Comparing articulatory images: An MRI / Ultrasound Tongue Image database

We report the development of a database that will contain paired ultrasound and MRI of tongue movements and shapes from 12 adults, illustrated with pilot data from one speaker. The primary purpose of the database will be to evaluate the informational content of ultrasound tongue images on the basis of the richer articulatory structures visible with MRI, and to provide tongue shape information that can later be incorporated into an image processing algorithm to enhance ultrasound tongue images. Ultrasound is an increasingly popular technique for studying speech production since it provides a real-time image of tongue movements. Its potential as a visualfeedback speech therapy tool has been recognised but has not yet been exploited to any great extent. In part this is because obstruents like /t/ /k/ /ch/,which are important targets for therapy, have tongue shapes in both canonical and common error productions which ultrasound displays rather poorly compared to the more easily-imaged vowels, glides and liquids. By enhancing ultrasound images in real time with information based on our corpus, we aim to create images which we hypothesise will A) be more easily understood by children for clinical feedback B) extend the range and utility of ultrasound generally.