Greg Kochanski

Prosodic Peak Estimation under Segmental Perturbations

This paper is also available as a preprint.

It points out why one would normally not trust measurements of fundamental frequency peaks in human speech to be particularly meaningful. That's because the position of the peak of a function is very sensitive to any small frequency perturbation that one might add to it. That's a very general problem with measuring peaks. In this case, one might have a broad prosodic peak that one might like to measure, and the perturbation is some sort of segment-related shift in fundamental frequency.

It also includes some tests on an algorithm for finding peak positions better, but that really isn't the core of the paper.

The full reference is: Greg Kochanski, "Prosodic peak estimation under segmental perturbations" (2010), J. Acoust. Soc. Am. Volume 127, Issue 2, pp. 862-873 (February 2010). Doi:10.1121/1.3268511

This work was partially funded by the UK's Economic and Social Research Council (the ESRC) via project RES-000-23-0149.