A LoRA Is All You Need to Turn a Speech LLM Full-Duplex
kastnerkyle · x · 2026-11-12
A technical writeup shows you don't need to train a full-duplex voice model from scratch: take an existing speech LLM, freeze it, and train only 0.3% extra parameters via LoRA for about 30 hours on a single GPU — and it learns full-duplex interaction.
More from Multimodal
- Runway open-weights Praxis-1 27B world action model for robots, with NVIDIA — tlakomy · 2026-11-12
- Psychedelic 'Claude Fable 5.2 self portrait' image goes viral on X — repligate · 2026-11-12
- NVIDIA and ElevenLabs expand voice agent partnership with 15MW Blackwell compute and sovereign deployments — nvidia · 2026-11-12
- Seedream V5.5 tested: dozens of characters, hundreds of objects in one complex scene — mark_k · 2026-11-12
- Claude flags mispronunciations in video audio it claims it can't hear — RileyRalmuto · 2026-11-12
- ComfyUI tip: paint inpaint masks as layered PNGs and reuse them across runs — Deftroyy · 2026-11-12