VoxPolyMem: Interaction-aware multimodal memory for multi-party spoken conversations beats baseline by 23.6 points
Wenxu Jia · hf · 2026-09-30
VoxPolyMem tackles long-term memory for multi-party spoken conversations, combining incremental speaker identification with a memory hierarchy of interaction memory, fact memory, and participant profiles. Retrieval is framed as sequential decision-making with an agent rewriting queries and selecting tools, trained via Evidence-Gain GRPO (EG-GRPO) for round-wise credit assignment. The authors also release VoxPolyBench. It scores 85.0 overall, beating the strongest baseline by 23.6 points, and exceeds public memory baselines by 8+ points on Mem-Gallery and H2HMem-Multi. Code and datasets are open-sourced.
More from coding & agent
- "Context + Harness is the new product": one tweet nails the agent stack shift — smtabatabaie · 2026-09-30
- 5 ways to cut LLM costs without changing models: optimize tokens, caching and calls — goyalshaliniuk · 2026-09-30
- Open Pstack: Codex as parent, Devin implements, Claude reviews — multi-provider agent orchestration — CombinationOk2374 · 2026-09-30
- VFX creator uses Opus 5.5 to drive Blender, freezing dancers into statues — anselm · 2026-09-30
- Cisco's 2026 campus hiring adds AI-agent project round with prompt documentation — cneuralnetwork · 2026-09-30
- One Year of Local Image Generation: Why Civitai and ComfyUI Both Fall Short — BenDLH · 2026-09-30