- Added try-except in AudioVAEComponentConfig.from_metadata() for JSON parsing
- Added try-except in model_detection.py for transformer config parsing
- Changed bare json.loads() calls to properly handle JSONDecodeError
- Provides clear error messages instead of crashing
- Added unit tests for the fixes
Fixes potential crashes when config contains invalid JSON
For LTX Audio VAE, remove normalization of audio during MEL spectrogram creation.
This aligs inference with training and prevents loud audio from being attenuated.
