Leveraging large amounts of loosely transcribed corporate videos for acoustic model training | IEEE Conference Publication | IEEE Xplore