Tollywood Emotions: Annotation of Valence-Arousal in Telugu Song Lyrics

R Guru Ravi ShankerB Manikanta GuptaBV KoushikVinoo Alluri

Mar 2023

摘要

Emotion recognition from a given music track has heavily relied on acousticfeatures, social tags, and metadata but is seldom focused on lyrics. There areno datasets of Indian language songs that contain both valence and arousalmanual ratings of lyrics. We present a new manually annotated dataset of Telugusongs' lyrics collected from Spotify with valence and arousal annotated on adiscrete scale. A fairly high inter-annotator agreement was observed for bothvalence and arousal. Subsequently, we create two music emotion recognitionmodels by using two classification techniques to identify valence, arousal andrespective emotion quadrant from lyrics. Support vector machine (SVM) with termfrequency-inverse document frequency (TF-IDF) features and fine-tuning thepre-trained XLMRoBERTa (XLM-R) model were used for valence, arousal andquadrant classification tasks. Fine-tuned XLMRoBERTa performs better than theSVM by improving macro-averaged F1-scores of 54.69%, 67.61%, 34.13% to 77.90%,80.71% and 58.33% for valence, arousal and quadrant classifications,respectively, on 10-fold cross-validation. In addition, we compare our lyricsannotations with Spotify's annotations of valence and energy (same as arousal),which are based on entire music tracks. The implications of our findings arediscussed. Finally, we make the dataset publicly available with lyrics,annotations and Spotify IDs.

展开全部

图表提取