Dev reverse-engineers Wispr Flow, fine-tunes LFM3-1.7B for free local voice dictation
teortaxesTex · x · 2026-11-06
Developer elliotarledge reverse-engineered paid dictation tool Wispr Flow and fine-tuned LFM3-1.7B (using DeepSeek) into a free, open-source alternative that runs entirely on a Mac.
- Key-up latency: 65 ms on M4 Max vs 520 ms for Wispr Flow on the same clips; a 6-minute dictation session lands in about a quarter second
- Rust backend with a native Metal engine, each decode step runs as one fused megakernel, plus a dflash2 speculative drafter trained on a local 4x RTX Pro 6000 server
- Pipeline: Parakeet transcribes while LFM3-1.7B refines, both stream during speech so key-up only waits on the last chunk
- Fully offline (no internet/ping issues), supports custom term dictionaries and per-app adaptation
More from Infra
- Google to spend $17B on massive AI data center in Wyoming — Polymarket · 2026-11-06
- Hugging Face burns 800K GPU hours to release free 70K-word MoE training handbook — Thom_Wolf · 2026-11-06
- Coherent revenue jumps 50% to $2.37B, signs ~$20B five-year PhotonLink deal — BenBajarin · 2026-11-06
- Chip designer pitches Apollo on small AI data centers across 20 US states — pstAsiatech · 2026-11-06
- SpaceX data center pledges 40% power discounts and billions in community benefits — Scobleizer · 2026-11-06
- Mystery seller seeks to dump $500 million of data center loans — dinabass · 2026-11-06