[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[AUDITORY] Slakh2100: new dataset with 145 hours of musical mixtures with isolated sources and MIDI



Hi all,

On this New Music Friday I'm excited to announce Slakh2100! Slakh2100 is a new dataset that contains 2100 songs, totaling over 145 hours of musical mixture data. Each mixture also comes with isolated sources and MIDI data. Slakh is synthesized by using professional-grade sample-based synthesizers to render 2100 files from the Lakh MIDI Dataset by Colin Raffel. As you can guess, there are no vocals (hard to synthesize :-P), but there are many, many more instrument classes than have existed in other datasets. We hope everyone finds this dataset useful!

For more information see our preprint here: https://arxiv.org/abs/1909.08494

Here's the landing page for the site (downloads, scripts, analysis, benchmarks): www.slakh.com

Thanks!
-Ethan Manilow, Gordon Wichern, Prem Seetharaman, and Jonathan Le Roux