← Back to the audio demoAudio credits
The original mixture and all nine processed versions use the same four recordings from Freesound, via FSD50K.
What changed
Converted to mono 16 kHz, cropped or padded, adjusted in level and mixed. Processed variants use the trained model and stated target/remainder gains. One shared playback scale is applied to the original mixture and all nine outputs. Original recordings retain their listed licenses.
Adaptation and model processing: Eduardo Cepeda Torres, 2026. This audio adaptation is offered under CC BY 3.0; the CC0 source retains its dedication. Attribution does not imply endorsement by the original creators.
About this example
A synthetic aircraft warning, a dog bark, a door knock, and running water. The aircraft warning is the source labeled as the siren class in the study. These are precomputed estimates; other sounds and artifacts can remain.
A single playback factor of 1.00000000 is shared by every file. There is no independent normalization between modes.
Audio provenance, transformations and file hashes · Study-wide results · Research report