Title: Comparing KEC Recordings with Resynthesized EMA Data
Authors: Konstantin Sering, Fabian Tomaschek
Abstract:
Comparing simulated articulatory movements to their real counterparts
is critical in establishing some trust in the findings produced by a simulator and the
simulator’s generalizability. Point-wise and global measures are compared in this
study on a dataset of human EMA recordings and resynthesised simulated EMA
data. Most measures show comparable sensitivity to the hypothesized structure of
dissimilarity along different groups of EMA trajectories but fail to fulfill metrical
properties. A measure based on the Jensen-Shannon-distance (JSD) seem to show
good sensitivity and fulfills the triangle inequality and is therefore suggested to be
used in evaluating resynthesis models.
Year: 2020
In session: (Speech) Signal Processing in Medical Applications
Pages: 77 to 84