Qwen3-TTS

Voice design, cloning & 97ms streaming

Qwen3-TTS is a speech synthesis model family offering 0.6 billion and 1.7 billion parameter versions that generate multilingual speech in ten major languages. It supports ultra‑low‑latency streaming, creates voices from natural‑language prompts, performs three‑second zero‑shot voice cloning, and allows precise control of style and prosody. The system is suited for real‑time assistants, audiobooks, and other interactive applications that need fast, expressive speech output.