| Paper ID | E-3-3.5 |
| Paper Title |
JOINT-DIAGONALIZABILITY-CONSTRAINED MULTICHANNEL NONNEGATIVE MATRIX FACTORIZATION BASED ON MULTIVARIATE COMPLEX STUDENT'S T-DISTRIBUTION |
| Authors |
Keigo Kamo, Yuki Kubo, Norihiro Takamune, The University of Tokyo, Japan; Daichi Kitamura, National Institute of Technology, Kagawa Collage, Japan; Hiroshi Saruwatari, The University of Tokyo, Japan; Yu Takahashi, Kazunobu Kondo, Yamaha Corporation, Japan |
| Session |
E-3-3: Advanced Signal Processing and Machine Learning for Audio and Speech Applications |
| Time | Thursday, 10 December, 17:30 - 19:30 |
| Presentation Time: | Thursday, 10 December, 18:30 - 18:45 Check your Time Zone |
|
All times are in New Zealand Time (UTC +13) |
| Topic |
Speech, Language, and Audio (SLA): Special Session: Advanced Signal Processing and Machine Learning for Audio and Speech Applications |
| Abstract |
In this paper, we propose the model generalization of a fast version of multichannel nonnegative matrix factorization (FastMNMF). FastMNMF is a blind source separation (BSS) method under the assumption that the spatial covariance matrices of multiple sources are jointly diagonalizable. To further improve its source-separation performance, we introduce a multivariate complex Student’s t-distribution as a generative model, which includes a multivariate complex Gaussian distribution used in conventional FastMNMF. We derive a new parameter update rule using the auxiliary-function-based method and show the validity of the proposed method on the basis of BSS experiments using music sources. |