Confidence Measures and Integrating Algorithm in Utterance Verification
语音识别确认中的置信特征和判定算法

In this paper, an approach is presented for integrating several confidence measures to verify utterance based on support vector machine (SVM). Segmental filler-based posterior probability parameters and linear predictive coding (LPC) recognition difference measure are derived from the verified utterance. An SVM classifier is trained to integrate several confidence measures to make final decision. Experimental results show that confidence measures and the SVM classifier are effective for utterance verification.