Improved spontaneous Mandarin speech recognition by disfluency interruption point (IP) detection using prosodic features

In this paper, a new approach for improved spontaneous Mandarin speech recognition with disfluencies well considered is presented. The basic idea is to detect the disfluency interruption points (IPs) prior to the recognition, and then to use these information during rescoring in the recognition process. For accurate detection of disfluency interruption points (IPs), a whole set of new features were proposed and tested by carefully considering the special characteristics of Mandarin Chinese. A new approach of incorporating the decision tree into the maximum entropy model training was also developed to enhance the IP detection accuracy. Experimental results indicated that the proposed set of features and the IP detection approach were very useful, and the obtained information about disfluency actually benefited the speech recognition performance.