Rungano

A native desktop shell for Higgs Audio v3 TTS — multi-language, voice-cloning, emotion/style/SFX control.

Rungano means "folktale" in Shona. Built to faithfully narrate Shona oral storytelling traditions — voice cloning, auto-chunking for long-form narration, and complete control over emotion, style, and sound effects.

What this is

Rungano wraps the SGLang-Omni serving stack in a real desktop application (Electron) with a clean dark UI, system-tray integration, and a one-click launcher. It exposes the full control surface of the Higgs v3 model:

• 21 emotions
• 3 styles (singing, shouting, whispering)
• 10 prosody tags (speed, pitch, pauses, expressiveness)
• 9 sound effects (laughter, crying, screaming, etc.)
• Zero-shot voice cloning from a 10-30s reference clip
• 102 languages, 24 kHz output

Architecture

┌─────────────────────────────────────────────────────────────┐
│  Electron Desktop App  (this repo)                          │
│  ┌──────────────────┐    IPC    ┌────────────────────────┐  │
│  │  Renderer (UI)   │◄────────►│  Main process          │  │
│  │  HTML/CSS/JS     │          │  - Window mgmt         │  │
│  │                  │          │  - System tray         │  │
│  └────────┬─────────┘          │  - Subprocess control  │  │
│           │ fetch              └──────┬──────────┬──────┘  │
│           │                          │          │         │
│           ▼                          ▼          ▼         │
└─────────────────────────────────────────────────────────────┘
                                    │          │
                    ┌───────────────┘          └──────────────┐
                    ▼                                        ▼
         ┌──────────────────────┐                ┌──────────────────────┐
         │  Python Flask proxy  │                │  SGLang-Omni (WSL2)  │
         │  :7861               │  OpenAI-compat │  :8000               │
         │  - /api/voices       │ ─────────────► │  - /v1/audio/speech  │
         │  - /api/tts          │                │  - 4B model on GPU   │
         │  - /api/health       │                │  - Voice cloning     │
         │  - /api/history      │                │  - 24 kHz output     │
         └──────────────────────┘                └──────────────────────┘

Install

Prerequisites (one-time):

• Windows 10/11 with WSL2 enabled
• NVIDIA GPU with CUDA drivers (RTX 3090 verified)
• Python 3.10+
• Node.js 18+

Run the installer:

scripts\install.bat

This will:

1. Create node_modules/ (Electron)
2. Create server\.venv\ and install flask + requests
3. Inside WSL2: install CUDA toolkit, uv, SGLang, SGLang-Omni (from source)
4. Download the 4B Higgs model (~8 GB)

Total install time: 15-30 minutes (mostly the SGLang-Omni source compile).

Run

Easiest: double-click the desktop shortcut created on first launch. Alternative:

scripts\launch-higgs.bat

The window opens once the Flask proxy is healthy (~30 sec on warm start). Closing the window hides to the system tray. Use the tray menu to quit.

License

Non-commercial use only. See LICENSE.md for the full text. Commercial deployment requires a license from Boson AI.