Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice: Technical Overview and Model Release
Qwen has released Qwen3-TTS, a multilingual speech generation model family featuring a self-developed 12Hz tokenizer, dual-track streaming architecture, and instruction-based voice control across ten major languages. The models achieve strong performance in zero-shot text-to-speech, voice cloning, and voice design benchmarks compared to existing systems. Open article →