Drawing MusicUsing Neural Networks to Compose Descriptive Music from Illustrations
- Lucía Martín-Gómez
- Javier Pérez-Marcos
- Alfonso José López Rivero
- Giovanny Mauricio Tarazona Bermúdez
- Daniel H. de la Iglesia (ed. lit.)
- Juan F. de Paz Santana (ed. lit.)
- Alfonso J. López Rivero (ed. lit.)
Editorial: Springer International Publishing AG
ISBN: 978-3-031-14858-3
Año de publicación: 2023
Páginas: 30-42
Congreso: DiTTEt: International Conference on Disruptive Technologies, Tech Ethics and Artificial Intelligence (2. 2022. Salamanca)
Tipo: Aportación congreso
Resumen
The creative capacity of machines is still questioned by researchers and users alike. For this reason, computational creativity does not only focus on the development of machines for the creation of artistic content but also on the evaluation of the generated content. This works presents a system that composes polyphonic music from the drawings of a user in real time. Our proposal provides an analysis of the Fantasia film, produced in 1940 by Walt Disney and deduces the relationship between its audio and images. As part of system development, an LSTM-based Recurrent Neural Network was trained with MIDI music files and a model was obtained. As a result, the proposed system generates polyphonic music with expressive timing and dynamics by inferring chords from the user’s drawings. To assess the creative ability of the machine a Turing test was conducted and the quality of the interconnection between drawings and music was measured by another user test. Additionally, the performance of the considered classifiers is discussed.