LuxTTS Gradio UI

January 24, 2026 ยท View on GitHub

LuxTTS is a lightweight zipvoice-based text-to-speech model designed for high quality voice cloning and realistic generation at speeds exceeding 150x realtime.

https://github.com/user-attachments/assets/a3b57152-8d97-43ce-bd99-26dc9a145c29

UI install

git clone https://github.com/NidAll/LuxTTS-Gradio.git
cd LuxTTS-Gradio
pip install -r requirements.txt
python gradio_app.py

Windows beginner setup

powershell -ExecutionPolicy Bypass -File scripts/setup_windows.ps1
.\.venv\Scripts\python.exe gradio_app.py

UI usage

  1. Open the local Gradio URL shown in the terminal.
  2. Enter text.
  3. Upload a prompt (WAV/FLAC recommended) or record from the mic.
  4. Wait for "Model status: ready".
  5. Click Generate and play the output.

What is in this repo

  • Gradio UI in gradio_app.py with upload/record tabs, prompt preview, and advanced controls.
  • Windows beginner setup script in scripts/setup_windows.ps1.

UI features

  • Model loads on startup; Generate stays disabled until ready.
  • Upload tab for files and Record tab for microphone input.
  • Prompt preview panel with status and format guidance.
  • Advanced settings accordion for model, prompt, and generation controls.
  • Vocos frequency range control for timbre shaping.
  • return_smooth toggle for alternative output behavior.

Advanced controls

  • Model: model_path, device, CPU threads.
  • Prompt: duration, RMS, vocos frequency range.
  • Generation: num_steps, guidance_scale, t_shift, speed, return_smooth.

Windows optimizations

  • Model loads at UI startup and the Generate button stays disabled until ready.
  • ASR pipeline disables torchcodec to avoid Windows DLL and ffmpeg mismatch crashes.
  • Preview and prompt handling use librosa/soundfile; WAV/FLAC work without ffmpeg, MP3 requires ffmpeg on PATH.