The objective of this paper is speaker recognition under noisy and\nunconstrained conditions.\n We make two key contributions. First, we introduce a very large-scale\naudio-visual speaker recognition dataset collected from open-source media.\nUsing a fully automated pipeline, we curate VoxCeleb2 which contains over a\nmillion utterances from over 6,000 speakers. This is several times larger than\nany publicly available speaker recognition dataset.\n Second, we develop and compare Convolutional Neural Network (CNN) models and\ntraining strategies that can effectively recognise identities from voice under\nvarious conditions. The models trained on the VoxCeleb2 dataset surpass the\nperformance of previous works on a benchmark dataset by a significant margin.\n
Self-supervised learning (SSL) achieves great success in speech recognition,\nwhile limited exploration has been attempted for other speech processing tasks.\nAs speech signal c...
While research in large vocabulary continuous speech recognition (LVCSR) has sparked the development of many state of the art research ideas, research in this domain suffers fro...
Social media, news, blog, policy document mentions