Polyhymnia: An Automatic Piano Performance System with Statistical Modeling of Polyphonic Expression and Musical Symbol Interpretation

We developed an automatic piano performance system called Polyhymnia that is able to generate expressive polyphonic piano performances with music scores so that it can be used as a computer-based tool for an expressive performance. The system automatically renders expressive piano music by means of automatic musical symbol interpretation and statistical models of structure-expression relations regarding polyphonic features of piano performance. Experimental results indicate that the generated performances of various piano pieces with diverse trained models had polyphonic expression and sounded expressively. In addition, the models trained with different performance styles reflected the styles observed in the training performances, and they were well distinguishable by human listeners. Polyhymnia won the first prize in the autonomous section of the Performance Rendering Contest for Computer Systems (Rencon) 2010.