中文摘要：

目前音乐分类已成为分类算法的重要应用领域,但有关民歌的分类研究比较少。以云南原生态民歌为数据对象,提取11维的CELP音频特征,采用单一分类器和集成分类器Bagging、AdaBoost、MCS及随机森林对不同标记训练样本数据进行实验比较分析。数据表明Ada Boost和MCS的性能优于单一基分类器,其中AdaBoost表现最佳,对5个民族的歌曲识别正确率都达到85%以上。实验结果表明集成分类器对云南民族歌曲能进行有效的识别与分类。

英文摘要：

Nowadays music classification has become an important application field for classification algorithms, there is few research about folk songs classification. Uses the original folk songs in Yunnan Province as research objects, extracts their CELP audio characters on 11 dimensions scale, compares and analyzes the classified experimental data from different proportion of marked training samples, which processed by single classifier and ensemble classifier（such as Bagging, Ada Boost, MCS etc.）. Among them, the Ada Boost is the most efficiency, and the accuracy rate is over 85%.The result shows the ensemble classifiers can mean the effectiveness of the study methodology.