M11: Recurrent Variations for String Orchestra
Hendrik Vincent Koops
RTL Netherlands
Hendrik Vincent Koops is a composer and Senior Data Scientist at RTL Netherlands. He received a B.A. degree with Honors in Audio and Sound Design, and a M.A. degree in Music Composition in 2008, both at the HKU University of the Arts Utrecht. In 2012 he received a B.S. degree in Cognitive Artificial Intelligence and in 2014 the M.S. degree in Artificial Intelligence at Utrecht University. After completing research at the Department of Electrical and Computer Engineering at Carnegie Mellon University, he received a Ph.D. degree in Computer Science from Utrecht University in 2019, where he studied the computational modelling of variance in musical harmony. At RTL Netherlands, he is responsible for developing scalable audiovisual machine learning solutions to make video content more discoverable, searchable, and valuable. Hendrik Vincent Koops was a guest editor for a special issue on AI and Musical Creativity at the Transactions of the International Society for Music Information Retrieval, which focuses on new research developments in the domain of artificial intelligence applied to modelling and creating music in a variety of styles. In addition to his industry and academic work, Hendrik Vincent Koops is active as a composer, his music has received airplay on numerous local and international platforms, including part of selected and nominated works at international film festivals. He is also a co-organizer of the AI Song Contest, an international competition exploring the use of AI in the songwriting process.
Hendrik Vincent Koops’ Recurrent Variations for String Orchestra are a set of variations for String Orchestra, created with generative audio machine learning models. It is based on his submission to ISMIR 2021, where he presented a String Quartet piece for the standard string quartet ensemble of two violins, viola and cello, and consisting of three movements. The Recurrent Variations for String Orchestra are the result of a co-creative process between the composer and generative models, similar to his String Quartet piece. However, in the The Recurrent Variations for String Orchestra, the composer explores the capabilities of the models to generate audio-based variations of the String Quartet. The models used to create this piece were trained on a large dataset of public domain string quartet and orchestral music, and variations were generated by using different parts of the original String Quartet as a seed for the generative models. Model output is automatically filtered and selected based on music feature extraction, and mixed to mimic the sonic qualities of a large string orchestra. The resulting Musique Concrete-like variations reflect the limitations of these models such as capturing musical features like local structure and dependency between voices, and thus knowing how to properly express them. This misinterpretation is reflected in the artwork, which was also co-created with AI, in which a depiction of string players are seen eating their instruments, instead of playing them. This submission presents three variations from the complete set of the Recurrent Variations for String Orchestra.