Independent Deeply Learned Matrix Analysis for Multichannel Audio Source Separation

• DOI: 10.48550/arxiv.1806.10307
• Publication Date: 2018-01-01
• Authors: Shinichi Mogami; Hayato Sumino; Daichi Kitamura; Norihiro Takamune; Shinnosuke Takamichi; Hiroshi Saruwatari; Nobutaka Ono
• Keywords: Source Separation; Matrix (chemical analysis); Deep Neural Networks

In this paper, we address a multichannel audio source separation task and propose new efficient method called independent deeply learned matrix analysis (IDLMA). IDLMA estimates the demixing in blind manner updates time-frequency structures of each using pretrained deep neural network (DNN). Also, introduce complex Student's t-distribution as generalized generative model including both Gaussian Cauchy distributions. Experiments are conducted music signals with training dataset, results show validity proposed terms accuracy computational cost.