InterSpeech 2021

Leveraging Real-time MRI for Illuminating Linguistic Velum Action
(3 minutes introduction)

Miran Oh (University of Southern California, USA), Dani Byrd (University of Southern California, USA), Shrikanth S. Narayanan (University of Southern California, USA)
Velum actions are critical to differentiating oral and nasal sounds in spoken language; specifically in the latter, the velum is lowered to open the nasal port and allow nasal airflow. However, details on how the velum is lowered for nasal production in speech are scarce. State-of-the-art real-time Magnetic Resonance Imaging (rtMRI) can directly image the entirety of the moving vocal tract, providing spatiotemporal kinematic data of articulatory actions. Most instrumental studies of speech production explore oral constriction actions such as lip or tongue movements. RtMRI makes possible a quantitative assessment of non-oral and non-constriction actions, such as velum (and larynx) dynamics. This paper illustrates articulatory aspects of consonant nasality, which have previously been inferred from acoustic or aerodynamic data. Velum actions are quantified in spatial and temporal domains: i) vertical and horizontal velum positions during nasal consonant production are quantified to measure, respectively, the degree of velum lowering and velic opening, and ii) duration intervals for velum lowering, plateau, and raising are obtained to understand which portion of the velum action is lengthened to generate phonologically long nasality. Findings demonstrate that velum action tracking using rtMRI can illuminate linguistic modulations of nasality strength and length.