Add NaN/Inf safety check to prevent JSON serialization errors

Fix torch version for PyPI compatibility

Fix model loading: upgrade transformers to 5.0.0 and handle safetensors metadata issue

Fix: Apply compatibility patches before speechbrain import

Fix Dockerfile: Copy all files to include src and config

Prioritize root in sys.path and add debug info

Fix Import Error: Add __init__.py files and debug logs

Fix ModuleNotFoundError by adding root to sys.path

Refactor project structure to Unified AI Voice Detection System

Fix SpeechBrain compatibility with newer torchaudio by mocking list_audio_backends

Fix model loading error by updating torch version and improving logging

Add fastapi to requirements.txt

Heavy & Accurate: Integrated SpeechBrain VAD + MMS-300M pipeline

Upgrade: Switch to MMS-300M (XLS-R) for robust multilingual deepfake detection

Fix: Require heuristic evidence for AI verdict on non-English audio to prevent false positives

Implement Conservative Consensus: SNR Analysis and Human Liveness Veto to eliminate false positives

Implement supreme Hybrid AI detection: sliding window, spectral and temporal features, and language-aware confidence.

Add comprehensive debug logging to track AI classification confidence calculations

Upgrade to specialized deepfake detection model and optimize audio processing

Fix: Align prediction label with pitch variance heuristic

Optimize: Truncate audio to 5s & add per-step timing logs

Fix: Final audio processing fixes and enhanced logging

Fix: Add debug logging and robust padding to audio processing

Fix: Robust Base64 decoding

Fix HF Spaces metadata

Configure Dockerfile for Hugging Face Spaces

Feat: ADD Dockerized the API

Initial commit: Multilingual Voice Detector with XLS-R Support