The expanded Turing test applied to algorithmic music composition

Abstract

The field of research in algorithmic music composition investigates the development of systems that have the ability to produce works automatically. In this context, gaps related to human-computer interaction via digital music are observed, with a demand for works that evaluate the material produced in relation to the listener. The present work describes the evaluation of an algorithmic music composition system through the application of the expanded Turing test. The experiment had the participation of 237 volunteers, subdivided into 3 groups: 10 professionals who work in the musical area, 39 participants with specialized musical knowledge and 188 participants with little or no musical knowledge. Excerpts from 3 compositions were used, one of them generated by the algorithmic composition system Fraseado and the other two classical compositions in the Western context. Among the results obtained, it was found that 77.1% of the participants identified the algorithmically generated excerpt as having been created by traditional composition methods, indicating the system's ability to exhibit a behavior equivalent to the human being.