摘要
面向MP3音乐的分类方法是利用MP3编解码的特点,将MP3音乐文件表示成特征向量,采用常用机器学习分类方法对音乐文件进行分类。重点对MP3音乐特征片段提取和分类方法进行讨论,提出基于离散余弦变换(MDCT)系数域3种特征参数的特征片段提取方法和基于LCS(学习分类器)的音乐分类方法。实验表明,特征片段提取方法能够在最短时间内找到最具有"特征"的特征片段,从而缩小了匹配时间,因此LCS分类方法提高了分类方法的命中率。
MP3-oriented music classification system makes use of the characteristic coefficients in the process of MP3 encoding/decoding. Focusing on characteristic segment method and classification method, the music segmentation method based on the feature parameters of discrete cosine transform domain and the music classification method based on LCS were proposed. The experiment shows that the music segmentation method can find the most featured segmentation in the shortest time, and the LCS makes the prediction more exact than other classification methods.
出处
《重庆邮电大学学报（自然科学版）》
2007年第4期417-421,共5页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
关键词
MP3
特征提取
音乐分类
学习分类器
MP3
feature extraction
music classification
learning classifier （LCS）
