Acoustic-to-Articulatory Inversion based on Dynamic MRI

A speech wave is the result of pushing the air through human vocal tract, where distinct vocal tract shapes lead to production of distinct sounds. The aim of acoustic-to-articulatory inversion research is to estimate the human articulatory con?guration when producing di?erent sound waves. In otherwords, acoustic-to-articulatory inversion is the problem of ?nding a function that maps the acoustic to the articulatory domain. In this research, I focus on developing machine-based approaches for modelling the articulation according to speech. I use data-oriented learning techniques for ?nding the correlation between di?erent sound units and vocaltract shape. The model is provided with the acoustic and the correspondingarticulatory data, and is trained to ?nd the mapping between them. The trained model can then be used to derive the articulatory parameters for unobserved acoustic data. This research is developed using a dynamic Magnetic Resonance Imaging database, which consists of midsagittal MRI imagesof human vocal tract (head) while speaking, and the corresponding acoustic data recorded simultaneously during articulation. In this poster, an overview of the current status of this project and a brief insight to the further stages is presented