Published Paper Details:

: In the contemporary music landscape, it's a fallacy to believe that only those with expert knowledge in music can create high-quality tunes. Music enthusiasts, regardless of their expertise, can produce music that resonates. Music is a universal language that many enjoy, and the advent of automatic music generation stands as a potential revolutionary milestone in the music industry. Traditionally, music was created manually using analog processes. However, recent advancements have seen music production pivot towards digital methodologies, with technology playing a supporting role alongside human creativity. This paper outlines the development of generative neural network frameworks capable of capturing the intricate aspects of harmony and melody autonomously. It provides a concise overview of music's fundamental characteristics, citing relevant sources where necessary. The core aim of the research is to develop a method for generating musical compositions using Long Short-Term Memory (LSTM) networks within the framework of Recurrent Neural Networks (RNNs). The generated compositions are initially in ABC notation, a method for music representation, which is subsequently transformed into MP3 format for broader accessibility. ABC notation serves as the chosen format for training the model, splitting into metadata that details the characteristics of the music, such as index, time signature, default note length, and tune type, and the actual musical notes represented by a series of characters. The algorithm is designed to master the patterns of monophonic music through a trio of single-layered LSTM networks, showcasing its ability to learn and replicate musical sequences efficiently.