Abstract
A speaker-dependent isolated word recognition system for Hindi digits is reported. The system uses energies and zero crossing rates of the speech signal and its first and second derivatives as recognition features. The system recognizes the input digit by comparing it with linearly time normalized reference patterns of all the digits and finding a best match. The recognition system has been tried out on the test set data containing 390 spoken digits (39 repetitions often digits) and a recognition score of 98.7% has been found.