To manage these complex conversational dynamics, a responsive system must continuously estimate fine-grained turn states

Introduction Achieving natural spoken dialogue requires systems to seamlessly handle continuous speech, backchannels, user interruptions while maintaining low latency [1]