2020/10/17 by Pavlos Avgoustinakis, Avgoustinakis, Pavlos, Giorgos Kordopatis-Zilos +7
Biochemistry, Genetics and Molecular Biology · Computer Science · #Audio and Speech Processing (eess.AS) #Cancer-related molecular mechanisms research #FOS: Computer and information sciences #FOS: Electrical engineering #Information Retrieval (cs.IR) #Machine Learning (cs.LG) #Multimedia (cs.MM) #Music and Audio Processing #Sound (cs.SD) #Video Analysis and Summarization #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2010.08737
openalex publication_date 2020/10/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
In this work, we address the problem of audio-based near-duplicate video\nretrieval. We propose the Audio Similarity Learning (AuSiL) approach that\neffectively captures temporal patterns of audio similarity between video pairs.\nFor the robust similarity calculation between two videos, we first extract\nrepresentative audio-based video descriptors by leveraging transfer learning\nbased on a Convolutional Neural Network (CNN) trained on a large scale dataset\nof audio events, and then we calculate the similarity matrix derived from the\npairwise similarity of these descriptors. The similarity matrix is subsequently\nfed to a CNN network that captures the temporal structures existing within its\ncontent. We train our network following a triplet generation process and\noptimizing the triplet loss function. To evaluate the effectiveness of the\nproposed approach, we have manually annotated two publicly available video\ndatasets based on the audio duplicity between their videos. The proposed\napproach achieves very competitive results compared to three state-of-the-art\nmethods. Also, unlike the competing methods, it is very robust to the retrieval\nof audio duplicates generated with speed transformations.\n