Unit-Based Agent for Semi-Cascaded Full-Duplex Dialogue Systems

2026-01-28

AI Summary

Achieving natural full-duplex human-machine spoken interaction requires balancing real-time responsiveness with contextual coherence. This work proposes a decoupled framework based on minimal dialogue units, leveraging a multimodal large language model to construct a semi-cascaded full-duplex system. By integrating voice activity detection (VAD) and text-to-speech (TTS) modules, the approach enables plug-and-play, training-free modular deployment. The system supports independent processing of input streams and state-transition prediction, facilitating seamless turn-taking and context-aware responses. Evaluated on the HumDial dataset, the method demonstrates strong performance and secured second place in the Full-Duplex Interaction Track of the Human-like Spoken Dialogue Systems Challenge.

Technology Category

• Natural Language Processing: Conversational AI/Dialog Systems
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Full-duplex voice interaction is crucial for natural human computer interaction. We present a framework that decomposes complex dialogue into minimal conversational units, enabling the system to process each unit independently and predict when to transit to the next. This framework is instantiated as a semi-cascaded full-duplex dialogue system built around a multimodal large language model, supported by auxiliary modules such as voice activity detection (VAD) and text-to-speech (TTS) synthesis. The resulting system operates in a train-free, plug-and-play manner. Experiments on the HumDial dataset demonstrate the effectiveness of our framework, which ranks second among all teams on the test set of the Human-like Spoken Dialogue Systems Challenge (Track 2: Full-Duplex Interaction). Code is available at the GitHub repository https://github.com/yu-haoyuan/fd-badcat.

Problem

• full-duplex dialogue
• human-computer interaction
• conversational units
• spoken dialogue systems
• real-time interaction

Innovation

• full-duplex dialogue
• conversational units
• multimodal large language model
• plug-and-play system
• voice activity detection