论文部分内容阅读
针对目前学术界主流的基于内容提取音乐结构开销过大和产生冗余碎片的问题,提出通过歌词分析对音乐结构进行自动提取,同时结合基于内容的SVM分类,以歌曲节拍为单位进行有无人声的判别,进一步对歌词提取的边界进行校正,精确了音乐结构的边界点。在相同数据集上的实验中,这种歌词与内容分析相结合的音乐结构提取算法对桥段和尾奏的分析准确率分别提高了9%和11%。
In view of the current mainstream problem of extracting music structure based on content overhead and generating redundant fragments, this paper proposes that the structure of music should be extracted automatically by lyric analysis. At the same time, based on content-based SVM classification, Discrimination, to further refine the boundaries of the lyrics extraction, the precise point of the music structure of the border. In the experiment on the same dataset, the accuracy of the analysis of the bridge segment and the tail is improved by 9% and 11%, respectively, for the music structure extraction algorithm combined with the content analysis.