Maximum likelihood constrained adaptation for multichannel audio synthesis

Multichannel audio environment can immerse a group of listeners in a seamless aural environment. Previously, we proposed a system capable of synthesizing the multiple channels of a virtual multichannel recording from a smaller set of reference recordings. This problem was termed multichannel audio resynthesis and the application was to reduce the excessive transmission requirements of multichannel audio. In this paper, we address the more general problem of multichannel audio synthesis, i.e. how to completely synthesize a multichannel audio recording from a specific stereophonic of monophonic recording, significantly enhancing the recording's quality. We approach this problem by extending the model employed for the resynthesis problem.