Utilising generative AI to assist in the creation and production of Chinese popular music

Authors : Xinhao Li, Hyuntai Kim

Pages : 703-722

Doi:10.31811/ojomus.1668286

View : 164 | Download : 270

Publication Date : 2025-10-29

Article Type : Research Paper

Abstract :Background: The rapid development of generative artificial intelligence (AI) has had a significant impact on the music industry, particularly Chinese popular music (C-pop), which presents unique difficulties due to its distinctive melodic structures and emotional depth. While AI can boost efficiency and inspire composers, current models are ineffective in feature selection and prediction accuracy, leading to compositions without stylistic integrity and commercial viability. Problem Statement: Current AI-generated melodies often lack the stylistic depth of human compositions and fail to predict market success accurately. These challenges highlight the need for a more effective AI framework capable of generating high-quality melodies. Objectives: This study introduces the GenAI Melody-LSTM algorithm. This generative AI-driven technique uses Long Short-Term Memory (LSTM) networks to create melodies inspired by Chinese pop songs and forecasts their success. The main goals are to build an AI pipeline for music preprocessing, improve feature selection, train a deep learning model for stylistically consistent melodies, and compare its effectiveness to other methods. Methodology: The methodology utilises a GenAI Melody-LSTM algorithm, which contains data preprocessing techniques such as mode and mean imputation, label encoding, and Min-Max scaling. Feature selection is improved using the Hybrid Filter-Wrapper Ensemble (HFWE) method, which combines Mutual Information, Chi-Square Test, ANOVA F-Test, and Recursive Feature Elimination (RFE) with Support Vector Machine (SVM), Random Forest, and Gradient Boosting Machine (GBM), with the final subset determined by majority voting. The selected features, like melody structure, key signature, tempo, rhythm complexity, instrumentation, and emotion, are used as inputs for an LSTM-based deep learning model that comprises multiple LSTM layers, dropout layers to prevent overfitting, and a dense output layer to create melodies and forecast their commercial success. Results: Performance evaluation using accuracy, precision, recall, F1-score, and Matthew’s correlation coefficient (MCC) provides better results with the GenAI Melody-LSTM algorithm, with 89.7% accuracy, 87.5% precision, 88.2% recall, an F1-score of 87.8%, and an MCC of 0.82. Conclusion: This research demonstrates that integrating generative AI, optimal feature selection, and deep learning significantly enhances Chinese pop music compositions. The LSTM-based model generates melodies and predicts their commercial viability, enabling composers to fine-tune AI-generated music for improved quality and market appeal.
Keywords : Üretken yapay zeka, Çin pop müziği, LSTM ağları, özellik seçimi, melodi tahmini

ORIGINAL ARTICLE URL