tiny-voca: 18MB TTS model runs in browser at 70x real-time on CPU

tiny-voca, a 4.5M-parameter English TTS model (18MB, 21MB ONNX), runs directly in the browser without GPU, account, or API, reaching 70x real-time on CPU. Caleb Fahlgren trained it from the open-source project for just $18.

2026-11-16 ~ 2026-11-17 · 4 related posts

2 near-duplicate retellings: thisdudelikesAI · thisdudelikesAI