From Dataset to Melody: Enhancing Music Composition with Computational Models

Abstract

This study explores the generation of melodic lines in the ethno-fusion genre using computational models, leveraging a unique MIDI-based dataset. The dataset, designed with two primary dimensions—solos and chords—aims to ensure precise, genre-specific outputs. The core techniques employed are Markov chains and generative grammar, chosen for their suitability in generating sequences aligned with rhythmic, pitch, and structural characteristics of the dataset. Generative grammar focuses on chord generation, isolating it from solos to enhance harmonic coherence, while Markov chains facilitate the creation of both solos and their combination with chords. The results, validated through extensive testing and feedback from music professionals, demonstrate high accuracy and utility in producing new compositional ideas, particularly within the Balkan music industry. This work provides a foundation for further exploration of AI-driven music generation, emphasizing scalability to other dimensions and instruments.

Received: 23 May 2025 / Accepted: 17 July 2025 / Published: 01 August 2025