The spectral dynamics of a non-linear model of the voice source signal was examined in two experiments. Controlled changes in the output spectrum were brought about by varying the amplitude of the model's excitation signal. A first experiment illustrated the model's ability to output continuously evolving spectra during an increase or decrease in time of the excitation function amplitude. In a second experiment, a comparison was carried out between the harmonic values of glottis cycles obtained by inverse filtering on the one hand, and the harmonics of synthesized glottis signals on the other; the identification of the parameters of the model was carried out on the basis of the same experimentally obtained cycles. Results show a very good agreement between real and synthetic voice source signals. In a second stage, we tested the ability of the model of a glottis cycle to approach the spectrum of other glottis signals produced by a same speaker under different conditions; here, the only control parameter was the amplitude of the model's excitation function. Results show that synthesized spectra manage to approximate the overall spectral trend, though without being able to reproduce individual spectral components exactly.
Bibliographic reference. Schoentgen, Jean (1989): "The spectral dynamics of a non-linear model of the glottal waveform", In EUROSPEECH-1989, 2481-2484.