With the increase in the number of cellular customers using the VoiceDialing system deployed by NYNEX and its customers, and the anticipated future cellular applications of speech recognition, the issue of robustness against noise has become a critical issue in our system. NYNEX is therefore incorporating noise-cancellation into the currently deployed system. A number of algorithmic and implementational questions were raised by the strong requirements of compatibility with existing landline and cellular users of speaker-dependent recognition. Mismatch-condition evaluations, post-processing and other algorithm issues are discussed here.