Phoneme-level framework for explainable speech deepfake detection

A phoneme-level framework using wav2vec 2.0 and HuBERT explains why speech is classified as real or fake. Other papers study robustness to real-world corruption and generalization to synthetic sound effects. A large-scale analysis of DETECT-3B-Omni confirms its independence from speech content and demographics.