Search papers, labs, and topics across Lattice.
4
0
6
Contemporary frontier speech models still catastrophically fail on living Arabic dialects, with top systems like GPT-4o achieving an error rate no better than 35.0% WER across 17 regional vernaculars.
Bridging the gap between audio embeddings and LLMs, this method achieves a remarkable 16.2% boost in deepfake voice detection accuracy on unseen domains.
Evidence Subspace Projection reveals how different factors influence deepfake detection decisions, providing a clearer understanding of model behavior in self-supervised learning.
The best deepfake audio detectors are surprisingly biased by audio quality, speaker gender, and language, undermining their real-world reliability.
