Reproduce “Phoneme Discretized Saliency Maps for Explainable Detection of AI-Generated Voice” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shubham Gupta, Mirco Ravanelli, Pascal Germain, Cem Subakan
The authors introduce Phoneme Discretized Saliency Maps (PDSM), which turn audio saliency maps into phoneme rankings for explaining voice detectors. An ASR system assigns time frames to phonemes, whose spans are scored and masked to test influence. PDSM raises masking-based faithfulness on trained CNN14 classifiers; for FastSpeech2, Integrated Gradients scores 0.35 versus 0.001 directly, a result limited to this test setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.