Junichi Yamagishi

A training-free self-embedding strategy can significantly enhance the detection of partial deepfake speech, even when only small segments are manipulated.

Embedding a neural codec representation enables full recovery of manipulated audio segments, a breakthrough for content integrity in speech recordings.

Artifact-specific experts can provide interpretable evidence in speech deepfake detection, transforming how we assess synthetic audio authenticity.

RL fine-tuning with GRPO can significantly boost the generalization of speech deepfake detectors to unseen attacks, outperforming standard supervised methods.