A method to determine the contribution of annotated performance directives in music performances

Interpreting notated music and performing it expressively is a complex skill that requires years of practice. In the quest for understanding this phenomenon, a question that arises naturally is to what degree performance directives annotated in the score affect expressive variations of tempo and loudness. Computational models of musical expression typically focus on musical structure and do not explicitly take into account annotated performance directives. The objective of the method presented here is to determine the degree to which loudness directives can account for expressive variations in loudness as measured from performances. To this end, we represent loudness directives by mathematical functions and use these to approximate measured loudness curves. This approximation yields coefficient values that represent how strongly each directive is reflected in the performance. Furthermore, the residual loudness curve after subtracting the model fit provides a clearer view on other, nonexplicit factors that influence expressive loudness variations.