| |
Abstract:
We have developed software to transform between pellet
configurations from the University of Wisconsin X-ray microbeam
speech-production database and a set of parameters describing
midsagittal vocal tract geometry (namely lip aperture, and tongue
constriction degree and location), using a principal components
analysis (PCA) to provide a low-dimensional intermediate
representation. The software can generally, but not always,
successfully recover approximate pellet positions from the
aperture/constriction parameters. We have also investigated
various methods of selecting pellet position data sets to subject
to PCA, for example via an acoustic amplitude criterion and via a
maximum constriction diameter criterion; the results of the PCA
seem quantitatively fairly insensitive to the method of data
selection.
|
