A transfer learning and progressive stacking approach to reducing deep model sizes with an application to speech enhancement | IEEE Conference Publication | IEEE Xplore