Dimensions acoustiques de la parole expressive : poids relatifs des paramètres resynthétisés par Praat vs. LF-ARX
Nicolas Audibert, Damien Vincent, Véronique Aubergé, Albert Rilliard, Olivier Rosec.

The emotional prosody is multi-dimensional. A debated question is whether some parameters are more specialized to convey some emotion dimensions. Selected stimuli carrying acted expressions of anxiety, disappointment, disgust, disquiet, joy, resignation, satisfaction and sadness on monosyllabic words were used to synthesize artefactual stimuli by projecting separately prosodic parameters on neutral expressions, with Praat and an LF-ARX algorithm. Perceptive evaluation of stimuli and comparison of results (1) indicate that F0 contours bring more information on positive expressions while voice quality and duration convey more information on negative expressions, and intensity alone is not informative enough (2) diagnoses minor artifacts of both synthesis methods which consequences may have interesting implications in expressive speech synthesis (3) validates the efficiency of the LF-ARX algorithm (4) measures the relative weights of each of the LF-ARX voice quality parameters.