Using linguopalatal contact patterns to tune a 3d tongue model

The six articulatory parameters of a three-dimensional tongue model were adjusted to replicate linguopalatal contact patterns measured with Electropalatography (EPG). The tongue model is based on artificially sustained articulations measured with MRI and the EPG data provides one possibility to tune the parameters to dynamic speech. A 3D model was generated of the palate and the electrode distribution, allowing the synthetic contact patterns to be calculated. The tongue parameters were then adjusted to minimise the deviation from the natural contact patterns. Substantial reduction of the false and missing electrode contacts was made in the tuning and the synthetic linguopalatal contact pattern is shown to replicate the total characteristics of the natural patterns rather well. The remaining error is often due to lateral asymmetry or central-to-edge contact variations. A new 3D tongue model [1] was recently developed within the KTH 3D Vocal Tract project [2]. The model was based on Magnetic Resonance Imaging (MRI) of a reference subject producing 43 artificially sustained articulations of Swedish and was able to replicate the tongue shapes in the corpus with an overall RMS reconstruction error of 1.2 mm. In a comparative study [3] of coarticulation measured by static MRI and combined dynamical electromagnetic articulography (EMA) and electropalatography (EPG), it was however found that the artificially sustained articulations were hyperarticulated. The conclusion of the study was hence that the static MRI data needed to be complemented with dynamical data, in order to generate a model representative of dynamic speech. Following this conclusion, the parameters of the 3D tongue model were tuned in this study, using data on the linguopalatal contact, collected with EPG. The concept of using three-dimensional models of the tongue and palate to determine synthetic linguopatal contact pattern that can be compared to natural EPG data has been proposed earlier in [4] and [5], but no actual training of 3D tongue models has yet been presented. The reference subject was the same, 28 year-old male native speaker of Swedish, used for the tongue model [1]. The corpus consisted of a subset of the MRI corpus [6], such that it included the 25 articulations with clear linguopalatal contact. /. The fricatives were acquired simultaneously with EMA data, allowing articulatory measures such as the jaw height and midsagittal tongue contour points to be determined at the same time, whereas the rest of the corpus was acquired with EPG only. The linguopalatal contact data was collected …