Full-Duplex Interaction in Spoken Dialogue Systems: A Comprehensive Study from the ICASSP 2026 HumDial Challenge
New benchmark lets AI converse naturally, managing interruptions and speech overlap in real time.
A team of researchers from various institutions has published a comprehensive study on full-duplex interaction in spoken dialogue systems, stemming from the ICASSP 2026 Human-like Spoken Dialogue Systems Challenge (HumDial Challenge). Traditional systems rely on rigid turn-taking, but this work aims to enable AI to handle interruptions, speech overlap, and dynamic turn negotiation—key elements of human conversation. The study introduces the HumDial-FDBench benchmark, built from a new high-quality dual-channel dataset of real human conversations that captures these complex interactions.
The benchmark assesses a system's ability to maintain conversational flow while managing interruptions. The researchers also created a public leaderboard to compare the performance of open-source and proprietary models, promoting transparent and reproducible evaluation. This initiative supports the development of more responsive and adaptive dialogue systems, moving beyond simple question-answer formats to truly human-like interaction.
- The HumDial Challenge introduces a dual-channel dataset of real human conversations with interruptions and overlapping speech.
- The HumDial-FDBench benchmark assesses systems on handling interruptions while maintaining conversational flow.
- A public leaderboard enables transparent comparison of open-source and proprietary models for full-duplex dialogue.
Why It Matters
Enables AI to converse naturally, handling interruptions and overlap, crucial for realistic voice assistants and customer service.