TABLE OF CONTENTS The PSOLA technique ("Pitch Synchronous OverLap and Add") is not, strictly speaking, a speech signal modeling technique. It is rather a speech signal processing technique, which can be applied to natural and synthetic signals and whose objective is to enable the control and modification of prosodic parameters. The originality and efficiency of such a technique consists of modifying independently the fundamental frequency and the duration of the signal, at a local level and without altering the timbre of the original voice. These functionalities are indeed of primary importance for a speech synthesis system.
This paragraph presents both from an historical angle and in a comparative way, the foundations and algorithms for modifying prosodic parameters in a speech signal. However, this presentation cannot be exhaustive, and we will focus mostly on the algorithms that are currently used in existing speech synthesis systemm [9].
A text-to-speech synthesis system must have the capacity to control three essential prosodic parameters: the duration of sounds, the pitch contour and the energy over time. The role of the linguistic processing step (see section 3.4) is to predict, from the written text, sequences of numerical values for each of the three prosodic parameters. It is therefore necessary to design operators that can be applied to the speech signal so as to reach particular prosodic targets. The fundamental frequency F 0 is a parameter specific to the glottal excitation. We have already discussed the difficulty of extracting this...
