SPEECH CONVERSATION SYSTEM OF THE MUSICIAN ROBOT.

In this paper, a speech input/output system to realize flexible conversation is described, which is implemented as a conversational subsystem of WABOT-2, a keyboard playing robot. In the speech input system, for the efficient treatment of the topics in the conversation, a network model having path weighting facility is proposed for the sentence recognition. In the speech output system, to make easy to produce arbitrary responses, the synthesis-by-rule is adopted. In this method, the deterioration of quality of the synthesized speech becomes main problem. Here, a new parameter interpolation method considering the sensitivity of formant frequencies for the k-parameter is proposed. With these methods, a conversational system with high flexibility is realized.