The compiled instrumental music emotion recognition (MER) dataset used in my Master's thesis (2022):
Jonas Klepper Rødningen, Emotion Recognition from Speech and
Instrumental Music: Proof of Shared Emotional
Code Through Transfer Learning, 2022, NTNU
See the tracklist files for the compiled instrumental music emotion recognition dataset (static dimensional valence and arousal values, normalized to [-1,1]): Soundtracks and DEAM. For information about the compilation process, I refer to the thesis.
The chosen segments from EMOPIA, and their Youtube-titles (++) are also included.
Please see the respective original datasets' links above for credits to the authors.
It is recommended to use StratifiedGroupKFold from Sklearn to create train/val/test splits that respects that some clips originate from the same song.