12dim VAE2Chords. Miles stellar library

Warning! The sonification provided by this prototype is highly data-dependent. Be aware of your headphones or speaker levels.

Demonstration video of VAE2Sound. The prototype allows stellar spectra-driven latent space sonification using variational autoencoders. This video shows the multimodal exploration of 25 spectra of the Miles stellar library from the Spanish Virtual Observatory. In this design, the 12 values of each autoencoder’s latent space vectors, are mapped to the fundamental frequencies of 12 independent sine generators. Each of these oscillators is sent to a loudspeaker to allow the acoustic spatialization of the representation in a surround sound system formed by 12 loudspeakers. A stereo bounce is used in the video for the demonstration. The graphical display provides the original spectrum (left), the recovered spectrum from the sonified latent vector (right), and a representation of the active speakers.

data: http://svocats.cab.inta-csic.es/miles/index.php?action=credits