Scenema-Audio: Gibberish Output and Hugging Face SSL/403 Download error

Hi people,

I’m trying to run Scenema Audio locally with Podman. I’m facing two issues:

Gibberish output: The model generates unintelligible audio instead of the expected speech.

Hugging Face download: The ~4.9 GB transformer checkpoint fails with SSL: CERTIFICATE_VERIFY_FAILED, Read timed out, and eventually 403 Forbidden – Make sure your token has the correct permissions.

My Hugging Face token is working (whoami() returns successfully), and I’ve also added the corporate CA certificate provided by IT. A direct HTTPS test to Hugging Face returns 200, but the model download still fails.

Has anyone faced these issues or knows the correct fix, particularly for the HF CDN/Xet download?