🤖 AI Summary
To address the scarcity of emotionally annotated musical scores, this work introduces WikiMT++, the first high-quality, multimodally annotated ABC-format music dataset comprising 1,010 manually verified scores. It systematically incorporates objective metadata—including album information, lyrics, and video links—as well as two complementary emotional annotation schemes: a 12-dimensional fine-grained emotion taxonomy and Russell’s circumplex-based quadrant labeling for subjective affect. To enhance annotation reliability, we propose CLaMP (Consistent Labeling and Mapping Pipeline), a novel annotation correction framework that improves label consistency by +37% and reduces annotation error rate by 52%. WikiMT++ enables emotion-conditional music generation, cross-modal retrieval, and emotion classification tasks. It has been adopted as a benchmark by multiple state-of-the-art music AI models, thereby filling a critical gap in structured, emotion-labeled symbolic music data.
📝 Abstract
WikiMT++ is an expanded and refined version of WikiMusicText (WikiMT), featuring 1010 curated lead sheets in ABC notation. To expand application scenarios of WikiMT, we add both objective (album, lyrics, video) and subjective emotion (12 emotion adjectives) and emo_4q (Russell 4Q) attributes, enhancing its usability for music information retrieval, conditional music generation, automatic composition, and emotion classification, etc. Additionally, CLaMP is implemented to correct the attributes inherited from WikiMT to reduce errors introduced during original data collection and enhance the accuracy and completeness of our dataset.
