Validation of Harmonic Progression Generator Using Classical Music

We evaluate the output of a Markov model-based harmonic progression generator, a classic model for corpus-based computational creativity. 87 participants performed a discrimination task classifying 20 musical excerpts as either human-composed or computer-composed. Also recorded was each participant's level of confidence in their choice. Results indicated that while overall performance was above what would be expected from random guessing, further analysis revealed this was due to the human-composed pieces being much easier to identify than computer-composed pieces. Assessed separately, participants were unable to identify computer-composed pieces above chance-levels. We suggest improvements to the experimental design that could be implement in future evaluations.