论文部分内容阅读
翻唱歌曲识别是音乐相似计算研究中的问题之一。通过对表征音乐和声的Chroma特征的分布分析,进行了基于近似子乐句的翻唱歌曲识别研究;在已收集的翻唱歌曲数据集上,完成了歌曲调的归一化处理、结构分割和特征抽取;比较了特征降维后,在EMD距离矩阵、欧式距离矩阵和余弦距离矩阵等不同测度下的音乐相似度计算。通过与MIREX中翻唱歌曲识别结果的比较表明,本研究中基于近似子乐句的研究方法在准确率上有一定程度的提升。
Cover song recognition is one of the problems in the study of music similarity calculation. Through the analysis of the distribution of Chroma features that characterize the harmony of music and music, the cover song recognition based on approximate sub-phrase is studied. On the collected cover song dataset, the normalization of the tune, structure segmentation and feature extraction are completed The similarity of EMD distance matrix, Euclidean distance matrix and cosine distance matrix are compared after dimensionality reduction. The comparison with the result of song recognition in MIREX shows that the research method based on approximate sub-phrase in this study has a certain degree of improvement in accuracy.