I’ve published a selection of recordings from an experimental generative setup written in Pure Data that works with two neural audio/ AI model architectures RAVE and vschaos2. Both are courtesy of IRCAM and can be used in realtime setups using the nn~ object.
The models have been trained on a dataset that contains parts of my own body of work. Titled “Saatgut Proxy”, the release is available via Bandcamp.
Track “Nausea Seeder” uses vschaos2 only in a similar setup as shown in a screencast posted earlier. “Tritt Nochmal Zu” comes with additional sound generators outside the audio generated by the models.
The patch setup and configuration that led to the recordings can be seen in the following screen video:
Sources
RAVE is “A variational autoencoder for fast and high-quality neural audio synthesis” created by Antoine Caillon and Philippe Esling. arxiv.org/abs/2111.05011
vschaos2 “is a vintage-flavoured neural audio synthesis package. It is based on unsupervised / (semi-)supervised training of spectral information using variational auto-encoders” created by Axel Chemla Romeu Santos. hal.science/tel-03543235
github.com/acids-ircam/RAVE
github.com/acids-ircam/vschaos2
github.com/acids-ircam/nn_tilde