EgoMem evaluation
EgoMem: Lifelong Memory Agent for Full-duplex Omnimodal Models
Yiqun Yao, Naitong Yu, Xiang Li, Xin Jiang, Xuezhi Fang, Wenjia Ma, et al. · 2026-01-30
- Modality
- Audio & speech
- Data
- Hybrid (real + synthetic/simulated)
- Task · ground truth
- Ret./Gen./Learn · ID+Fact+Resp.
- Scope
- Memory-specific benchmark
What it tests
- CECross-modal evidenceDirectly evaluated
- LELong-range evidenceDirectly evaluated
- SCState changesDirectly evaluated
- CFConflicting memoriesNot evaluated
- UQUnanswerable queriesNot evaluated
- MAMemory-grounded actionPartially evaluated
- SPSpeaker trackingDirectly evaluated
- ACAcoustic-cue retentionPartially evaluated
- RPRepair persistencePartially evaluated
Survey notes
Held-out omnimodal dialogue streams evaluate identity-linked retrieval and personalized response consistency with stored profile and social-relation facts.