Universal speech models for speaker independent single channel source separation | IEEE Conference Publication | IEEE Xplore