Google has created a dataset of 5,500 music samples with descriptive annotations to train music generation models. The dataset contains samples generated from a variety of text-based prompts, including rich narrative captions, short texts, a sequence of texts, a combination of melody and text, descriptions of famous pieces of artwork, instrument names, genres of music, musician experience levels, places, epochs, and accordion solos, as well as multiple music samples generated from the same prompt.
Image credit: Flickr user Jade Palmer
