Publication

Abstract : In today’s diverse musical landscape, ensuring the authenticity of compositions has become increasingly challenging, with plagiarism threatening artistic integrity. To address this, we propose Melodic Integrity, an AI-driven system that facilitates both music generation and plagiarism detection, including from hummed melodies—a capability not present in most existing models. The system employs Long Short-Term Memory [LSTM] networks to generate original compositions and a Siamese Convolutional Neural Network [CNN] to compare audio inputs with a database of MIDI files for plagiarism detection. Feature vectors are extracted from hummed melodies and paired for classification using similarity scoring. Experimental evaluations show that the plagiarism detection module achieved 98.44% accuracy for humming-based verification and 87.5% for MIDI file comparisons. The LSTM-based music generation module achieved a sequential note prediction accuracy of 40%. These results demonstrate the effectiveness of deep learning in both sup-porting innovative musical composition and enhancing originality assurance through reliable plagiarism detection mechanisms.