New methods for robust speech recognition