Freesound Loops 4k (FSL4)

This dataset contains ~4000 user-contributed loops uploaded to Freesound. Loops were selected by searching Freesound for sounds with the query terms <em>loop</em> and <em>bpm</em>, and then automatically parsing the returned sound filenames, tags and textual descriptions to identify tempo annotations made by users. For example, a sound containing the tag <em>120bpm</em> is considered to have a ground truth of 120 BPM. This dataset is part of the publication: Font, F., &amp; Serra, X. (2016). "Tempo Estimation for Music Loops and a Simple <br> Confidence Measure". In Proceedings of the Int. Conf. on Music Information Retrieval (ISMIR). The dataset contains a number of raw audio files in different formats (wav, aif, flac, mp3 and ogg), and a <em>metadata.json</em> file with metadata about the audio files. Audio files are all contained in the same directory and use Freesound IDs as filenames (e.g. 1234.wav, where the Freesound ID is 1234). The <em>metadata.json</em> file contains a dictionary with keys corresponding to Freesound IDs (note that the <em>.json </em>extension had to be removed because of Zenodo's limitations). Tempo (BPM) annotations can be found in the <em>annotations </em>property of each entry in the dictionary. The BPM ground truth collected from user annotations is typically correct, but we observed that around 5% of the cases could be wrong or very ambiguous annotations. For research using loops collected from Freesound, <strong>we recommend using this newer dataset (link to be added soon) </strong>which is more comprehensive and in which all tempo annotations have been manually verified. Note that even though the dataset metadata is released under CC-BY-4.0 license, the individual audio files have their own Creative Commons licenses. Licenses for these files are listed both in the <em>metadata.json</em> and <em>about.txt</em> files.