Improving the Performance of Out-of-vocabulary Word Rejection by Using Support Vector Machines

Support Vector Machines (SVM) represents a new approach to pattern classification developed from the theory of structural risk minimization [1]. In this paper, we propose an approach to improve the performance of confidence measurements for out-of-vocabulary word rejection by using SVM. Confidence measures are computed from the information of n-best candidates and anti-word by a Hidden Markov Model (HMM) based speech recognizer. The acceptance/rejection decision for a word is based on the confidence score which is provided by SVM classifier. And the decision is performed for each word in vocabulary separately. The performance of the proposed SVM classifier is compared with method based on posterior probability and anti-word probability. Experiments of Mandarin command recognition have showed that better performance can be obtained when using the proposed method.