The multi-channel AR model for real-time audio restoration

In this paper we propose a multi-channel autoregressive (AR) model that can be applied to real-time audio restoration. The model is built on the assumption that redundant audio information exists in independent multi-channels and a single corrupted channel can be modeled as linear combinations of scaled time shifts of other channels. The new model has similar computational complexity as the single-channel AR model, but is capable of using lower fixed orders to restore longer sections of audio segments without audible distortion. The model can apply to psychoacoustic motivated techniques as well as source-separated audio.