论文部分内容阅读
针对基本反复模型音乐分离方法自适应性差的问题,提出一种基于美标度倒谱系数(MFCC)的多反复结构模型的音乐分离方法。首先,提取出音乐信号的MFCC系数矩阵(39维的数据构成);然后利用余弦特性得到其相似矩阵,进而将相似度一致的片段划分到一起,建立不同的反复结构模型;之后结合理想二元掩蔽(]BM)分离出背景音乐及歌声的频谱,相应的时域信号则由傅里叶逆变换获得;最后,在不同类型、长度的音乐文件上测试了算法性能,将提出的算法与Rafii的反复算法和Ozerov的灵活窗非负矩阵分解方法进行对比。实验结果表明,改进方法在分离性能上最高提高3 dB左右,并且对于曲调变换大的音乐提高效果更为明显,从而证实了改进方法是一种有效的音乐分离方法,并且更具稳定性。
Aimed at the poor adaptability of music separation method based on the basic repetitive model, a music separation method based on MFCC is proposed. Firstly, the MFCC coefficient matrix (39-dimensional data structure) of the music signal is extracted; then the similarity matrix is obtained by using the cosine property, and the fragments with the same degree of similarity are grouped together to establish different repetitive structure models; (BM) to separate the spectrum of background music and song, the corresponding time-domain signal is obtained by inverse Fourier transform. Finally, the performance of the algorithm is tested on music files of different types and lengths. The proposed algorithm is compared with Rafii Iterative algorithm and Ozerov flexible window non-negative matrix decomposition method for comparison. The experimental results show that the improved method can improve the separation performance up to about 3 dB, and the effect of music enhancement on the large melody transform is more obvious, which proves that the improved method is an effective music separation method and more stable.