Modeling stop-consonant releases for synthesis

Abstract

This study is part of a project leading to rule-based speech synthesis using the HLsyn synthesizer. In HLsyn, stop-consonant releases are generated by controlling the time variation of a constriction that is formed by the lips, the tongue blade, or the tongue body. In order to generate a consonant release that is perceptually acceptable and that has acoustic characteristics that match those of normal speech, it was found that the trajectory of the consonant release (cross-sectional area versus time) had to exhibit an initial rapid rise, followed by a delay in which the rise was interrupted, followed finally by a final rise. The burst at the consonant release is generated by the airflow through the constriction during the time that the rise is delayed. A model of the time course of this three-stage release of the articulator has been developed, taking into account the role of the intraoral pressure, the tapering formed by the articulator, and the glottal area. The model shows that the delay between the initial and final area increases is progressively longer for labials, alveolars, and velars, in agreement with acoustic data on the burst duration.