Computer Science Department
School of Computer Science, Carnegie Mellon University
A Probabilistic Method for Tracking a Vocalist
Lorin V. Grubb
This work presents an implemented system and method for automatically accompanying a singer given a musical score. Specifically, I offer a method for robust, real-time detection of a singer's score position and tempo. Robust score following requires combining information obtained both from analyzing a complex signal (the singer's performance) and from processing symbolic notation (the score). Unfortunately, the mapping from the available information to score position does not define a function. Consequently, this work investigated a statistical characterization of a singer's score position and a model that combines the available musical information to produce a probabilistic position estimate. By making careful assumptions and estimating statistics from a set of actual vocal performances, a useful approximation of this model can be implemented in software and executed in real time during a musical performance.
As part of this project, a metric was defined for evaluating the system's ability to follow a singer. This metric was used to access the system's ability to track vocal performances. The presented evaluation includes a characterization of how tracking ability can be improved by using several different measurements from the sound signal rather than only one type of measurement. Examined measurements of the sound signal include fundamental pitch, spectral features dependent upon the score's phonetic content, and amplitude changes correleted with the start of a musical note. The evaluation results demonstrate how incorporating multiple measurements of the same signal can improve the accuracy of performance tracking, for individual performances as well as on average. Overall improvement of the performance tracking system through incremental specification, development, and evaluation is facilitated by the formal statistical approach to the problem.