GAN Comparative Study

Generative Adversarial Networks
for Audio Generation: A Comparative Study

Ernest Vanmosuinck, Antonio Mone,
Alessandro Marincioni, Adetunji Ayodele

IRMAS Dataset

Training Samples

Sample 1 - voice

Sample 2 - electric guitar

Sample 3 - saxophone

Sample 4 - trumpet

Sample 5 - piano

Sample 6 - flute

WaveGAN

10k steps

Sample 1

Sample 2

Sample 3

Sample 4

Sample 5

Sample 6

SpecGAN

10k steps

Sample 1

Sample 2

Sample 3

Sample 4

Sample 5

Sample 6

20k steps

Sample 1

Sample 2

Sample 3

Sample 4

Sample 5

Sample 6

20k steps

Sample 1

Sample 2

Sample 3

Sample 4

Sample 5

Sample 6

30k steps

Sample 1

Sample 2

Sample 3

Sample 4

Sample 5

Sample 6

30k steps

Sample 1

Sample 2

Sample 3

Sample 4

Sample 5

Sample 6