Music source separation is important for applications such as karaoke and remixing. Much of previous research
focuses on estimating magnitude short-time Fourier transform (STFT) and discarding phase information. We observe that,
for singing voice separation, phase has the potential to make considerable improvement in separation quality. This paper
proposes a complex-domain deep learning method for voice and accompaniment separation. The proposed method employs
DenseUNet with self attention to estimate the real and imaginary components of STFT for each sound source. A simple ensemble
technique is introduced to further improve separation performance. Evaluation results demonstrate that the proposed method
outperforms recent state-of-the-art models for both separated voice and accompaniment.