Phonetogram averaging methods compared (or) Fourier based phonetogram contour averaging revisited

Peter Pabon

In 1995, Sulter introduced the use of “Fourier descriptors” as part of his approach to a structured phonetogram analysis.
More than ten years on, the method for averaging phonetogram contours with so called Fourier descriptors has no prevalence in phonetogram norm building. A possible reason for this could be the seemingly complexity of the method.
However, this method should deserve more study. It is very elegant and relatively easy to use and enables averaging of phonetogram contour shapes, independent of the overall position within the Fo/SPL range or the size of the encapsulated area.
By including a standard deviation calculation of the contributing phonetograms, this spectral technique could add a deviation band to reflect the proportional local variability along the average contour curves. Moreover, this method has the potential to create fluent morphs between contours, by a simple weighting of the contributing contour spectra as well as facilitate the recognition/classification of contours by spectrum matching techniques.
To lead to a better understanding of the method itself, simple examples showing applications of this method will be presented. Other potential properties mentioned above will be further discussed and the contour spectrum method will be compared with other current methods of norm phonetogram building and phonetogram averaging. Test material will be taken from previously published norm values from (Heylen **), Speyer and Pabon as well as newly collected data from Lamarche.

Peter Pabon, Royal Conservatory, The Hague / University Utrecht / Voice Quality Systems Renée Speyer, Academisch Ziekenhuis Maastricht
Anick Lamarche, Kungliga Tekniska Högskolan (KTH), Stockholm
(Louis Heylen, He already indicated he wants to cooperate, but up to this moment it is still unclear if he wants to be mentioned as an author too **)

Peter Pabon, pabon@koncon.nl