HiThink Turn: An Intent-Aware Turn-Taking Control Module for Full-Duplex Dialogue

2026-09-28

AI Summary

This study addresses the challenge of achieving timely and selective interruptions in full-duplex dialogue systems, where end-of-utterance prediction proves insufficient. To this end, we propose an intent-aware streaming turn-state predictor that decouples response intent from semantic completeness. We introduce a novel minimal-intent-sufficient-prefix supervision strategy, leveraging large language model-assisted annotation combined with speech alignment. Furthermore, truncated audio training is incorporated to enhance robustness against partial utterances. The system performs streaming inference and playback state decisions based on 240ms chunks, enabling low-latency interruption control. Experimental results demonstrate that our method achieves state-of-the-art performance across multiple benchmarks, yielding a 98% interruption success rate and reducing the average stopping latency by 60.9%.

Abstract

Full-duplex dialogue requires timely yet selective interruption handling, which end-of-turn prediction alone cannot achieve: complete utterances may need no response, while unfinished requests may warrant interruption. To address this challenge, we propose HiThink Turn, an intent-aware streaming turn-state predictor that separates response intent from semantic completeness and conditions decisions on system playback state. A key contribution is minimal intent-sufficient prefix supervision, constructed through LLM judgments and speech alignment, while training on audio truncated at chunk boundaries improves robustness to partial speech. These components support streaming inference with 240-ms audio chunks, enabling low-latency, accurate full-duplex turn control. Experiments show that HiThink Turn leads the compared methods in Easy Turn macro accuracy, Full-Duplex-Bench average interaction rate score (0.933), and non-target-speech average playback resume rate (0.735). Additionally, intent-prefix triggering raises interruption success from 89\% to 98\% and reduces mean stop latency by 60.9\%.

Problem

• full-duplex dialogue
• turn-taking control
• interruption handling
• intent awareness
• end-of-turn prediction

Innovation

• Full-Duplex Dialogue
• Intent-Aware Turn-Taking
• Streaming Inference
• Minimal Intent-Sufficient Prefix Supervision
• Low-Latency