2021/06/21 by Anurag Kumar, Yun Wang, Kumar, Anurag +5
Computer Science · Engineering · #Acoustic Wave Phenomena Research #Artificial Intelligence (cs.AI) #Audio and Speech Processing (eess.AS) #FOS: Computer and information sciences #FOS: Electrical engineering #Music and Audio Processing #Sound (cs.SD) #Speech and Audio Processing #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2106.11335
openalex publication_date 2021/06/21 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Transfer learning is critical for efficient information transfer across\nmultiple related learning problems. A simple, yet effective transfer learning\napproach utilizes deep neural networks trained on a large-scale task for\nfeature extraction. Such representations are then used to learn related\ndownstream tasks. In this paper, we investigate transfer learning capacity of\naudio representations obtained from neural networks trained on a large-scale\nsound event detection dataset. We build and evaluate these representations\nacross a wide range of other audio tasks, via a simple linear classifier\ntransfer mechanism. We show that such simple linear transfer is already\npowerful enough to achieve high performance on the downstream tasks. We also\nprovide insights into the attributes of sound event representations that enable\nsuch efficient information transfer.\n