JAIST Repository >
School of Information Science >
Articles >
Journal Articles >

Please use this identifier to cite or link to this item: http://hdl.handle.net/10119/15477

Title: Weighted Robust Principal Component Analysis with Gammatone Auditory Filterbank for Singing Voice Separation
Authors: Li, Feng
Akagi, Masato
Keywords: Singing voice separation
Robust principal component analysis (RPCA)
Gammatone auditory filterbank
IBM/IRM estimation
Issue Date: 2017-10-26
Publisher: Springer
Magazine name: Lecture Notes in Computer Science
Volume: 10639
Start page: 849
End page: 858
DOI: 10.1007/978-3-319-70136-3_90
Abstract: This paper presents a proposed extension of robust principal component analysis (RPCA) with weighting (WRPCA) based on gammatone auditory filterbank for singing voice separation. Although the conventional RPCA is an effective method to separate singing voice and music accompaniment, it makes some strong assumptions. For example, drums may lie in the sparse subspace instead of being low-rank, which decreases the separation performance in many real-world applications, especially for drums existing in the mixture music signal. Accordingly, the proposed WRPCA method utilizes different weighted values between sparse (singing voice) and low-rank matrices (music accompaniment). In addition, we developed an extended RPCA on cochleagram using an alternative time-frequency (T-F) representation based on gammatone auditory filterbank. We also applied IBM/IRM estimation to improve the separation results. Evaluation results show that WRPCA achieves better separation performance than the conventional RPCA, especially for the IBM estimation method.
Rights: This is the author-created version of Springer, Feng Li and Masato Akagi, Lecture Notes in Computer Science, 10639, 2017, 849-858. The original publication is available at www.springerlink.com, http://dx.doi.org/10.1007/978-3-319-70136-3_90
URI: http://hdl.handle.net/10119/15477
Material Type: author
Appears in Collections:b10-1. 雑誌掲載論文 (Journal Articles)

Files in This Item:

File Description SizeFormat
2749.pdf292KbAdobe PDFView/Open

All items in DSpace are protected by copyright, with all rights reserved.


Contact : Library Information Section, Japan Advanced Institute of Science and Technology