LubDubDecoder: Bringing Micro-Mechanical Cardiac Monitoring to Hearables

Paper Title

LubDubDecoder: Bringing Micro-Mechanical Cardiac Monitoring to Hearables

Publication Info

• Topic area: Micro-cardiac monitoring using hearables for health applications.
• Keywords: Hearables, seismocardiography, gyrocardiography, cardiac sensing, micro-mechanical vibrations, health monitoring, motion artifact removal, ear-based sensing, personal informatics, cross-device calibration.

Background and Problem

• Problem / challenge: Current methods for measuring micro-cardiac signals (SCG and GCG) require clinical settings and chest-mounted sensors, which are impractical for everyday use. Smartphone-based approaches suffer from placement variability, and existing hearable-based methods rely on specialized sensors unavailable in low-cost devices.
• Significance: Enabling unobtrusive, everyday monitoring of micro-cardiac events can reduce clinical stressors, support timely interventions, and benefit populations with mobility restrictions, such as older adults.
• Motivation and related work: Prior works have explored cardiac monitoring using IMUs, microphones, and radar systems, but these approaches are limited by device availability, user effort, or environmental constraints. LubDubDecoder aims to address these gaps by leveraging the ubiquity of hearables.

Solution

• Proposed approach: LubDubDecoder transforms the built-in speaker of hearables into a sensor to capture coarse heart sounds and reconstruct fine-grained SCG and GCG signals for micro-cardiac event timing.
• Novelty:
  1. Repurposing in-ear speakers for cardiac sensing across diverse hearables.
  2. Zero-effort cross-device normalization for generalization without explicit user calibration.
  3. Motion artifact removal pipeline for robust signal extraction during everyday activities.
  4. Experimental validation across hearables, users, and real-world conditions.
• Procedure and key techniques:
  • Ear-based cardiac sounds are captured using hearable speakers or microphones.
  • Signals are conditioned, segmented, and filtered for noise removal.
  • A temporal autoencoder reconstructs SCG and GCG signals from ear-based sounds.
  • Fiducial points are labeled using automated algorithms.
  • Calibration strategies adapt to individual physiology and device variability.

Results

• Concrete findings:
  • Within-user SCG and GCG reconstruction achieved Pearson correlations of 0.94–0.95.
  • Cross-user correlations reached 0.88–0.91 with five calibration cycles.
  • Cross-device normalization enabled correlations of 0.91 across four hearables.
  • Motion artifact detection achieved 97.7% accuracy.
  • Timing errors for fiducial points were ≤30 ms at the 95th percentile across scenarios.
• Advantage over baselines: Comparable performance to IMU-equipped earbuds (0.92 correlation) and mmWave radar (0.72 correlation), while expanding compatibility to low-cost hearables.
• Experiments / evaluation:
  • Feasibility study with 25 participants using five hearable types.
  • Evaluation across intra-session, inter-session, inter-user, and inter-device conditions.
  • Longitudinal testing over days and weeks, and scenarios like sleep and post-exercise.
  • User experience survey rated ease of use (4.6/5) and trustworthiness (4.3/5).
• Limitations and future work:
  • Challenges with open-ear earbuds and motion-heavy activities.
  • Clinical validation in patients with cardiovascular conditions.
  • Exploration of joint SCG-GCG reconstruction and deployment on hearing aids.

Summary

LubDubDecoder introduces a novel system for micro-cardiac monitoring using hearables by repurposing built-in speakers to capture and reconstruct SCG and GCG signals. It achieves high accuracy across users, devices, and real-world conditions, supported by robust motion artifact removal and zero-effort calibration strategies. The system demonstrates potential for unobtrusive health monitoring in everyday scenarios, with promising applications for older adults and individuals with mobility restrictions. Future directions include clinical validation, integration with hearing aids, and enhanced contextual tracking for self-experimentation and clinical workflows.