Implementation of Computationally Efficient And Accurate Music Auto Tagging

In recent past, Convolutional Neural Network becomes a prominent methodology used in Music Information Retrieval (MIR) projects, such as audio classification, mood recognition, auto tagging of music then on. Music auto tagging is a preeminent research crux and modern advances in audio analysis algorithms hasten the improvement of automatic music tagging system. In particular, many new systems use Convolutional Neural Network and function on raw audio wave form as input or Mel-spectrogram representation as input data. However, due to diversity in the exploratory setups developed by researchers, for instance, using different dataset split and software versions, it is challenging to compare the proposed architectures. We looked at the EfficientNets model family in this research, which achieves state-of-the-art performance on the music auto tagging and provides assessment measures such as Precision, Recall, F1score, Accuracy, Confusion Matrix & ROC-AUC.