FOUNDATIONS FOR ON-THE-FLY LEARNING IN THE CHUCK PROGRAMMING LANGUAGE
Rebecca Fiebrink, Ge Wang, Perry R. Cook · 2009
Machine learning techniques such as classification have proven to be vital tools in both music information retrieval and music performance, where they are useful for leveraging data to learn and model relationships between low-level features and high-level musical concepts. Explicitly supporting feature extraction and classification in a computer music programming language could lower barriers to musicians applying classification in more performance contexts and encourage exploration of music performance as a unique machine learning application domain. Therefore, we have constructed a framework to execute feature extraction and classification in ChucK, forming a foundation for porting MIR solutions into a real-time performance context as well as developing new solutions directly in the language. We describe this work in depth, prefaced by introductions to music information retrieval, machine learning, and the ChucK language. We present three case studies of applying learning in real-time to performance tasks in ChucK, and we propose that fusing learning abilities with ChucK’s real-time, on-the-fly aesthetic suggests exciting new ways of using and interacting with learning algorithms in live computer music performance. 1.