VoiceMem: Streaming Dual-Brain Memory for Real-Time Interaction
VoiceMem splits speech memory into factual and emotional tracks, then retrieves in 134 ms.
The paper describes a “dual-brain” memory system for real-time conversational speech models: a left side for informational recall and a right side for affective and persona memory. Its authors report stronger retrieval than Mem0 under their top-5 setup, plus state-of-the-art results across three persona benchmarks. They argue the latency fits inside standard voice-activity detection timing, so it adds no extra conversational delay. HF Daily Papers' note
The paper describes a “dual-brain” memory system for real-time conversational speech models: a left side for informational recall and a right side for affective and persona memory. Its authors report stronger retrieval than Mem0 under their top-5 setup, plus state-of-the-art results across three persona benchmarks. They argue the latency fits inside standard voice-activity detection timing, so it adds no extra conversational delay. HF Daily Papers' note
score 5