README.md

metadata

title: Qwen3-TTS Demo
emoji: 🎙
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.25.0
app_file: app.py
pinned: false
license: apache-2.0

Qwen3-TTS Demo

Open-source text-to-speech with four modes:

1. Custom Voice - Pick a preset speaker with optional emotion instructions
2. Voice Design - Describe any voice in natural language and the AI creates it
3. Voice Clone - Clone a voice from a 3-second audio sample
4. Multi-Speaker Story - Auto-detect characters, assign voices, add emotions, generate full audiobook

Setup

Hardware: Requires GPU (runs on ZeroGPU for free on HF Spaces).

Optional: Add DASHSCOPE_API_KEY in Secrets for the Multi-Speaker Story mode (character detection + emotion analysis).

Supported Languages

English, Chinese, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian

Models Used

• Qwen3-TTS-12Hz-1.7B-CustomVoice (preset voices)
• Qwen3-TTS-12Hz-1.7B-VoiceDesign (natural language voice design)
• Qwen3-TTS-12Hz-1.7B-Base (voice cloning)

All models are Apache 2.0 licensed.