This code is a two channel implementation of DeepConvSep architecture.
A new training stage for binaural mixtures has been included: it
compares Interaural Level Differences (ILDs) from both channels to
enhance spatial stability in the final estimates.
This code is a two channel implementation of DeepConvSep architecture. A new training stage for binaural mixtures has been included: it compares Interaural Level Differences (ILDs) from both channels to enhance spatial stability in the final estimates.