How affects can perturbe the automatic speech recognition of domotic interactions

Frédéric Aman (1) , Véronique Auberge (2) , Michel Vacher (1)

1 GETALP - Groupe d’Étude en Traduction Automatique/Traitement Automatisé des Langues et de la Parole
2 LIG - Laboratoire d'Informatique de Grenoble

In Smart Home, the vocal home automation orders, for comfort purposes, or assistive devoted, have been pointed as the more relevant interaction for ambient assisted living. Even if the orders are very strictly formulated, when they are daily used (directed to the smart home, or to a robot mediator), they become often pronounced with various affects. In this paper we have evaluated how some state of the art ASR systems shut down with expressive orders, acted or spontaneous, and how the ASR training with neutral and/or acted and/or spontaneous expressive commands corpus can greatly modify the ASR performances.

Domaines: Intelligence artificielle [cs.AI]

Frédéric Aman, Véronique Auberge, Michel Vacher. How affects can perturbe the automatic speech recognition of domotic interactions. Workshop on Affective Social Speech Signals, Aug 2013, Grenoble, France. pp.1-5. ⟨hal-00953247⟩