What AI models and data does Songbox use?
Every model and dataset behind our AI, with its source and licence.
We think you should be able to see exactly what powers our AI, so here is the full list. Everything we use is open and permissively licensed, and the two classifiers we trained ourselves (one for instruments, one for genre) learned only from music that artists released under an open licence for this purpose.
- LAION-CLAP, for genre, mood and instrument understanding. Released by the non-profit LAION under a public-domain licence (CC0).
- Beat This!, for tempo and beat detection. From the Institute of Computational Perception at Johannes Kepler University Linz, under the MIT licence.
- librosa, for musical key detection. An open audio library under the ISC licence.
- OpenMIC-2018, the openly licensed dataset our instrument detector learned from: 20,000 Creative Commons clips (CC-BY 4.0).
- Free Music Archive, the openly licensed collection our genre detector learned from: tracks that artists chose to share under Creative Commons.
We even passed on more accurate models because their licences were restrictive. We would rather be slightly less clever and completely clean.
Was this article helpful?
Thanks for your feedback!
