| 1. | This feature perfectly combine the frequency in acoustics level and the temperament in music semantic level , we use the cosine distance of this feature to represent the similarity of two music clips , then we design a group of algorithms that is inspired from the thought of edit distance and dynamic programming . they segment the feature vectors into groups at first , then through group similarity match , group recurrent detect , merge recurrent group and structure label joined algorithms to complete the music structure label task . because this is a really new field of research and no good method of evaluation had been finding , we propose a new evaluation method and the results of the experiments show that it is a good method
然后设计了一组源于编辑距离和动态规划思想的音乐结构分析算法,首先将特征向量分组,然后经过组相似匹配、组重现检测、重现组归并和自动标注四个前后衔接的环节实现了音乐结构的自动标注,较好地实现了将音频形式的音乐自动标注为表示音乐结构的三元组列表形式,由于这是一个新的领域,目前还没有比较好的量化评价方法,本文提出一种新的评价方法,并用它来评价结构分析的结果,取得了较好的效果。 |
