Speech and Autonomic Cardiac Dynamics: A Technical Survey and Multimodal Modeling Framework

Modeling the interactions between speech and cardiac autonomic regulation requires bridging physiological mechanisms with robust computational frameworks. In this paper, we argue that heart rate variability (HRV) acts as a mechanistically grounded intermediate target for multimodal speech-ECG learning, offering a way around the brittle nature of handcrafted acoustic biomarkers. We review the current literature to expose distinct structural gaps in how cardiac voice models are currently trained. To address these, we introduce a multimodal HRVspeech alignment framework that pairs self-supervised speech encoders with synchronized ECG data. Ultimately, this approach prioritizes physiological interpretability and temporally aligned learning objectives, setting a more rigorous baseline for deploying machine learning in healthcare.