TABLE OF CONTENTS This section presents the two main methods that have served as the support for developing new speech synthesis systems. The goal is to relate the acoustic content of the message which we have so far considered as a phoneme sequence, i.e. from a linguistic viewpoint to the description of a speech signal that ultimately expresses itself as a variation of acoustic pressure.
As it is not conceivable to consider a priori all the possible acoustic configurations of a same phoneme [7], and because the phoneme remains the only link with the text to be pronounced, two methods link the abstract phonetic description and the concrete acoustic reality:
rule-based synthesis is based on an explicit description of the control parameters over time representing the speech signal. It consists of a set of rules derived from an explicit knowledge of production phenomena, at least at the acoustic level;
unit-based concatenative synthesis rejects the above hypothesis and proposes a black-box approach. Information which is impossible to model from the speech signal is encapsulated into a speech unit which expresses the link between the linguistic and acoustic levels. The definition of this unit is solely based on external characteristics for instance, its phonetic description.
Independently from the signal model used in the synthesis system, a rule-based approach for speech synthesis involves modeling the transitions between phonemes in terms of rules dynamically. As opposed to the unit-based concatenative approach, the rule-based approach provides...
