Create Beat Saber Song Mappings with @beatsage_ai #MachineLearning #ArtificialIntelligence #BeatSaber @Denizen_Kane @chrisdonahuey @Naysy

@naysy playing a custom Beat Saber mapping by Beat Sage AI. Check out the video –> https://youtu.be/nDZ61cRBhzU

 

Last month @Denizen_Kane (Abhay Agarwal) and @chrisdonahuey announced the release of Beat Sage. This web service utilizes machine learning to create custom song mappings for Beat Saber (“mapping” means putting the blocks and obstacles to the music). Beat Saber mappings are typically generated by “mappers” and are only available for select songs. Beat Sage allows you to create custom mappings for any song on YouTube.

The underlying AI builds off of @chrisdonahuey’s earlier work with Dance Dance Revolution. In general, it utilizes two neural networks to place the blocks and the directionals for those blocks:

Beat Sage uses two neural networks to map an audio file into a plausible Beat Saber level. These neural networks were trained on Beat Saber levels created by humans. The first neural network listens to the audio and predicts at what points in time blocks should be placed. The second neural network looks at the predicted timings and maps each to a timestamp to a block type (e.g. red up, blue down, red up + blue down).

Although there are a few limitations to the current Beat Sage offerings, it hasn’t stopped the community from enjoying it! As of the last count, Beat Sage has been used to generate over 50k custom mappings for Beat Saber.

If you’d like to get involved with the Beat Sage community check out their Discord or subreddit. You can also support their work on Patreon. If you’d like to see Beat Sage in action, @Naysy did a fantastic review on YouTube. If you’d like to see some related work, check out our post on @chrisdonahuey’s LakhNes.

 

Written by Rebecca Minich, Product Analyst, Data Science at Google. Opinions expressed are solely my own and do not express the views or opinions of my employer.