Unsupervised Data Drift Monitoring for Deployed Deepfake Speech Detectors
Deepfake speech detectors are typically evaluated as binary classifiers on fixed benchmarks, but deployed systems face evolving attacks, codecs, speakers, and recording conditions, with labels unavailable during operation.
ProofPaper ↗
Key points
- We propose an unsupervised drift monitor that operates on the internal embeddings of a deployed detector.
- The monitor compares each input against bona fide and spoof references using regularized Mahalanobis distances.
- We evaluate the monitor on four modern detectors on three datasets.
- Our findings suggest that deployed deepfake speech detectors should be treated not only as a binary classification, but also as a reliability problem.
Sources (1)
- [1]Unsupervised Data Drift Monitoring for Deployed Deepfake Speech DetectorsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 02:08 PM
Deepfake speech detectors are typically evaluated as binary classifiers on fixed benchmarks, but deployed systems face evolving attacks, codecs, speakers, and recording conditions, with labels unavailable during operation.
We propose an unsupervised drift monitor that operates on the internal embeddings of a deployed detector.
Extractive summary: sentences quoted from the sources.