Samples for DAFx00
In the paper Model-Based Synthesis and Transformation of voiced sounds we propose a physically informed model for the glottis system, loosely based on the well known Ishizaka Flanagan model. Here are some sound examples summarizing the basic procedure we used for analysis, identification, synthesis and transformation of voiced sounds (see the paper for details).
- A simple voice signal (radiated pressure) is considered. We take the voiced part of this signal.
- Using standard inverse-filtering techniques, the vocal tract all-pole filter is identified and the glottal flow signal is reconstructed from the radiated pressure.
- The glottal model is identified using the above glottal flow signal. Then the system can be run using different values for the physical parameters. Here only the resonance frequency has been changed, in order to lower the pitch and add some vibrato to the glottal signal.
- The resynthesized glottal signal is convolved with the identified vocal tract. We are thus able to modify some key features of the voiced sound (pitch, energy, open phase / closed phase ratio...) without altering voice identity.