Dulan S Dias and TGI Fernando published a research paper titled “Komposer – Automated Musical Note Generation based on Lyrics with Recurrent Neural Networks” at IEEE First International Conference on Artificial Intelligence and Data Sciences (AiDAS), 2019 organized by Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA Perak Branch, Tapah Campus and Faculty of Science and Information Technology, Universiti Teknologi Petronas (UTP), Malaysia held on September 19, 2019 at Casuarina@Meru Hotel, Ipoh, Perak, Malaysia.

Abstract:

Musical creativity being one of the strong-hold characteristics that differentiate humans from computers in today’s technologically advanced society, algorithmic composition and song writing are the research areas that aim to bridge this gap. With the introduction and development of various neural network-based methodologies that have shown quite a promise in applications to a wide range other fields, it is promising to see how these new technologies can cater to the domain of musical creativity. Even though there has been significant amount of research done focusing on musical composition, it is not the same for musical song writing. The main objective of this research study is to apply Long Short-Term Memory Recurrent Neural Networks in constructing a machine learning model that can generate musical melody notes when it is provided with a lyrical input (musical song writing). In this study, we were able to successfully generate musical melody notes for provided lyrical inputs with consistencies of over 80%. In addition to that, a web-based inference tool was developed as a result of this study, which allows us to easily generate musical melody sheets when we provide with a lyrical input.