Qwen3‑TTS is a new open‑source suite of advanced text‑to‑speech models released by Alibaba’s Qwen team. It brings cutting‑edge speech synthesis capabilities — including voice cloning, voice design, and multilingual generation — to developers, researchers, and creators. (qwen.ai)
Open‑Source Release
• The entire Qwen3‑TTS model family is published under the Apache 2.0 license, meaning you can use and build on it in both research and commercial projects. (Let’s Data Science)
Multi‑Model Architecture
• The suite comprises several models across two main parameter sizes (around 0.6B and 1.7B parameters).
• Models include:
– Base: Efficient text‑to‑speech and fast voice cloning.
– CustomVoice: Preset expressive voices with style control.
– VoiceDesign: Create new custom voices via natural‑language descriptions. (MarkTechPost)
Multilingual Support
• Qwen3‑TTS supports speech generation in at least 10 languages, including Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian. (MarkTechPost)
Voice Cloning in Seconds
• The Base models can clone a voice using as little as 3 seconds of input audio. (DEV Community)
Real‑Time Streaming & Low Latency
• Thanks to a specialized 12Hz speech tokenizer and efficient architecture, Qwen3‑TTS can begin streaming speech in ~97 ms, making it suitable for interactive applications. (GIGAZINE)
Multilingual & Cross‑Lingual Capabilities
• You can clone voices in one language and generate speech in another, enabling cross‑lingual voice applications. (DEV Community)
Benchmarks & Quality
• Independent benchmarks report that Qwen3‑TTS models deliver high speaker similarity and low error rates compared with competitors like MiniMax or ElevenLabs. (DEV Community)
• Hugging Face hosts the Qwen3‑TTS models and demos where you can experiment with voice cloning and speech generation. (Hugging Face)
• Browser‑based demos are available, allowing easy testing of voices without setup. (Qwen3 TTS)
In summary: Qwen3‑TTS is among the most advanced open‑source text‑to‑speech model families available in early 2026. It combines high‑quality, natural speech, multilingual support, voice cloning, and real‑time performance — all under a permissive license that encourages widespread use and development. (qwen.ai)