Tried the Qwen3 TTS, and it's awesome
Unproofread notes
Previously, I have used several local models and setups for text-to-speech and nothing works satisfactory as paid APIs like form Elevenlabs and OpenAI. But recently learned about the new Qwen3 TTS model from this Simon Willison's post and I tried it.
And to my surprise, it's awesome. The audio quality is great, and the expressions are natural and understandable as well. I tried the model Qwen3-TTS-12Hz-1.7B-Base by downloading the weights from HuggingFace and it just works.
But if you would first like to try this out before installing it locally, here's the live demo on the HuggingFace website.
Comment via email