The accuracy of this algorithm has continuously improved over time with the development of our flagship project, called “AI DJ”. By letting deep learning based AI model “listen” to a library of music, we have succeeded in quantifying the relevance of each song, automating the song selection know-how by professional DJs.
Playlist generation without relying on metadata by indexing arbitrary music libraries using deep learning models
Playlist content can be dynamically generated and changed based on previously played music content
Proven in floor performances around the world. Tasteful music selection that even professional DJs unanimously approve of
Automatic playlist generation
Based on the currently playing song, subsequent playlists can be automatically generated to continue playing background music in a natural flow.
Can be used as a recommendation engine for producers by enabling intuitive indexing and search through their collection.
This technology is used in the AI BGM selection function in the U Music terminal provided by USEN, which automatically selects music according to the store environment, time of day, and season.
USEN – U MUSIC (https://iot.usen.com/u-music/)
To index the music library for song selection, we use our proprietary deep learning model to create a vector representation of the content so that musically similar songs appear in close proximity. When selecting the next song based on the currently playing song, the system searches for the optimal song in this vector space. The selection can be guided using metadata such as playlist themes, environmental variables, sentiment, and time of day.
License term: Monthly Developer's license: Yes
Input: Playback information for the past several songs Output: Songs recommended by the algorithm as continuation
Cloud computing: Standard API provided On-premise environment: Possible by consultation