ISDEAI: A method for assigning error and defect in PSC test corpus

This paper proposes a method to assign the on-the-spot scores of the syllable level to the initial, final and tone level in PSC Test Corpus by integrating the statistic data with a small amount of experts' annotation information (ISDEAI method hereafter). Comparing with the result of previous data-driven statistic method (DDS), the results of ISDEAI are more consistent with the actual pronunciation features of the provincial Putonghua in China. This method can be applied to assign the errors and defects semi-automatically in PSC Test corpus and thus benefit to the research for establishing a large-scale speech corpus with the pronunciation errors annotations.