Speech imitation with a child’s voice: addressing the correspondence problem

摘要

We hereby present our first steps towards linking an embodied speech acquisition system and a speech production module, in order to provide a robot with the ability to reproduce acquired speech representations. Due to the type of interaction intended for the robot, we endowed it with a child-like voice, concretized with the use of a new vocoder-like technique for speech synthe- sis. The task in hand consists of finding and using a correspon- dence between configurations in the tutor’s acoustic parameter space, which might be inaccessible for the system’s voice, and phonetically equivalents in the robot’s. This mapping is learned by having a cooperative tutor imitating the robot’s monophonic utterances, giving the robot the necessary knowledge to map a tutor’s utterance to its own vocal space, and imitate it.