AI
Duplex-MPE: Benchmarking Multi-Party Interaction in Full-Duplex Dialogue
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 24/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 10:25:22 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 05:18:25 UTC.
Briefings
The Duplex-MPE benchmark evaluates multi-party dialogue by testing whether speech models should answer, remain silent, or stop when a human speaker resolves a request. It uses 2,000 scenarios with three to four human speakers and one assistant, featuring both explicit and implicit addressing. Five open-weight speech systems were tested, with MiniCPM-o 4.5 leading on three scored capabilities. A transcript-based Gemini 3.1 Pro reference showed a 64.3 percentage point higher response rate for explicit requests compared to implicit ones, while paired tests found no significant difference in response rates between the speech systems.
Key points
- The benchmark introduces a new evaluation framework for full-duplex dialogue where models must decide whether to answer, stay silent, or stop speaking when a human speaker resolves a request.
- Five open-weight speech systems were evaluated, with MiniCPM-o 4.5 leading on three scored capabilities, while other systems showed mixed performance with frequent speech and inaccurate answers.
Context and limitations
The benchmark relies on continuous conversation audio without transcripts or supplied turn boundaries, and evaluation is limited to five specific speech systems.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 10:25:22 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihunotpwgo3w2x5wg6ed4ax6fmasioprenacd2fkyhsdjkix4aeye