Robust Speech Recognition Based on Spectro-temporal Features