Qwen3-TTS

Qwen3-TTS is a set of open-source text-to-speech (TTS) models that the Qwen team publishes on GitHub, in 0.6B and 1.7B sizes. It can design a new voice from a written description, read with a preset voice, or clone a voice from reference audio; the notes for the Base model say 3 seconds of audio is enough. The README says it covers 10 major languages, including Chinese, English, Japanese and Korean, plus several dialect voices. Voice design, voice cloning and natural-language control of tone all come in the same set of models.

Who it's for:developers and audio creators who want to run a speech model on their own computer or server. Readers who don't write code won't get much from it.

Before you start: running it locally means supplying your own compute and model files. You can also use Alibaba Cloud's DashScope API, which needs an account and an API key and is billed by usage; your audio then goes to Alibaba Cloud. The code is under the Apache-2.0 licence; we didn't read the licence for the model weights. The project doesn't say what permission you need to clone someone else's voice, so only use your own voice or one whose owner has agreed.

Based on the project's public documentation. Not tested by AI Inspiration World.