SWARAM: Spectral Waveform Analysis using ResNet for Acoustic Music Genre Recognition

Music is a beautiful and enchanting art form that stirs emotions and creates a special connection with our feelings. With the help of streaming services and new releases, over the last decade, there has been a remarkable expansion in the music industry. The increased demand for music has brought attention to how crucial it is to organize music well to improve accessibility and offer more accurate suggestions. Deep learning techniques, which are effective for image classification, are a promising solution for music recognition. However, transfer learning, which uses pre-trained models to improve the performance of new models, has rarely been used for this task. This paper proposes a unique method for categorizing music genres that makes use of transfer learning and ResNet50. MFCCs are used as a representation of the audio data. The GTZAN dataset, a benchmark for music genre recognition, was utilized to assess the proposed method. Techniques for data augmentation have also been studied to increase the capability of the model for generalization. The proposed approach outperforms numerous recently published studies and achieves 94.5 % accuracy.