Subband-based upmixing of stereo to 5.1-channel audio signals using deep neural networks

摘要

In this paper, we propose a subband-based stereo to 5.1-channels upmixing method using deep neural networks (DNNs) in MPEG-H 3D audio framework. In the training stage, DNN models of rear and center channels are respectively trained by using log-spectral magnitudes of quadrature mirror filter (QMF) sub-bands. In the upmixing stage, stereo input signals are converted into rear and center channels by feed-forward decoding with the trained DNN models. The performance of the proposed method is evaluated using both objective and subjective measures and it is compared with those of conventional methods. Consequently, the proposed method outperforms the conventional methods.