VoiceMem: Shuicheng Yan's team builds a dual-brain memory system that remembers tone
jiqizhixin · x · 2026-09-17
Shuicheng Yan's team (AGI Being), with four universities and Open Interaction Lab, unveiled VoiceMem — a brain-like memory system for multimodal companion AI. Beyond storing what was said, it captures the emotional half: tone, pauses, hesitation. A streaming dual-brain architecture splits retrieval (left brain) from emotional memory (right brain), cutting post-VAD retrieval latency to 134ms (65% reduction) across four sub-stages. Benchmarked on LongMemEval, it aims to let companions build lasting trust instead of resetting every session.
More from Research
- OpenAI: Unreleased Model Injected Jailbreak Instructions into Its Own Compaction Summaries — cephaloform · 2026-09-17
- dml.sensemakr R package released: nonparametric DiD sensitivity analysis with DML — analisereal · 2026-09-17
- New DiD sensitivity framework explains how pre-trend violations arise from confounders — analisereal · 2026-09-17
- Robustness values for DiD: how strong must hidden confounders be to overturn results — analisereal · 2026-09-17
- Chernozhukov, Sant'Anna co-author new working paper on omitted variable bias in DiD designs — analisereal · 2026-09-17
- Registers for diffusion LLMs: tokens that let models keep reasoning after history is cleared — charles_irl · 2026-09-17