Study of talker individuality by using ARX speech analysis-synthesis-editing system

Abstract

Experiments were performed to investigate roles of voice source and vocal tract characteristics in the perception of talker individuality by using a new formant-type speech analysis synthesis-editing system based on the ARX (auto-regressive with exogenous input) speech production model. One of the key features of the system is implementation of the algorithm that can automatically estimate voice source parameters and vocal tract parameters of the synthesizer directly from a speech utterance. Results of the experiments among five male talkers show that vocal tract characteristics contribute more to the perception of talker individuality than the voice source and the static component of the formant trajectories is a primary cue to talker individuality.