Edited by Anne Lacheret-Dujour, Sylvain Kahane and Paola Pietrandrea
[Studies in Corpus Linguistics 89] 2019
► pp. 261–269
The quality of the speech recordings in the Rhapsodie project is highly variable, ranging from excellent to very poor. The main enemies of a fundamental frequency analysis are echo, microphone low-pass filtering, overlapping sound sources, creak, etc. As some fundamental frequency tracking algorithms perform better than others on specific speech segments, alternative methods can prove to be more efficient. This chapter describes approaches to identify problems and provides guidelines to select and apply one of the five available fundamental frequency tracking algorithms (spectral comb, AMDF, autocorrelation, Cepstrum, harmonic selection) on specific speech segments. These methods are implemented in a single software program and can be applied to specific recording segments by simple graphic commands. The resulting reliable pitch curves enable automatic stylization and annotation processing, which is usually not possible using a single F0 tracking method.