Discovering new music is difficult, making it a frustrating experience for both listeners and services. Identifying what one person liked about a specific song is a challenge when music is so subjective. Two different people may love the same song, but for different reasons that affect their wider tastes. In an attempt to improve the situation, Danning Liang and Artem Laptiev from MIT’s School of Architecture and Planning built a kind of AI-powered boombox called VBox that helps listeners discover music in a new way.
Most existing services use some combination of listener data and qualitative categorization of songs to aid in music discovery. But those connections are obvious and tend not to identify the factors that actually predict a listener’s enjoyment of a song. Artificial intelligence models, on the other hand, excel at finding connections and patterns that we might not see ourselves. In this case, VBox uses OpenAI’s natural language models to categorize music and find similar songs. As a song plays, VBox will list keywords related to the music. If a specific keyword resonates with the listener, they can select it to influence the next song choice.
There aren’t a lot of technical details available, but we do know that an Arduino board is somewhere in the mix. It uses RFID to identify genre cards that start the music discovery process. The keywords scroll across a LED matrix display and a crank handle attached to a rotary encoder lets listeners move through the keyword list. The enclosure is made of gorgeous stamped sheet metal with a leather strap and some 3D-printed internal framework for the electronic components. Music pumps out through a pair of small speakers.
