Abstract: A novel mapping of embedding music to Rn space and a new kernel-based classification method are presented.First，analyzing melody contours from a huge music dataset，it translats music datasets to strings with an alphabet including only eight letters.Second，three string kernels are obtained by mapping music strings to high dimension Rn space using contiguous substrings embedding method.Finally，performances of three string kernels on five music datasets are evaluated by SVM algorithm and ROC graph.