Architecture for Automated Tagging and Clustering of Song Files According to Mood
This addresses the difficulty for users in recalling songs in expanding digital libraries by offering a mood-based classification system, though it appears incremental as it combines existing methods.
The paper tackles the problem of organizing digital song libraries by proposing an architecture for mood-based classification, using audio content analysis and lyrics to map songs into a 2D emotional space and form clusters, but does not provide concrete performance numbers.
Music is one of the basic human needs for recreation and entertainment. As song files are digitalized now a days, and digital libraries are expanding continuously, which makes it difficult to recall a song. Thus need of a new classification system other than genre is very obvious and mood based classification system serves the purpose very well. In this paper we will present a well-defined architecture to classify songs into different mood-based categories, using audio content analysis, affective value of song lyrics to map a song onto a psychological-based emotion space and information from online sources. In audio content analysis we will use music features such as intensity, timbre and rhythm including their subfeatures to map music in a 2-Dimensional emotional space. In lyric based classification 1-Dimensional emotional space is used. Both the results are merged onto a 2-Dimensional emotional space, which will classify song into a particular mood category. Finally clusters of mood based song files are formed and arranged according to data acquired from various Internet sources.