Fig. 4From: Time-domain adaptive attention network for single-channel speech separationVisualization results of the spectrogram of the separated samples. First row: spectrogram of mixed speech. Second row: spectrogram of clean speech. The fourth, fifth, sixth, and seventh rows represent the separation results by using baseline, local, global, and local and global attention networks, respectively, and then subtracting the spectrogram of clean speechBack to article page