Characterization of Speech during Imitation

Abstract

The present work analyzes some effects of imitation on the produced voice, focusing on both macro-level parameters such as the pitch contour, as well as on micro-level acoustical structures such as the formant frequencies. The following speech production characteristics are addressed: (1) Tracking of pitch contours is carried out using a super-resolution algorithm for pitch detection. (2) Formant frequencies are estimated for the vowel phonemes using a linear prediction technique, approximating the vocal tract by an all-pole model. (3) Cross-sectional tube areas of the vocal tract are estimated using inverse Levinson algorithm. A strong tendency of pitch contours during imitation to resemble those of the target speaker was found. A significant tendency of the second and third formants during imitation to resemble those of the target person was identified. A tendency of the cross sectional areas to be smaller during imitation was found.