The main purpose of this research is specifying the articulation difference between L1 and L2 speakers by digitizing tongue motions and analyzing their differences between utterances. Differences in tongue motion directly influence speakers' pronunciation, so it may be possible to determine a speaker's L1 from tongue motion data. By comparing L1 and L2 speakers' tongue motion, we can also guide L2 speakers to improve their L2 pronunciation. In this research, we use coronal cross sections of the tongue taken by an ultrasound scanner to carry out the following: first, record the ultrasound of a speaker's tongue motion using the story “The Boy Who Cried Wolf.” Then, sample mobility information by using histogram of oriented gradients. Next, use Karhunen-Loeve expansion to reduce the vector dimensions. At this time, we get the average difference between the starting vector of tongue motion and the subsequent vectors, then normalize the direction of the two averages. Finally, we use dynamic time warping to compare each vector per frame. The experiment results allowed us to compare speakers' tongue mobility information in words which were recorded in different experiment environments or by different speakers.
Meeting abstract. No PDF available.
Normalization and matching routine for comparing first and second language tongue trajectories
Shusuke Moriya, Yuichi Yaguchi, Naoki Terunuma, Takahiro Sato, Ian Wilson; Normalization and matching routine for comparing first and second language tongue trajectories. J. Acoust. Soc. Am. 1 November 2013; 134 (5_Supplement): 4244. https://doi.org/10.1121/1.4831607
Download citation file: