# Multi-TTS-Server Pinokio Launcher

One-click Pinokio launcher for [hermes/multi-tts-server](https://gitea.amakusa.ru/hermes/multi-tts-server).

## What it installs

This launcher downloads the TTS server source code, creates a Python virtual environment, installs dependencies, installs the correct PyTorch build for your GPU, and downloads the required voice models.

## Requirements

- Windows, macOS, or Linux
- NVIDIA GPU with CUDA 12.8, AMD GPU with DirectML (Windows only), or CPU
- At least 16 GB of system RAM
- ~8 GB of free disk space (venv + models)

## Included engines

- **F5-TTS** — Russian model `Misha24-10/F5-TTS_RUSSIAN` (~1.3 GB)
- **XTTS-v2** — Russian model `NeuroDonu/RU-XTTS-DonuModel` (~5.3 GB)
- **CPU fallback** for machines without a compatible GPU

## One-click usage

1. Open Pinokio.
2. Install this launcher.
3. Click **Install**.
4. When install completes, click **Start**.
5. The Gradio Web UI and FastAPI docs will open automatically.

## Features

- OpenAI-compatible speech endpoint: `POST /v1/audio/speech`
- MultiTTS Android-compatible endpoint: `GET /tts?text=...&voice=...&engine=f5-tts`
- Engine load/unload controls: `POST /engines/{name}/load` and `/engines/{name}/unload`
- MultiTTS Android export: `GET /export/multitts`

## Reset

Click **Reset** to remove the virtual environment, cached files, and downloaded models. The source code in `app/` is kept so the next install only recreates the venv and re-downloads models.

## Update

Click **Update** to pull the latest launcher and app source code, then re-run the install steps.

## License

This launcher wraps the multi-tts-server project. Use in accordance with the licenses of F5-TTS, XTTS-v2, and the upstream project.