Synthesis by Rule of Disordered Voices

Abstract

The synthesis of disordered voices designates the use of numerical methods to simulate the vocal timbre of speakers suffering from laryngeal pathologies or dysfunctions to investigate the link between perceived timbre and speech signal properties. The simulation is based on a mapping of the amplitude of a narrow-band input signal onto the amplitude of a desired output signal, while the cycle lengths of the input and output are identical. The proposed amplitude-to-amplitude mapping, also known as waveshaping, makes possible simulating a wide range of timbres by fixing the control parameters of a cascade of elementary waveshapers. These enable evolving sample by sample the open quotient, pulse onset and offset rounding, speed quotient and formant ripple of the glottal airflow rate. Preliminary perceptual tests show that the perceived naturalness of the synthetic timbres is comparable to or better than the perceived naturalness of timbres generated via template-based waveshaping.