Music era classification using hierarchical-level fusion

Music era is one of Music Information Retrieval research that connecting several songs with similar characteristics from similar year or decade but not limited to particular genre and mood. Previous researcher tried to recognize musical era with classification model using single audio feature like spectrogram and chromagram, but the performance was poor. Feature and model selection affect classification era performance. One of the challenge in selecting feature is whether the using of multimodal or combination of audio features can improve music era classification performance. In this research, Hierarchical-level fusion model is used to combine several audio features like spectrogram and chromagram to determine music era. We obtained both 83% and 73% overall accuracy for Indonesian Music Dataset (IMD) and Million Song Dataset (MSD) of era classification tasks using Hierarchical-level fusion model. This research result also strengthened with overall precision, recall, and F-score result 0.83,0.82, 0.82 for IMD dataset and 0.73, 0.72, 0.72 for MSD dataset experiment.