Researchers at ETH Zurich have created a dataset of over 15.3 million songs from over 400,000 musicians. The dataset includes audio and text embeddings, links to a YouTube video and Spotify preview of each song, the track names on YouTube and Spotify, video view counts, video descriptions, and release dates, as well as comparisons between the YouTube and Spotify samples. Researchers can use the dataset to improve AI systems for music-related applications.
Image credit: Flickr user Keith Cooper
