Settings
Device, models, and storage. Everything stays on your machine.
This device
WebGPUchecking…
Active inference pathnot loaded yet
Model in memoryno
WebGPU uses the full-precision model (~326 MB download, fastest). Without it, the app uses the quantized model (~92 MB) on the CPU — slower than real time, but fully offline after download.
Downloaded models
No model files downloaded yet. Generate speech once in the Studio and the model will be cached here.
About
FreeTTS v1 — free, local-first AI text-to-speech.
Engine: Kokoro-82M (Apache-2.0) via kokoro-js, ONNX Runtime Web.
App license: GPL-3.0-or-later (see THIRD-PARTY-NOTICES.md).