Deep Learning-Driven Non-Contact Sound Source Localization via Multi-Axis Analysis with Laser Doppler Vibrometry

This study presents a non-invasive sound source localization methodology that leverages multi-axis vibration analysis combined with deep learning classification. Accurate sound source localization is critical in various fields, including clinical diagnostics, where determining the origin of acoustic signals provides valuable physiological insights. Using Laser Doppler Vibrometry (LDV), sound-induced surface vibrations are measured, and directional information is extracted from the Log Power Spectra (LPS). The proposed framework integrates theoretical modeling of multi-axis vibrations, experimental validation, and deep learning techniques, utilizing convolutional operations and Bayesian inference to estimate the direction of arrival (DoA) of sound sources. Experiments conducted with two distinct materials at varying frequencies demonstrated an average classification accuracy exceeding 97% across angles from -90° to 90°. These findings highlight the potential of multi-axis vibration analysis for precise DoA estimation, with promising applications in clinical diagnostics, acoustic engineering, and assistive hearing technologies.Clinical Relevance - The proposed methodology introduces a non-invasive approach to sound source localization, with promising applications in clinical diagnostics, acoustic engineering, and assistive hearing technologies. By leveraging LDV-based vibration measurements and deep learning, it overcomes the limitations of traditional diagnostic tools in handling complex sound propagation. The ability to precisely localize acoustic events could enhance diagnostic accuracy across various medical disciplines. For instance, in cardiology, this method may improve the localization of heart murmurs caused by turbulent blood flow, aiding in the diagnosis of valvular defects. In pulmonology, accurately identifying the source of adventitious lung sounds, such as crackles or wheezes, could help pinpoint pathological changes like inflammation or airway obstructions. In assistive hearing technologies, multi-axis vibration analysis could contribute to the development of hearing aids that more effectively localize and process sound in real-world environments. This methodology represents a major advancement in non-invasive diagnostics, with the potential to significantly improve patient care across multiple medical specialties.