The reverberation is an acoustic phenomenon which reduces the intelligibility of the voice recordings and the quality of the audio recordings. For this reason, new works have been made to remove the reverberation of the audio recordings, the process of dereverberation.
In the last years, Deep Learning technics have begun to be used with promising results. This work continues this line using the architecture Demucs, which is designed to separate music tracks, to implement the new task of separate the original audio, which comes from the audio source, from the reverberation created by the room.
You can test it here https://colab.research.google.com/drive/1DA4hHL1KahFxTa3vs6k8I1ycBk5ZMCsy