Stream tqdm progress to Gradio UI

Preload cu13 NVRTC and default Whisper validation on

Run inline on Space instead of via FastAPI

Update app.py

Update app.py

Update app.py

Copy reference audio before processor deletes it; tidy UI into advanced accordion

Disable Whisper validation (libcublas missing on ZeroGPU)

Resolve model paths absolute; drop persistent-storage assumption

Preload AudioProcessor at module import for ZeroGPU snapshot

Initial Gradio ZeroGPU app for Scenema Audio