Anticipation of Forged Video Evidence using Machine Learning

Abstract

To detect audio manipulation in a pre recorded evidence videos by developing a synchronization verification algorithm to match the lip movements along with its audio pitch values. Audio video recognition has been considered as a key for speech recognition tasks when the audio is sullied, as well as visual recognition method used for speaker authentication in multispeaker scenarios. The primary aim of this paper is to point out the correspondence between the audio and video streams. Acquired audio feature sequences are processed with a Gaussian model. [1].This proposed method achieves parallel processing by effectively examining multiple videos at a time.In this paper, we train the machine by convolutional neural network (CNN) and deep neural network (DNN).CNN architecture maps both the modalities into a depiction space to evaluate the correspondence of audiovisual streams using the learned multimodal features. DNN is used as a discriminative model between the two modalities in order to concurrently distinguish between the correlated and uncorrelated components. The proposed architecture will deploy both spatial and temporal information jointly to effectively discover the correlation between temporal inf different modalities. We train a system by capturing the motion picture. This method achieves relative enhancement over 20% on the equal error rate and 7% on the average precision in comparison to the state of the art method.