Research of Korean Speech Identification Based on Rhythm Feature

In this paper, an automatic Korean speech identification method was proposed for multi-language speech data environment including Korean, Chinese, Russian and English spoken data. Firstly, shifted delta coefficients of speech pitch were extracted and applied as input feature for the training of classifiers. Secondly, SVM was used as classifier between Korean and non-Korean speech, and the results of three continuous votes were adopted to decide the final result and calculate the accuracy rate. Experimental results show that the proposed method could solve the issue of Korean speech identification effectively.