Tongue contour extraction from ultrasound images based on deep neural network

Aurore Jaumard-Hakoun (1, 2) , Kele Xu (2, 1) , Pierre Roussel-Ragot (2) , Gérard Dreyfus (2) , Maureen Stone (3) , Bruce Denby (2, 1)

1 UPMC - Université Pierre et Marie Curie - Paris 6
2 ESPCI Paris - Ecole Superieure de Physique et de Chimie Industrielles de la Ville de Paris
3 VTVL - Vocal Tract Visualization Lab [Baltimore]

Studying tongue motion during speech using ultrasound is a standard procedure, however automatic ultrasound image labelling remains a challenge, as standard tongue shape extraction methods typically require human intervention. This article presents a method based on deep neural networks to automatically extract tongue contours from speech ultrasound images. We use a deep autoencoder trained to learn the relationship between an image and its related contour, so that the model is able to automatically reconstruct contours from the ultrasound image alone. We use an automatic labelling algorithm instead of time-consuming handlabelling during the training process. We afterwards estimate the performances of both automatic labelling and contour extraction as compared to hand-labelling. Observed results show quality scores comparable to the state of the art.

Mots clés: Tongue shape; Medical imaging; Machine learning; Ultrasound

Domaines: Linguistique

Aurore Jaumard-Hakoun, Kele Xu, Pierre Roussel-Ragot, Gérard Dreyfus, Maureen Stone, et al.. Tongue contour extraction from ultrasound images based on deep neural network. The International Congress of Phonetic Sciences, Aug 2015, Glasgow, United Kingdom. ⟨hal-01366237⟩