Date d'évènement : 01/09/2026
Thomas Risse - IRCAM
Real-time vocal synthesis using self-oscillating physical models
Abstract:
This work focuses on developing methods and associated simulation codes that can enable the stable, real-time synthesis of voiced sounds from reduced-order, self-oscillating physical models. This type of model is typically composed of a mechanical model of vocal fold dynamics coupled with various acoustic components, such as the trachea and vocal tract. It has been used for analysis and vocal synthesis for several decades. VocalTractLab is an example of a comprehensive and evolving self-oscillating articulatory synthesiser. However, most (if not all) existing works of this kind are deployed in offline environments. To achieve real-time synthesis and enable direct user interaction with the synthesis parameters, the solving methods must be fast and stable enough to prevent the solution from blowing up while playing.
|
The model is written as an assembly of constitutive blocks in the port-Hamiltonian framework. In the vocal tract, a time-varying Webster equation is used, coupled to yielding vocal tract walls and a radiation condition at the lip boundary. A standard FDTD scheme is used for time-space discretisation. The vocal folds and glottal flow are described using a modified version of Story and Titze's classical body-cover model. This modification restores the mass and energy balances of the glottal flow. Due to nonlinearities in the constitutive laws of both the vocal fold model and the glottal flow law, a specific time-discretisation strategy is employed. This strategy, which is based on a scalar auxiliary variable transformation, results in a numerical scheme that is almost explicit and satisfies a discrete power balance, and is therefore stable. A C++ implementation of the solver is proposed. Computation times are close to 100 times faster than real-time on a laptop. The solver can therefore be used in real-time environments, as demonstrated by a Max/MSP example. Bringing synthesis to such environments raises the question of interface design for parameter control. First experiments and perspectives in this regard will be presented. |
 |
 |
‼️ Le séminaire peut être suivi en ligne via le lien : https://univ-amu-fr.zoom.us/j/95011527683?pwd=bTeNvaaZY75buaa7bRL0pcucXSASm7.1 |
Le mardi 1er septembre 2026 à 11h00 / Amphithéâtre François Canac, LMA
Voir en ligne : plus d’informations concernant l’orateur