Tongue contour extraction from ultrasound images based on deep neural\n network

Abstract

Studying tongue motion during speech using ultrasound is a standard\nprocedure, but automatic ultrasound image labelling remains a challenge, as\nstandard tongue shape extraction methods typically require human intervention.\nThis article presents a method based on deep neural networks to automatically\nextract tongue contour from ultrasound images on a speech dataset. We use a\ndeep autoencoder trained to learn the relationship between an image and its\nrelated contour, so that the model is able to automatically reconstruct\ncontours from the ultrasound image alone. In this paper, we use an automatic\nlabelling algorithm instead of time-consuming hand-labelling during the\ntraining process, and estimate the performances of both automatic labelling and\ncontour extraction as compared to hand-labelling. Observed results show quality\nscores comparable to the state of the art.\n