Marcelo ParadaSANCHES, I.2022-01-122022-01-122017-01-05PARADA, M.; SANCHES, I. Visual Voice Activity Detection Based on Motion Vectors of MPEG Encoded Video. Proceedings - UKSim-AMSS 11th European Modelling Symposium on Computer Modelling and Simulation, EMS 2017, p. 89-94, 2017.https://repositorio.fei.edu.br/handle/FEI/3858© 2017 IEEE.This article presents a visual voice activity detector (VVAD) that relies on features extracted from an MPEG encoded video, e.g. MPEG-4 AVC (H.264) or MPEG-H part 2 HEVC (H.265). The technique uses the inter frames resulting from the motion compensation and estimation of the temporal redundancy analysis, in which blocks of the image comprises the representation of motion vectors. The proposed algorithm relies on this information already available in the encoded video to estimate the lip motion and thus reducing processing time and code complexity. A standard voice activity detector (VAD) based on the audio signal is also performed and similarities on the audio and video approaches can assure not only a more reliable voice activity detector, especially in noisy environments but can also be useful as liveness detection for biometric systems.Acesso RestritoArtigo de evento10.1109/EMS.2017.26biometryliveness detectionVADVVAD