EarFusion: Quality-Aware Fusion of In-Ear Audio and Photoplethysmography for Heart Rate Monitoring

Wearable devices are increasingly adopted for continuous heart rate (HR) monitoring, yet existing solutions remain constrained by modality-specific vulnerabilities. Photoplethysmography (PPG), the most widely used technique, suffers from motion artifacts, skin-sensor variability, and ambient light, whereas in-ear audio sensing, an emerging alternative, is robust to many motion disturbances but easily corrupted by ambient noise and self-speech. These complementary failure modes motivate an integrated approach that can dynamically adapt to signal reliability. This paper introduces EarFusion, the first framework that integrates in-ear audio and PPG for HR estimation through quality-aware fusion. We design modality-specific Signal Quality Indices (SQIs) and incorporate them into a fusion algorithm that allocates trust across streams in real time. Experiments on the OmniBuds platform across diverse activities with synchronized ECG ground truth validate both the effectiveness and feasibility of the approach. EarFusion reduces HR estimation error compared to single-modality baselines, consistently achieves lower variance than either modality alone, and further benefits from SQI-driven integration relative to using a single quality measure.