Qwen3-TTS

Qwen3-TTS is a next-generation multilingual speech synthesis (TTS) model series open-sourced by the Qwen team at Alibaba Cloud’s Tongyi Lab, officially released in January 2026. The models adopt an innovative Discrete Multi-codebook LM architecture, powered by the self-developed Qwen3-TTS-Tokenizer-12Hz for efficient acoustic compression and high-dimensional semantic modeling. The series offers 0.6B and 1.7B parameter variants across multiple model types — CustomVoice, VoiceDesign, and Base — covering everything from lightweight deployment to high-quality synthesis.

Qwen3-TTS sets a new benchmark for open-source TTS in terms of feature completeness. Voice cloning — the Base model can accurately replicate a speaker’s timbre with just 3 seconds of reference audio, supporting cross-lingual cloning and reusable speaker embeddings. Voice design — the VoiceDesign model allows users to define custom voice styles through natural language descriptions (e.g., “a playful, childish anime girl voice”), achieving a “what you imagine is what you hear” creative experience. Streaming synthesis — built on a Dual-Track hybrid streaming architecture, end-to-end latency is as low as 97ms, with the first audio packet delivered immediately after a single character is input, meeting the rigorous demands of real-time interactive scenarios. Instruction control — supports flexible control over timbre, emotion, speed, and prosody via natural language instructions, with the model deeply understanding text semantics and adaptively adjusting expression.

Qwen3-TTS supports 10 major languages: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian, along with multiple dialectal voice profiles to meet global application needs. The models are released under the Apache 2.0 license, with a complete Python package (qwen-tts), a WebUI demo, and DashScope API services. vLLM-Omni provides official Day-0 support. The project has accumulated over 12,000 stars on GitHub and is well-suited for smart voice assistants, audio content creation, virtual digital humans, cross-lingual dubbing, and real-time speech interaction scenarios.