A temporal simulator for developing turn-taking methods for Spoken Dialogue Systems

Ethan Selfridge and Peter Heemam

Developing sophisticated turn-taking behavior is necessary for next-generation dialogue systems. However, incorporating real users into the development cycle is expensive and current simulation techniques are inadequate. As a foundation for advancing turn-taking behavior, we present a temporal simulator that models the interaction between the user and the system, including the user's speech, voice activity detection, incremental speech recognition, and natural language generation. We describe the details of the simulator and demonstrate it on a sample domain.