Speech-Guided Multimodal Learning for Vocal Tract Segmentation in Real-Time MRI

2026-05-18

AI Summary

This work addresses the challenge of accurately segmenting articulatory organs in real-time MRI (rtMRI), which is hindered by low image contrast, rapid motion, and limited spatial resolution. The authors propose a three-stage framework that leverages both speech and phonological supervision during training but requires only rtMRI images during inference. Key innovations include transforming phonological representations into spatial priors, designing a dual-level cross-modal contrastive pretraining strategy to align visual and acoustic encoders, and introducing a cross-attention decoder to effectively fuse multimodal representations. This enables successful knowledge transfer from multimodal training to unimodal inference. Evaluated on the 75-Speaker~Annot-16 and USC-TIMIT datasets, the method significantly outperforms existing unimodal and multimodal approaches, achieving clinically viable, precise vocal tract segmentation without any audio input at test time.

Abstract

Segmenting vocal tract articulators in real-time MRI (rtMRI) is a challenging dynamic image segmentation problem characterized by low contrast, rapid motion, and limited spatial resolution. However, while rtMRI acquisitions may provide synchronized acoustic signals, existing methods discard this information, and the few multimodal approaches that incorporate audio cannot be deployed when audio is unavailable. We propose a three-stage framework that leverages acoustic and phonological supervision during training while requiring only the rtMRI image at inference: phonological representations are converted into spatial bounding-box priors for articulator localization, visual and acoustic encoders are aligned via dual-level cross-modal contrastive pretraining, and the learned representations are fused through a cross-attention decoder, effectively transferring multimodal knowledge into a single-modality inference pipeline. Evaluated on 75-Speaker~Annot-16 and USC-TIMIT datasets, our method outperforms existing unimodal and multimodal methods, demonstrating that multimodal supervision provides transferable benefits for precise and clinically deployable vocal tract segmentation.

Problem

• vocal tract segmentation
• real-time MRI
• multimodal learning
• dynamic image segmentation
• speech-guided

Innovation

• multimodal learning
• cross-modal contrastive pretraining
• cross-attention decoder
• phonological priors
• real-time MRI segmentation