Combining of confidence measures under Bayesian framework for speech recognition

Bayesian combining of confidence measures is proposed for speech recognition. Centralized and distributed schemes are considered for confidence measure combining under a Bayesian framework. Centralized combining is a feature level fusion which combines the values of individual confidence scores and makes a final decision. In contrast, distributed combining is decision level fusion which combines the individual decision makings made by each individual confidence scoring method. Both methods are basically based on the statistical modeling of confidence features. In addition, adaptation of the confidence score using statistical models is also presented. The proposed methods reduce the classification error rate by 17% from conventional single feature based confidence scoring method in an isolated word out-of-vocabulary rejection test.