Skip to main content
Fig. 6 | EURASIP Journal on Audio, Speech, and Music Processing

Fig. 6

From: Stripe-Transformer: deep stripe feature learning for music source separation

Fig. 6

Visualization of deep stripe feature learning results. Two song clips are taken from the Musdb18 test set. Magnitude spectrograms of the mixture are presented, in which four different colors represent the four constituent sources. On the left and bottom sides of each spectrogram, there are score bars which represent the average attention scores of horizontal and vertical stripes, respectively. The higher the score is, the greater attention the network pays to the position

Back to article page