Singing Voice Analysis and Synthesis System through Glottal Excited Formant Resonators.
Piero Pierucci, Andrea Paladin · University of Michigan Library Repository · 1997
This paper presents a system for singing voice analysis and synthesis, that makes use of formant resonators to model the vocal tract behaviour. This well known approach is complemented with the introduction of a modeling of the source signal which is based on an inverse filtering method. Using this source modeling in combination with a cascade of formant resonators produces a synthesized signal that preserves the timbre characteristics and fine details of the original voice, with the advantage of a compact representation. Furthermore a considerable separation among source and vocal tract characteristics could be obtained, which allows for easy pitch and rate modifications. The synthesis system is suitable for low cost real-time implementation using fixed point DSP hardware. A MARS platform implementation of the synthesis system is outlined in the paper.