A dataset of Valence/Arousal detection
For copyright reasons, it is not possible to make available audio and lyrics of the songs in the dataset. Then only deezer track ids and Million Song Dataset ids are provided as input of the task. The Million Song Dataset provides audio and lyrics features while the Deezer API provides metadata that might be used to retrieve audio or lyrics data from other sources.
Details are provided in the paper: "End-to-end Multimodal Music Mood Detection Based On Audio And Lyrics", Rémi Delbouys, Romain Hennequin, Francesco Piccoli.