MOSS-Live: A Full-Duplex Real-Time Interaction System
Blog post from Hugging Face
MOSS-Live is an OpenMOSS real-time interaction system designed for continuous, full-duplex conversations that combine simultaneous listening and speaking, selective attention to directed speech, visual input, conversational memory, and background tool execution. Its architecture separates a foreground layer that manages live turn-taking, interruptions, acknowledgments, and topic shifts from a background layer that performs longer tasks such as research or planning, with a coordination layer connecting task state and conversation context. Demonstrations include event planning, road-trip adjustments, and camera-based exploration of Shanghai, illustrating its ability to retain changing constraints and resume interrupted topics. Across benchmarks including FDB, Big-Bench Audio, and τ-Voice, MOSS-Live reports particularly strong performance in full-duplex interaction metrics, tool-use success, backchannel selectivity, speech understanding, and post-interruption response quality, while trailing some competing models on certain measures such as overall audio accuracy and multi-turn task success.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| Real-time | 12 | 2,990 | 636 | 190 | -28% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.