fp16 ONNX export of Irodori-TTS, packaged for in-browser WebGPU inference with chrome-on-aituber. The browser app downloads these files directly (CORS) and stores them in browser storage; nothing is sent to a server.

The ONNX export and the inference runtime are built with ngc-shj/irodori-tts-webgpu. The reproducible build recipe lives in the chrome-on-aituber repository.

This repository redistributes derivative weights and runtime code. See licenses/ for the full texts and licenses/THIRD_PARTY_NOTICES.txt / licenses/NOTICE for details.

This model performs reference-audio-based voice synthesis (voice cloning). Do not use it for impersonation, fraud, or generating misleading content. Only use reference audio you have the right to use. Note that, unlike the Python reference implementation of Irodori-TTS, this browser runtime does not apply SilentCipher watermarking to generated audio (see licenses/THIRD_PARTY_NOTICES.txt).