Desenvolvimento de um sistema gerador de música simbólica para jogos eletrônicos do gênero RPG

Abstract

This work details the development process of a melody generation system, focused on RPG games for legacy video game consoles, employing machine learning techniques. The system offers the generated melodies in both audio (WAV) and MIDI formats. Its primary goal is to serve as a tool for inspiring composers, helping them overcome potential creative blocks. To build this system, a database of MIDI files was used, representing musical pieces from games of this genre, specifically from the Nintendo Entertainment System (NES) game console. During data preprocessing, the melodies from each track were separated into different labels to improve contextualization for the learning model, which employed a Neural Network approach, specifically Long Short-Term Memory (LSTM) networks. The system includes a graphical user interface, through which users input a partial melody and request the generation of a new one, based on their input and the selected emotion. The implementation produced melodies with a certain degree of structural organization and low dissonance between notes, although without clearly defined musical patterns. This project aims to demonstrate its utility for future implementations that seek to enhance the integration of artificial intelligence within musical production, potentially expanding to other genres or different aspects of musical structure.