Characterizing phonetic convergence with speaker recognition techniques
Résumé
Speakers are known to accommodate to each other's behavior when interacting. Information circulates via multiple sensory-motor loops operating at various levels of the interaction and this closed-loop process induces modifications in all levels of representation from social and psychological evaluation to low-level gestural behaviors such as gaze, respiratory patterns, or speech. Various authors have proposed that these representations tend to converge or diverge according to cognitive demand. While quite plausible, claimed observations of such behavior in speech are extremely controversial. The effects are rather small, and are difficult to capture and characterize objectively. This paper focuses on the study of the convergence between phonetic representations - spectral realizations of speech sounds - using automatic classification techniques developed for speech and speaker recognition. Using data collected during a novel language game we term 'verbal dominoes', we show that scores are comparable between global techniques and a more fine-grained analysis focused on vocalic segments.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...