A highly robust deep learning technique for overlap detection using audio fingerprinting

Due to the proliferation of video-based applications, there is a high demand for automated systems to support various video-based tasks that are free from human intervention, i.e., manual tagging. In this paper, we present a novel approach for detecting the presence of overlap between two videos by exploiting their corresponding audio signals, which is a crucial preprocessing step for audio, and further video alignment and synchronisation. Several existing approaches have limitations related to timestamps, overlapping regions, and the length of video clips. For the proposed work, we target the challenging scenario consisting of simultaneously recorded videos in an unconstrained manner by multiple users attending performance events. xOur work is an attempt towards developing a robust framework that not only considers noisy components present in the audio but is also free from the limitations mentioned above. We compare our framework with several other existing approaches. Our proposed framework outperforms other approaches by an average of 13.71% in terms of accuracy.

Automated summary

Due to the proliferation of video-based applications, there is a high demand for automated systems to support various video-based tasks. This paper presents a novel approach for detecting overlap between two videos by analyzing their audio signals. The proposed framework outperforms other approaches by an average of 13.71% in terms of accuracy.