University seat number recognition: An application of connected alpha-digit recognition

University Seat Number (USN) of a student acts as a unique identity number for the student in an institution. Recognition of USN is necessary in order to get information about the student. USN recognition is one of the applications of connected alpha-digit recognition. Recognition of spoken alphabets and digits is difficult task in automatic speech recognition due to phonetic similarities among certain group of vocabulary sets. In this paper, knowledge based features are used along with conventional Mel Frequency Cepstral Coefficients (MFCC) to enhance the overall correctness of the USN recognizer by overcoming phonetic similarity difficulties. The USN correctness obtained using only MFCC feature is 78.57%. By adding additional knowledge based features, the overall USN correctness obtained is 82.14%, when tested on 616 USN samples.