H2HMem
H2HMem: A Multimodal Memory Benchmark for Agents in Human-Human Interactions
Shiping Zhu, Yibo Yang, Zhengyang Wang, Tiancheng Shen, Dandan Guo, Ming-Hsuan Yang · 2026-06-08
- Modality
- Audio & speech
- Data
- Hybrid (real + synthetic/simulated)
- Task · ground truth
- QA/Learn · Ans.+Ev.
- Scope
- Memory-specific benchmark
What it tests
- CECross-modal evidenceDirectly evaluated
- LELong-range evidenceDirectly evaluated
- SCState changesDirectly evaluated
- CFConflicting memoriesDirectly evaluated
- UQUnanswerable queriesDirectly evaluated
- MAMemory-grounded actionNot evaluated
- SPSpeaker trackingPartially evaluated
- ACAcoustic-cue retentionNot evaluated
- RPRepair persistencePartially evaluated
Survey notes
Timestamped multimodal human-human sessions are converted into memory before subsequent retrieval and QA; prior interactions supply the required evidence.