In this paper, an automatic assessment system has been developed for Putonghua articulation and pronunciation assessment. The framework of developing the system consists of designing the vocabularies respectively for the two assessment parts with standard scoring regulations, constructing the speech recognition module which makes use of a HMM based acoustic model with 39 parameters including logarithmic energy, the previous 12 MFCCs and their first and second difference extracted from the records of speech and the Speech SDK which is a developing kit from Microsoft to improve the accuracy, and building disorder diagnosis module according to the clinical requirement. In order to measure accuracy of this objective method, a comparative experiment is carried out with the subjective one, the results show that the two methods are almost the same and have a extremely correlation in both the articulation and pronunciation assessments.
展开▼
