Qwen3-TTS
lastmod 2026-01-26created 2026-01-261 edits
Provides voice design, cloning and generation
About
- Qwen3-TTS is a series of multilingual speech generation models that support voice cloning, voice design, ultra-high-quality human-like speech generation and natural-language-based control
- Supports 10 major languages, including Chinese, English, Japanese and Korean, plus various dialects, and comes in two model sizes, 1.7B and 0.6B
- Its in-house Qwen3-TTS-Tokenizer-12Hz encoder compresses speech signals efficiently while fully preserving paralinguistic information and the acoustic environment
- Its Dual-Track streaming architecture outputs the first audio packet immediately after a single character of input, achieving real-time synthesis with 97 ms latency
- Released as open source, so developers and companies can use high-quality speech generation technology directly