ElevenLabs Eleven v3 Delivers Emotional Voice Expressiveness at Sub-150ms Latency; Community Showcases Multimodal Edge AI on M3 Pro

ElevenLabs launched Eleven v3, a real-time voice AI model with emotional expressiveness — including sighing, whispering, and laughing — at under 150ms latency for speech-to-text, which practitioners are calling a step-change for voice product development. Separately, a developer demonstrated Parlor, a real-time multimodal AI system combining audio/video input with voice output running efficiently on consumer Apple M3 Pro hardware using Gemma E2B, earning 150 upvotes on Hacker News and showcasing advancing edge AI inference capabilities. Developer GuppyLM, a minimal LLM implementation for educational transparency into model internals, attracted 623 Hacker News upvotes, reflecting strong community appetite for open, interpretable AI tooling.