The features include being able to experience all 3 types of Hugging Face Space demos currently available,
and the model is saved in the root folder rather than the cache folder.
Alibaba seemed reluctant to release Qwen3 TTS, so I had given up on it,
but they're releasing it as open weights like this. ㅠㅠ They're truly a beacon of light in the open model world.
---------------------
Qwen3-TTS_autopatch.7z is a version with several features patched.
On macOS, you can perform inference using CPU and MPS,
and on Windows, you can perform inference using available acceleration such as ML, CUDA, and CPU.
Additionally, since SoX installation is cumbersome on Windows, SoX dependencies have been removed from both macOS and Windows.
A button has been added to automatically fetch the qwen3tts repository and apply necessary patches.
In other words, after extracting and setting up the Python environment, the remaining necessary files from the qwen3tts repository will be prepared automatically.
* Since there is a pyproject.toml file, you can build a virtual environment with uv sync.
Download
https://github.com/DINKIssTyle/DINKIssTyle-Qwen3TTS-GUI/archive/refs/heads/main.zip
GitHub
https://github.com/DINKIssTyle/DINKIssTyle-Qwen3TTS-GUI
