Explore machine learning magic in this casual simulation that turns over 44,000 Freesound audio samples into a navigable 3D sound cloud. Players sift through the dataset's t-SNE embeddings, pick standout clips, and replay them to build custom instruments like pianos from synth notes or marimbas from pops and crackles.
Simple mechanics drive endless creativity: move across the vast 3D map, select audio to save, and trigger playback for rhythmic fun. The indie experiment reveals the computer's grasp on sound data, perfect for explorers who savor discovery over construction. Audio enthusiasts and casual tinkerers find a playground in its sheer scale and sonic variety.
Feel the magic of Machine Learning in this new experiment that combines over 44,000 sounds from Freesound, a data set created by Kyle McDonald and a touch of Cabbibo weirdness.
Run your hands through a cloud of sound, learn the shape of the computers mind, select your favorite audio delicacies and make a drum set from this decadent aural smorgasbord.
The mechanics may be simple ( navigate through a 3D map of sounds , select individual sounds to keep, and hit those sounds again to play them ), but because of absolutely massive size of the data set, the possibilities are endless!
Want to create your own piano out of synth notes? Go ahead!
How about a marimba of pops and crackles? Be my guest!
And if you are just the exploring type, the data is beautiful enough to begin with! There is no need to make an instrument, or play it, because in from of you lies a tool for discovering and understanding a beautiful algorithm, and even more beautiful data: Sound
This project would not be possible without the amazing community of FreeSound.org
As well as Kyle McDonald, who scraped audio from freesound.org & made t-SNE embeddings of the data!
Created during Unity Artist in Residency 2017
