The classification of performance skills of national vocal music is an existing problem in music research and education, and the traditional manual classification method is time-consuming and labor-intensive. To solve this problem, this article studied the automatic classification system of performance skills of national vocal music. Specifically, it is to select audio samples from the MTG-QBH dataset for audio preprocessing, feature extraction, and training and testing of classification models. In the audio preprocessing stage, spectrum subtraction is used for denoising, data standardization, mute processing and data enhancement to ensure the consistency and quality of audio data. MFCC method is used to extract audio features and convert audio signals into feature vectors for classification. Based on these feature vectors, SVM model is used for training and classification, and different vocal skills are accurately identified. The experimental results show that the system has excellent performance in classification accuracy, precision, recall rate and F1 score, all reaching about 0.95. The experiment shows that the system can effectively classify various ethnic vocal techniques, improving classification efficiency and accuracy.