This paper proposes an extension of robust principal component analysis (RPCA) with weighted values for monaural singing voice separation.Although the conventional RPCA is an effective method to separate singing voice and music accompaniment from the mixted audio sig- nal, it fails when one singular value is much larger than all others. For example, drums may lie in the sparse subspace instead of being lowrank, which lead that the separation performance is decreased in many real world applications, espe- cially for drums existing in the mixture music sig- nal. Therefore, in order to solve this problem, we utilize different weighted values between sparse (singing voice) and low-rank matrices (music ac- companiment). Evaluation results on ccMixter and DSD100 datasets show that the proposed method achieves better separation performance than the conventional RPCA.
Singing Voice Separation and Vocal F0 Estimation Based on Mutual Combination of Robust Principal Component Analysis and Subharmonic Summation
