π Fish Audio S2 Pro is now on Pinokio: 1-click install
Fish Audio's 4B-parameter text-to-speech model runs fully on your own machine.
What it does
- π Reads text aloud in 80+ languages. English, Japanese and Chinese sound best.
- ποΈ Clones a voice from 10β30 seconds of reference audio. Save it once and reuse it any time.
- π Takes direction inside the text: drop tags like [whisper], [laughing], [excited], [pause] or [professional broadcast tone] where you want the delivery to change.
- π₯οΈ Comes with a clean web UI: choose mp3, wav, opus or pcm output and adjust temperature and chunk length.
- π Includes a local API (/v1/tts) for your own apps and scripts.
Example
β I can't believe it [laughing] you actually did it! [whisper] Don't tell anyone.
Before you install
- You need an NVIDIA GPU with ~24 GB of VRAM (RTX 3090, 4090 or 5090) and ~25 GB of free disk space. CPU, AMD ROCm on Linux and Apple Silicon also work, just slower.
- The model is gated: accept the terms on huggingface.co/fishaudio/s2-pro (https://huggingface.co/fishaudio/s2-pro) before clicking Install.
- License: the model weights are for research and non-commercial use only.
π Install β Start β generate.
π Fish Audio S2 Pro is now on Pinokio: 1-click install
Fish Audio's 4B-parameter text-to-speech model runs fully on your own machine.
What it does
Example
β I can't believe it [laughing] you actually did it! [whisper] Don't tell anyone.
Before you install
π Install β Start β generate.