Comparative Performance Evaluation of Transformer GPT and DCGAN Models for Monophonic Music Generation Using ABC Notation

This paper presents a comparative performance evaluation of Transformer-based GPT and Deep Convolutional Generative Adversarial Network (DCGAN) models for monophonic music generation using symbolic ABC notation. The study focuses on analyzing structural accuracy, tonal diversity, repetition control, and musical coherence. Experimental results demonstrate that the Transformer GPT model significantly outperforms DCGAN in terms of melodic consistency, transition learning, and resistance to mode collapse. Objective metrics such as repetition score, length similarity, pitch histogram distribution, and transition matrices are used along with qualitative musical observations.