Functional Principal Component Analysis of Vocal Tract Area Functions
Abstract
This paper shows the application of a functional version of principal component analysis to build a parametrization of vocal tract area functions for vowel production. Sets of measured area values for ten vowels are expressed as smooth functional data and next decomposed into a mean area function and a basis of orthogonal eigenfunctions. Interpretations of the first four eigenfunctions are provided in terms of tongue movements and vocal tract length variations. Also, an alternative set of eigenfunctions with closer association to specific regions of the vocal tract is obtained via a varimax rotation. The general intention of the paper is to show the benefits of a functional approach to analyze vocal tract shapes and motivate further applications.