The source-filter paradigm has been widely used to model, synthesize and analyse vocal sounds. In spite of their efficiency, most of such models neglect some physical phenomena which could significantly improve naturalness. In this paper, we consider a modified source-filter model which includes a simplified aeroacoustic coupling between the glottal airflow and the vocal tract while keeping low-cost computation and efficient analysis methods. We introduce a glottal area waveform model derived from observations on high-speed video-endoscopic recordings and based on the so-called Liljencrants-Fant (LF) model. The vocal tract is modeled by a concatenation of straight pipes with lossless plane waves propagation. The coupling is ensured by the standard Bernoulli equation, a flow-separation model and continuity constraints for acoustic pressure and flow at the inner end of the vocal tract. This voice production model is driven by the sub-glottal pressure, the glottal area and the vocal tract geometry. Moreover, we introduce a sound analysis method for this model. In conclusion, we present some synthesis, analysis and transformation examples to evaluate the performances of this model and compare it with a classic source-filter model.
Contribution au colloque ou congrès : Acoustics 2012