NeuralFunk, or This break does not exist. We trained WaveGAN neural network on 5000 vintage drumloops (breaks) from around 1500 recordings. All loops were fit to 120bpm tempo and sampled at 16384Hz. The trained model is able to generate non-existing breaks and smooth transitions between them. This is our early result. We're about to improve both the dataset and the model. Project by @ @ The result on the video was obtained by moving a point slowly through
Hide player controls
Hide resume playing