A new framework for robust speech recognition in complex channel environments