FSDKaggle2018 dataset: 11k audio clips and 18h of training data unequally distributed in 41 classes of everyday sounds. It was collected for the DCASE2018 Challenge Task2: General-purpose tagging of Freesound audio with AudioSet labels. The dataset is described in our paper General-purpose tagging of Freesound audio with AudioSet labels: task description, dataset, and baseline.
FSDnoisy18k: a dataset featuring 20 classes and 42.5 hours of audio for the investigation of label noise in audio. Submitted to ICASSP2019, will be available in the coming days. In the meantime, you can check its companion site for a detailed dataset description.
FSD: This is the first and main dataset being collected with the Freesound Datasets platform. It is currently work-in-progress and we hope to make the first release in early 2019. FSD will be a large-scale, general-purpose audio dataset consisting of audio clips from Freesound organised following the AudioSet Ontology.