Modeling of the glottal flow by physical models with waveform matching properties

Physical modeling of sounds and voice is used to study and simulate the
sound production mechanisms of various acoustic systems and of
phonation.
One major drawback is that fitting a physical model to real data is
usually a very difficult task. We describe an approach to sound and voice
analysis/synthesis which combines the physical modeling and the
black-box
modeling approaches. We discuss in particular the problem of fitting a
physical model of the glottis to glottal flow waveforms obtained by
inverse-filtering of voiced speech signals. A waveform-matched
low-dimensional model of the glottal source is presented, and a
data-driven identification procedure is outlined. A set of inverse-filtered glottal
flow waveforms with different characteristics is used to test the
effectiveness of the method. Experimental results are shown to demonstrate how the
model can reproduce a wide range of target waveforms.