EmoMusic: Learning to Represent and Interact with Music via Emojis

Aman Shukla

Abstract: Both music emotion recognition and emotion-based inter- faces have been popular topics of computer music research. However, very few systems enable user interactions with rich and fine-grained emotions. In this paper, we use emojis, a prevalent means of expressing emotions in text-based communication, as the media to convey emotion in music audio. Our EmoMusic system learns to represent music with emojis in a weakly supervised fashion by regarding lyric texts as the intermedi- ate weak labels. The front end of the system is an emoji-based music interface. The interface not only displays both song-level and segment- level music emotions through emojis but also allows user interactions by searching songs via emojis. The whole system demonstrates the interplay between emotion and music in a simple and intuitive way and aims to enhance the listening experience of the user through emojis.