Interspeech'2005 - Eurospeech
The presentation concerns a synthesizer of disordered voices. The synthesizer consists of a non-linear model of the phonatory excitation and a vocal tract simulation based on a concatenation of cylindrical tubelets. For each tubelet, viscous, thermal and wall vibrations losses are modeled by means of numerical filters. A conical tubelet is added at the lip-end to simulate the transition from one-dimensional to three-dimensional wave propagation. A source-tract interaction is included. The synthetic phonatory excitation signal is obtained via a shaping function that transforms an harmonic driving function into the desired waveshape. The instantaneous frequency and the spectral slope of the phonatory excitation are controlled by the instantaneous frequency and amplitude of the harmonic driving function. Several types of dysperiodicities are simulated by modulating these two parameters. The voice disorders that the synthesizer is able to simulate are pathological vocal jitter and vocal tremor, biphonation, diplophonia and random vibrations of the vocal folds. Turbulence noise is modeled via additive white noise.
Bibliographic reference. Hanquinet, Julien / Grenez, Francis / Schoentgen, Jean (2005): "Synthesis of disordered speech", In INTERSPEECH-2005, 1077-1080.