Extended Pipeline for Content-Based Feature Engineering in Music Genre Recognition

Abstract

We present a feature engineering pipeline for the construction of musical signal characteristics, to be used for the design of a supervised model for musical genre identification. The key idea is to extend the traditional two-step process of extraction and classification with additive stand-alone phases which are no longer organized in a waterfall scheme. The whole system is realized by traversing backtrack arrows and cycles between various stages. In order to give a compact and effective representation of the features, the standard early temporal integration is combined with other selection and extraction phases: on the one hand, the selection of the most meaningful characteristics based on information gain, and on the other hand, the inclusion of the nonlinear correlation between this subset of features, determined by an autoencoder. The results of the experiments conducted on GTZAN dataset reveal a noticeable contribution of this methodology towards the model's performance in classification task.

MS-SincResNet: Joint Learning of 1D and 2D Kernels Using Multi-scale SincNet and ResNet for Music Genre Classification

“…As shown in The SOTA methods GTZAN dataset (%) Bisharad et al [7] 85.36 Bisharad et al [8] 82.00 Raissi et al [42] 91.00 Sugianto et al [45] 71.87 Ashraf et al [3] 87.79 Ng et al [39] (FusionNet) 96.50 Liu et al [30] 93.90 Nanni et al [37] 90.60 Ours (MS-SincResNet) 91.49…”

Neural Network Music Genre Classification

“…The research presented in this section have a common goal of condensing the dataset by striping out non-essential data points while maintaining the critical information for the NN to learn from. The features extracted from the dataset should be comprehensive, compact, and effective; this means they should represent the music well, require a smaller amount of storage, and require little computation to extract [1,[48][49][50]. If the correct features are not extracted or there is any loss of data while extracting, the machine learning phase will lack the ability to make use of vital information and thus the features chosen will significantly affect the final results and accuracy of the work [40].…”

Combining CNN and Broad Learning for Music Classification

“…LSTM) can grasp the prominent long-term dependency based properties, such as recurrent harmonics and music structure contained in the music. These are the possible reasons why deep learning architecture based schemes have achieved tremendous success in various MIR tasks, such as onset detection [6], emotion recognition [7], chord estimation [8], rhythm stimuli recognition [9], source separation [10], music recommendation [11] and auto-tagging [4], [12], [14], [15]. For music classification tasks, CNN and RNN are the two most adopted deep learning architectures.…”