Try it in your browser

Speak with Miro & Dii

Pick a voice, type a sentence, and hear it. Each voice is a tiny ~15.6M-parameter model, downloaded straight from Hugging Face and run entirely in your browser with onnxruntime-web — nothing is sent to a server.

Preview models. These voices are a work in progress — they are still being trained and do not represent the final quality. Expect rough edges; they will keep improving.
Advanced (piper inference)

Tiny models — runs on a potato 🥔

Each voice is a ~15.6 million-parameter model — small enough to run on a Raspberry Pi, a phone, or the very browser tab you're in right now, fully offline, no GPU and no cloud. While the big TTS labs brag about billions of parameters, we optimize for the opposite: tiny, fast, private, on-device voices that anyone can run and own.

Runs in your smart home

The eSpeak voices are piper / Home Assistant compatible — the exact .onnx you just heard drops into theHome Assistant piper add-on. One voice identity, on the web and on your local voice assistant. Read the story inTwo voices, every language.

First synthesis of a voice downloads its model (~63 MB) from Hugging Face and caches it; later runs are instant and fully offline.

An OpenVoiceOS collaboration

Miro & Dii are built withOpenVoiceOS, the community non-profit voice platform — a collaboration in progress. The models are published openly onHugging Face and run on thephoonnx engine. Everything here is FOSS and self-hostable.

Open source — train your own voice

phoonnx is fully open source, including the training code (see itsTRAINING.md). No gatekeeping, no API keys — the goal is to democratize text-to-speech for everyone, especially low-resource and minority languages. Build datasets and train your own voices with ourml-notebooks (TTS, wake-word, and intent-classifier pipelines).

Want to use phoonnx voices in your browser?

This whole demo is powered byphoonnx.js, our open-source library for in-browser VITS TTS with onnxruntime-web — Unicode and espeak phonemizer paths, no server. Install it straight from GitHub and drop it into your own site or app:

npm i github:TigreGotico/phoonnx.js onnxruntime-web

Source & docs:github.com/TigreGotico/phoonnx.js.