Automated Classification of Chinese Traditional Music Genres Using Multi-Modal Knowledge Graph Convolutional Networks

Abstract

In the digital music era, effective classification of music genres is crucial for organizing vast music databases. This study addresses the challenges of manual annotation by proposing an automated approach for classifying Chinese traditional music genres. Leveraging a MIDI dataset, the music excerpts undergo preprocessing using the Adaptive Multi-Scale Improved Differential Filter (AMSIDF) to enhance signal quality. Subsequently, distinctive features are extracted using the Synchro-Transient-Extracting Transform (STET), enriching the data with essential musical characteristics. The preprocessed and feature-enriched data is then fed into a Multi-Modal Knowledge Graph Convolutional Network (MKGCN) for classification. The MKGCN model is adept at integrating multi-modal data sources, enabling the fusion of audio features with metadata or textual information. Through knowledge graph convolutional layers, the MKGCN model captures intricate relationships within the data, facilitating accurate classification of Chinese traditional music genres. Then MKGCN optimized with Black Winged Kite Algorithm (BWKA) for accurate classification of Chinese traditional music such as Dance music, Metal, Rural, Classical, and Folk genres. This study achieves significant advancements in the automated classification of Chinese traditional music genres. The proposed Chinese Traditional Music Classification using Multi-Modal Knowledge Graph Convolutional Network (CTMC-MKGCN-BWKA) approach is implemented in Python. The performance of the proposed CTMC-MKGCN-BWKA approach attains 21.14%, 24.31% and 23.78% higher accuracy, 22.74%, 21.01% and 15.28% higher Precision and 22.14%, 24.01% and 22.71% higher Recall compared with existing methods such as Chinese Traditional Music Classification using Depp learning (CTMC-DL), Chinese Traditional Music Classification using Deep Neural Network (CTMC-DNN), Chinese Traditional Music Classification using Deep Belief Network (CTMC-DBN).