoidal Modeling of Speech

This paper describes the psychoacoustical approach to the sinusoidal analysis/synthesis of the speech. With the SN (sinusoids plus noise) model, the periodic part of the speech signal is described by time-varying sinusoids and the non-periodic part by noise. The number of sinusoids needed for reconstruction can be then reduced by application of the psychoacoustical principles. The proposed compression scheme allows us to reduce the amount of physical data needed to store the parameters. The whole method can be used in speech synthesis for storing speech corpuses that are prepared for easy prosodic manipulation.