title	emoji	colorFrom	colorTo	sdk	sdk_version	app_file	pinned	license	suggested_hardware
Qwen3-TTS Demo	🎙️	blue	purple	gradio	5.33.0	app.py	false	apache-2.0	zero-a10g

Qwen3-TTS

Text-to-Speech model based on Qwen3 architecture with voice cloning and voice design capabilities.

🚀 Quick Start

Option 1: Docker Setup (Recommended)

Using Pre-built Docker Image

A pre-built Docker image is available on DockerHub at starlento/qwen3-tts:latest.

Prerequisites:

• Docker and Docker Compose installed
• NVIDIA GPU with CUDA support
• NVIDIA Container Toolkit installed

Steps:

1. Start the container:

docker-compose up -d

The application will be available at http://localhost:7860

Available Apps:

• app_custom_voice.py - Custom voice synthesis (default)
• app_voice_clone.py - Voice cloning
• app_voice_design.py - Voice design

To use a different app, modify the command line in docker-compose.yaml:

command: python app_voice_clone.py --server-name 0.0.0.0 --server-port 7860

Option 2: Host Setup with UV

Prerequisites

• Python 3.10
• NVIDIA GPU with CUDA 12.9 support
• UV package manager (will be installed automatically if not present)

Installation Steps

2. Run the UV setup script:

chmod +x setup_uv_env.sh
./setup_uv_env.sh

The script will:

• Install UV if not already installed
• Create a virtual environment with Python 3.10
• Install PyTorch 2.8.0 with CUDA 12.9 support
• Install all required dependencies

3. Activate the environment:

source .venv/bin/activate

4. Run the application:

# Default app
python app.py

# Or choose a specific app
python app_custom_voice.py
python app_voice_clone.py
python app_voice_design.py

⚠️ Important Notes

Flash Attention

Note: Flash Attention is NOT installed by default in either setup method. If you need flash-attn for optimized attention mechanisms, you'll need to install it manually.

Model Storage

Models are downloaded to:

• Docker: /models directory (mapped to ~/models on host)
• Host: Default HuggingFace cache (~/.cache/huggingface)

📦 Docker Image

• DockerHub: starlento/qwen3-tts:latest
• Base: Python 3.10-slim
• Includes: PyTorch 2.8.0 with CUDA 12.9 support
• Size: ~8GB