Indonesian Lyrics Classification using Feature Level Fusion

The Lyrics classification is one of remain were challenging topics in Music Information Retrieval. In this research, we build Indonesian Lyrics Dataset (ILD) collecting derived more than a thousand popular Indonesian song from 1970 until 2010. The data was crawling from various offline sources or online sources. We use pre-processing techniques like stopword, Indonesian stemming, dictionary checking, spell correction, and others. We adopt the Feature Level Fusion method that combining bottleneck features of Neural Networks. We tested a model into the genre, mood, and year label of Indonesian lyrics. Experiments show that Feature level Fusion model outperforms single Long ShortTerm Memory and Recurrent Neural Networks for Indonesian lyrics classification.