The AudioObservatory (2019)
A system that takes a large number of audio files, analyzes them for a number of feature trajectories, computes distance matrices based on these features (using dtw for example), and then creates a weighted combination of these to use in embedding the audio content in a 3D space. This combination can be updated to match user interest, resulting in a reorganization of the material It also defines a number of interactions to explore the space, including building probabilistic paths to guide the user navigation based on the user's choices, and provides multimodal input methods (e.g. face-tracking, LeapMotion, MIDI controllers, etc.) to quickly have local and global access to sounds within the source collection.