This paper describes a hidden Markov model (HMM) based utterance verification system designed in the frame-work of statistical hypothesis testing. The two major issues of how to design ke}'word and string scoring criteria are addressed. We motivate the need for discriminative hypothesis testing for verification. One such approach based on minimum classification error is investigated. When the proposed verification technique was integrated into a state-of-the-art connected digit recognition system, the string error rate for valid digit strings was found to decrease by 57% when setting the rejection rate to 5%. Furthermore, the system was able to correctly reject over 99.9% of non-vocabulary word strings.
Bibliographic reference. Rahim, Mazin G. / Lee, Chin-Hui / Juang, Biing-Hwang (1995): "Discriminative utterance verification for connected digits recognition", In EUROSPEECH-1995, 529-532.