Models for Automatic Musical Composition

The engineering thesis, "Models for Automatic Musical Composition," explores the evolution of automatic music composition, focusing on key algorithms and apparatus. It offers an in-depth review of advancements from Guido of Arezzo's "Micrologus" and Athanasius Kircher's "Arca Musarithmica" to 18th-century musical games (musikalisches würfelspiel), culminating in Winkel's 19th-century musical automaton, the "Componium." The study delves into 20th-century information theory applications, highlighting the first computer implementation in the "Illiac," and examines formal grammar experiments of the Chomskian kind. The second part analyzes power density spectral analysis of classical compositions, emphasizing their 1/f characterization ("pink" musical structures) and its correlation with the prediction-surprise binomial (Clarke Voss's musicality criterion), which is crucial to any musical language. Finally, the third part implements these concepts into automatic music generation systems using multilayer recurrent perceptrons (neural networks), producing compositions beautifully performed and recorded by pianist Rafael Lamas. Presented in 1994 at the Technical University of Madrid's School of Telecommunications Engineering, the thesis was awarded the highest mark by a jury of distinguished professors. It later received the Best Engineering Thesis prize from Spain’s National Radio and Television and the Official College of Telecommunication Engineers.