Description

Our data consists of a single male English native speaker, reading 1899 sentences providing a total of 2.5 hours of speech audio. A subset of 720 sentences is from the Harvard set which was read twice at a normal and fast pace. The remaining sentences were a subset of the TIMIT dataset.
Acoustics and articulatory movement were recorded using a Carstens AG501 EMA device. Passive transducers were attached to speech articulators using medical-grade cyanoacrylateglue. Three sensors were placed midsagittally on the tongue surface, one sensor on the tongue dorsum (TD), one on the tongue blade (TB), and one behind the tongue tip (TT). Two more sensors were parasagittally placed to the left (BL) and right (BR) of the tongue blade. Three additional sensors were placed on the lips, two were midsagittally attached on the upper(UL) and lower lips (LL) at the vermillion border, and one on the right corner (LC) of the lips. Additionally, two sensors were placed on the jaw on the gingiva below the medial incisors (LI)and between the canine and first premolar (LJ).