Evaluation of a New Model for Vowels Synthesis with Perturbations in Acoustic Parameters

Voice signal contains intrinsic irregularities which become more evident in the presence of pathologies. The acoustic parameters are very useful for the clinical assessment of voice and pathologies detection. Most existing voice models handle irregularities as additive noise and not as information carriers. In this work, a new model is proposed allowing to generate synthesized voices with previously selected acoustical parameters shimmer and jitter. Artificial voices are generated from a glottal source signal, obtained by conveniently disturbing amplitudes and periods, and then filtered using an auto-regressive linear filter. Models were developed for amplitude and period perturbations based on statistical methods. Several signals were generated and the performance of the model was analyzed. The quality of synthesized voices was evaluated using an objective quality measurement. The obtained jitter and shimmer values mostly agreed with the theoretically predicted values. These results suggest that this model is useful for artificial voices generation. Keywords— irregular-voice, voice synthesis model, acoustic parameters, jitter, shimmer.