Qwen3-TTS Full Demo

All released variants • CPU-friendly • No streaming (full generation only)

Uses 9 built-in premium voices + optional style instruction

Model

1.7B-CustomVoice 0.6B-CustomVoice

Precision

float32 float16

Text to speak

Language

Speaker

Style instruction (optional)

Generated Speech

Notes
• First generation per model loads weights (may take 1–5 min).
• Use float32 if float16 causes crashes (common on CPU).
• 0.6B models are faster / lighter on CPU.
• No streaming yet in official qwen-tts package — generations are full-text → full-audio.
• Repo & docs: https://github.com/QwenLM/Qwen3-TTS