Achieving sufficient temporal and spatial resolution with adequate signal-to-noise ratio (SNR) in dynamic soft palate imaging is challenging. Triggered acquisitions require repeated reproducible speech samples, and while real-time imaging is more reliable, it may lack SNR. Adaptive averaging was implemented to improve SNR in nongated imaging during repetition of a speech task. Similar images were identified using localized cross-correlation before averaging. Adaptive averaging was applied to the soft palate region of images from six volunteers and one patient acquired with various sequences. In volunteers, soft palate SNR increased by 53 ± 17% with four averages. The additional SNR was used to enable reduced slice thickness in two example subjects. Adaptive averaging was also compared to pseudotriggered images in one example and, in another, it was applied to an unrepeated speech task. In a patient with a repaired cleft palate, the technique was used to demonstrate residual velopharyngeal insufficiency. This initial work demonstrates that increased temporal or spatial resolution may be traded for reduced SNR, which can be recovered using adaptive averaging. This will be a valuable tool in assessing velopharyngeal function, particularly in pediatric patients where cooperation may make gated studies difficult or when their head sizes require increased spatial resolution while maintaining temporal resolution.