Supervised and Unsupervised Discovery of Latent Directions for Drum Synthesis

View the source code on GitHub aframires/stylegan2-ada-pytorch

Supervised and Unsupervised Discovery of Latent Directions for Drum Synthesis

António Ramires, Jordan Juras, Julian D. Parker, Xavier Serra

Random Generated Sounds by the GAN

In this section, we show some random sounds generated by the trained GAN network. These were randomly selected and not cherry picked examples.

Latent Space Interface

Generating Drums in the Z-Space

The first interface presented to the users for generating drums is the 512 dimensional latent Z-Space.

We show examples of the interface along with the sounds they produce:

ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples
ZSpace Examples

Modifying a Seed with SeFa

The second interface presented to the users for generating drums is the 7 directions obtained from the SeFa algorithm.

We show examples of modifying each of the first 5 directions, from -80 to +80:

1st Direction

2nd Direction

3rd Direction

4th Direction

5th Direction

Modifying a Seed with InterfaceGAN

The third interface presented to the users for generating drums is the 7 directions obtained from the InterfaceGAN algorithm.

We show examples of modifying each of the first 5 directions, from -80 to +80:

Hardness

Depth

Brightness

Roughness

Warmth