Timbre Space
Real-time Deep Learning tools for experimental electronic musicby Kıvanç Tatar, Daniel Bisig, and Philippe Pasquier
The model learns from a dataset of audio files and synthesizes sounds in real-time. The model let users to explore a universe of timbre that it learns from the dataset. More details of the Deep Learning framework is upcoming in a few weeks.
Initial examples
Here are our initial generations. Original 1 and Original 2 are the excerpts of original samples. Well, these two tracks are actually generated only using the original magnitude spectrums, and phase is added after using a reconstruction technique. Likewise, our Deep Learning model generates only the magnitude spectrum, and phase is added later using a reconstruction technique. Hence, the original 1 and original 2 are our baselines, and they are the ideal qualities that the Deep Learning model aims to achieve during the training.
Reconstructions-> X Interpolations 0.0 and X Interpolations 1.0 are reconstructions of the original audio files using the Deep Learning model, original 1 and 2 respectively.
Timbre Interpolations-> X Interpolations 0.1 means that this sample is generated using 90% of original 1 timbre and 10% of original 2 sample. Think 0.1 as a slider value...
Timbre Extrapolations-> X Extrapolations 1.1 means that we are drawing an abstract line between timbre example 1 and 2, and then following the direction of that line, we are moving further away from the timbre 2 by 10%. X Extrapolations -0.1 means we are drawing a line between timbre 2 and 1, and moving further away from timbre 1 in that direction by 10%.
Acknowledgements
This work has been supported by the Swiss National Science Foundation, and Social Sciences and Humanities Research Council of Canada.
Ce travail est supporté par le Fonds national Suisse de la recherche scientifique, et le Conseil national des sciences humaines et sociales du Canada.
