PHONEXIAbuyer room · privateManage buyer room
Proprietary labelled speech-authenticity evaluation corpora, disclosed indirectly but decisively through the product's calibration statement: the Replay Attack Detection score is a Log-Likelihood Ratio whose zero point corresponds to the Equal Error Rate (EER) on their evaluation datasets, with practical scores falling between -4.5 and +1.0. Producing an EER-calibrated threshold requires a held-out corpus with per-clip genuine-versus-replay ground truth, so the vendor necessarily maintains a labelled authenticity evaluation set. Content spans the acoustic dimensions named as the detection features — room reverberation, microphone quality, background noise and transmission characteristics — implying paired live-versus-replayed recordings of the same speakers across differing capture and transmission conditions. Part of a wider Deepfake Detection suite, so synthesis labels likely accompany replay labels
Get notified as soon as Phonexia publishes. Until then, discover alternatives today with a 7-day trial.