Unlocking Cross-Modal Biosignal Synthesis: A Temporally-Aware VAE-Diffusion Model

Synthesizing authentic phonocardiograms (PCG) from ubiquitous electrocardiograms (ECG) is a critical task for accessible cardiac monitoring. Existing generative models, however, struggle to capture the heart's complex electromechanical coupling, failing to meet the dual requirements of temporal precision and physiological fidelity needed for clinically relevant waveform analysis. We introduce the Temporally-Aware VAE-Diffusion model, a synergistic hybrid architecture that resolves this trade-off. Our architecture enforces tight physiological coupling through an Enhanced Condition Fusion mechanism and explicitly models long-range cardiac dynamics via Temporal Attention Blocks. On the EPHNOGRAM benchmark, our model sets a new state-of-the-art, achieving a Pearson correlation of 0.910$\pm$0.008, 95.95\% S1 detection accuracy, and a precise 12.0 ms timing error, significantly outperforming leading diffusion and Transformer baselines. Crucially, our work provides a reproducible zero-shot transfer evaluation for ECG-to-PCG synthesis. Evaluated on the synchronized PhysioNet/CinC 2016 training-a/MITHSDB subset without target-domain training, our model preserves high waveform fidelity and clinically relevant timing structure under domain shift, including on pathological recordings. These results support cross-dataset robustness of the proposed synthesis framework, while downstream diagnostic validation remains an important direction for future work.

Heart monitoring often relies on two types of signals: electrical signals from an electrocardiogram (ECG) and mechanical heart sounds from a phonocardiogram (PCG). While ECGs are easy to capture with common devices, PCGs usually require specialized sensors, limiting access for many people. We asked: can we generate realistic heart sounds directly from widely available ECG recordings? To solve this, we designed a hybrid AI model that combines two powerful techniques. First, it learns a simplified representation of real heart sounds, capturing the essential patterns. Then, it uses this representation to generate high-quality heart sounds that match the timing and rhythm of the ECG. Our model explicitly learns how electrical signals drive mechanical heart activity, producing accurate waveforms even on previously unseen datasets. This approach could make dual-modality heart monitoring more accessible, supporting research, algorithm development, and data augmentation without needing extra hardware. While the generated sounds are not yet approved for medical diagnosis, they reliably mimic real heart signals and could help advance low-cost cardiac assessment and AI-assisted heart studies.