Doppler-Based Sound Localization and Its Application in AI-empowered Traffic Warning for Deafness

Abstract

Hearing is essential for information processing and our safety, especially in traffic. For patients with hearing impairments, sound localization devices help identify threatening sound sources and reduce traffic safety risk. Existing methods typically rely on measuring time delay using microphone arrays. Unfortunately, this method faces critical challenges for dynamic sound sources, posing major threats. Addressing this challenge, we propose a wearable sound localization system that combines interaural time differences and Doppler frequency shifts to determine the position, direction, and speed of moving sound sources. The method's feasibility is first evaluated through a combination of theoretical calculations and experimental verifications. A proof-of-concept setup was established by engaging three microphones as the receiver and a toy car emitting a constant tone as the sound source. An AI algorithm based on artificial neural network was further trained using the received sound signal when the source moved at different locations and directions. The results demonstrated accurate detection of Doppler shifts, with classification accuracies of 100% for front/back, 87.5% for distance, and 62.5% for left/right. Finally, the system was integrated into a wristband with feedback motors, providing vibrational alerts based on the detected motion and proximity of sound sources. These results validate the feasibility of using Doppler shifts and machine learning for motion detection in real time. The proposed system offers a portable solution to enhance the awareness of the environment for individuals with hearing impairments and lays the groundwork for future warning devices in traffic safety applications.