Evaluation of the measurement precision in three-dimensional Electromagnetic Articulography (Carstens AG500)

Abstract

Three-dimensional Electromagnetic Articulography (EMA) measures location and orientation of the moving speech articulators in real time by means of small, wired sensors. We evaluated the measurement accuracy of the Carstens AG500 EMA system using data acquired simultaneously with the Vicon optical motion tracking system (OPT). EMA sensors and OPT markers were combined in a single rigid object to be able to predict location and orientation of the EMA sensors from OPT motion tracking data. The error was computed as the root mean squared (RMS) error. We found that deviations from constant inter-sensor distances (relative error) were in general below 1 mm and 0.6° while the difference between the measured and estimated positions (absolute error) ranged between 1 and 2 mm and 0.5° and 0.7°. By examining error patterns, four critical orientation regions were detected, but no discernible location dependent error patterning. Sensor velocity appeared to have little impact. The RMS error of the original position calculation has not been found to be a reliable predictor. In the absence of a clear error structure we recommend careful analysis of unexpected findings in speech production data acquired with EMA. Avenues for further improvement of the system are discussed.