Combining spectral feature mapping and multi-channel model-based source separation for noise-robust automatic speech recognition | IEEE Conference Publication | IEEE Xplore