MusicLM by Google

The MusicCaps dataset consists of 5,521 music clips, each lasting 10 seconds. These clips are labeled with an aspect list and a free-text caption provided by musicians. The aspect list includes adjectives that describe the sound of the music, such as genre, specific instruments, and mood. The free-text caption provides a more detailed description of the music, including specific instruments used and the overall mood. The dataset is sourced from the AudioSet dataset and is divided into an evaluation and training split. It is licensed under a Creative Commons BY-SA 4.0 license. Each clip is labeled with metadata, including the YouTube ID of the video it appears in, the start and end positions of the clip in the video, labels from the AudioSet dataset, aspect list, caption, author ID, whether it is a balanced subset, and whether it is part of the AudioSet evaluation split. The dataset is designed for use in music description tasks.