🎙️ Qwen3-TTS Enhanced

Clone any voice in seconds. 100% local, runs on your GPU.

An enhanced interface for Qwen3-TTS with multi-reference cloning, variation generation, and audio preprocessing.

---

✨ Key Enhancements

• Multi-Reference Cloning - Combine up to 5 audio samples for higher quality
• Variation Generation - Create 1-5 outputs and pick the best one
• Audio Preprocessing - Automatic normalization + optional noise reduction
• One-Click Install - Just run install.bat on Windows

---

Features

Feature	Description
🎤 Voice Clone	Clone voices from short audio (3+ seconds)
🎭 Create Voice	Combine multiple samples with per-file transcripts
👤 Custom Voice	9 preset speakers with emotion control
✨ Voice Design	Create voices from text descriptions
💾 Save & Load	Keep voices as portable .pt files
⚙️ Settings	Configure data folder, persists across updates

---

Requirements

• Windows 10/11 or Linux (Docker)
• NVIDIA GPU with 8GB+ VRAM (RTX 20 series or newer)
• Python 3.10-3.12 (auto-installs a bundled version if needed)

GTX 10 series (Pascal)? Use v1.2.3 which includes CUDA 12.4.

---

Quick Start

Windows

install.bat    # One-time setup
run.bat        # Launch app

Docker

docker-compose up --build

Open http://localhost:8000

First run downloads ~4GB of models.

---

Tips

• Longer audio = better clones (10-30 seconds ideal)
• Add transcripts for improved accuracy
• Enable "Clean audio" for noisy recordings
• Try multiple variations to find the best result

---

License

Apache 2.0 - See LICENSE

Built on Qwen3-TTS by Alibaba Cloud.