Voice Interaction in an Interface Agent for Process Plant

摘要

An interface agent that mediates between an operator and a process plant in operation was developed. The agent has the ability of speech recognition and understanding, and provides an operator with the functionality of oral command. In this paper, we describe the oral command system of the interface agent, which contains a speech recognition engine and a natural language processor. The combination of keyword spotting by an commercially available speech recognition engine and case-based language understanding allowed us to build a flexible and intelligent oral command system which accepts ill-formed spoken sentences such as utterance hesitating and word omission. We built a voice interaction model in locutionary act level and introduced the notion of context into the speech recognizing and the language processing in order to overcome errors and ambiguities caused by the recognition engine. The demonstration showed excellent capability for voice interaction between an operator and the agent. This study was achieved as one of the Human Media Research and Development Projects under the Industrial Science and Technology Frontier Program in Japan.