Normalization of vocal tract shape using radial basis function (音声)

概要

論文の詳細を見る
A simple normalization procedure was applied to Electromagnetic Midsagittal Articulographic (EMMA) data using a radial based function to get the general description of articulatory data. The transformation function used in the normalization procedure consisted of not only linear parts but also nonlinear factors. The purpose of the normalization processing was to reduce the cross-subject morphological differences of vocal tract shape and size among raw data. A physiological articulatory model has been used to serve as the template. Data sets of three subjects obtained from NTT EMMA database were adopted in the experiments. The landmarks were defined obeying the same rules among the template and subjects in the vocal tract space. The evaluation results show that the variances among subjects have been reduced after normalization.
社団法人電子情報通信学会の論文
2007-07-19