This paper describes the main features of a real time hardware continuous speech recognition system. Using both parallel and pipelined processing techniques, matching up to several hundred words from a previously stored vocabulary of whole word "templates" is possible in real time. An efficient single pass dynamic programming algorithm is used to find the sequence of templates that best represents the input. Continuous recognition is achieved using a traceback technique on partial recognition results. The acoustic analysis contains a number of features to improve performance. In particular a novel noise compensation algorithm is briefly described.