Copyrighted Music Detection Using Machine Learning Techniques

This research aims to develop a model for detecting copyrighted music using machine learning techniques and audio fingerprinting. The training data for the model were obtained from JSON files containing extracted audio features, which were processed using a 1D Convolutional Neural Network (Conv1D) architecture with Dense Layers to classify music based on MFCCs, Delta, Delta-Delta, Spectral Peaks, Spectral Flux, Harmonic Peaks, and Frame Energy. The model was improved using SMOTE to address data imbalance issues and evaluated through K-Fold Cross Validation to enhance its learning efficiency and classification accuracy. The experimental results indicate that the training accuracy ranged from 52.23% to 54.55%, while the validation accuracy varied between 51.21% and 54.79%, suggesting that the model effectively learned from the training data while maintaining stable validation performance. However, the accuracy gap indicates a degree of overfitting, requiring further improvement. Additionally, validation loss fluctuated between 1.1611 and 1.3744, as shown in Figure 2, indicating that while the model achieved lower training loss, its generalization capability remains a challenge. Evaluation through the Classification Report revealed an overall accuracy of 40.18%, demonstrating strong classification capability. The highest precision was observed for (0.57) and “Missed Call” (0.42), while the lowest recall was found in the “LONDON” class (0.31), as highlighted in Figure 1. The confusion matrix further reveals misclassification trends, particularly in overlapping classes, where was frequently misclassified as . To enhance the model's robustness, future work should focus on refining feature extraction, implementing better regularization techniques, and optimizing hyperparameters to reduce overfitting. Additionally, expanding the dataset and incorporating real-time matching techniques will further improve the system's efficiency in detecting copyrighted music in practical applications.