VoiceMem: dual-brain memory brings near-zero latency long-term recall to voice agents
青稞AI · wechat · 2026-09-07
Researchers from NTU, Tsinghua, CUHK and others (first author Zhifei Xie, corresponding author Shuicheng Yan) released VoiceMem, a long-term memory framework for real-time voice agents. It pairs an informational 'left brain' (Schema-Entity indexing with cluster emergence) with an emotional 'right brain' (persona and affect attribution), and streams retrieval across four stages so the 134ms dual-brain search fits inside the VAD window for near-zero added latency. On LoCoMo it scores 91.2 retrieving only 5 memories, beating Mem0's top-200 by 30 points, and leads on 11 of 14 capability categories in its new ChatMem-Bench. Framework, dataset and demo are open source, with Qwen Audio Agent integration.
Related event: VoiceMem Cuts Voice AI Memory Retrieval Latency to 134ms(2 posts)→
More from Research
- Hashed n-gram value embeddings push GPT to 3.2115 loss, beating 30M track record — RobertTLange · 2026-10-27
- TMLR paper: enforced thinking doesn't reliably make LLM agents disclose key info to users — SharonYixuanLi · 2026-10-27
- Diamandis: AI Cancer Navigator and $100 Genome Sequencing Close the Care Gap — PeterDiamandis · 2026-10-27
- EMNLP Oral: training-free Nexus Sampling cuts KV-cache memory 10x; OpenResearcher tops BrowseComp-Plus at 54.8% — yuz9yuz · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- Forking gap follows a power law in context frequency; masking rare n-grams shrinks it — ZimingLiu11 · 2026-10-27