Automatic Recognition of Emotions in Songs: a Summary

400 Words2 Pages
Automatic Recognition of Emotions in Songs: A Summary By Martin Fracker The paper “Lyrics, music, and emotions” written by Rada Mihalcea and Carlo Strapparava explored the notion of automatically recognizing emotions at the line level in songs. Annotated emotions is similar to the genre to which a particular song belongs, as both can be used to categorize and filter a list of songs. Through crowd-sourced data obtained from Amazon Mechanical Turk, and three experiments exploring the methods by which emotional recognition can be accomplished, the automatic recognition of emotion at the line level in songs was proven to be possible. The project started off with the selection of 100 popular songs. For each song ten annotation jobs were posted on Amazon Mechanical Turk. After isolating the reliable annotators from the spammers, roughly two to five annotations for each song were left over. Each annotation consisted of a line-by-line emotional analysis for each song. Each line was rated on the presence of the following emotions and their intensity: anger, disgust, fear, joy, sadness, and surprise. Among these crowd-sourced annotations, it was observed that joy, and sadness occurred the most often. After gathering the crowd-sourced annotations, three experiments were performed, each one attempting to automatically annotate the songs line-by-line. In the first experiment, the songs were annotated using textual features alone in order to determine their usefulness. In the second musical features alone were used; and, in the third, both were used together. It was no surprise that the automatic notation was more successful when both textual and musical features were using in combination. In addition to joy and sadness, it was observed that anger also occurred rather frequently, which was not apparent in the crowd-sourced annotations. They admitted that further

More about Automatic Recognition of Emotions in Songs: a Summary

Open Document