Liveness Verification in Audio-Video Speaker Authentication
Girija Chetty, Michael Wagner · 2004
In this paper we propose liveness verification for audio-video speaker authentication systems to guard against possible replay attacks that employ pre-recorded audio and/or still face samples from a photo. The technique uses a fusion of acoustic features and visual features, automatically extracted from the lip region, to differentiate synchronous audio-video presentations from asynchronous replay attacks. Experiments with the multi-modal VidTIMIT database achieve liveness verification equal-error rates of less than 1 % using lip-voice feature vectors comprising mel-cepstral coefficients, eigenlips and lip region measurements. 1.