Speakers as sensors: Machine learning-based earphone impedance analysis for ear canal insertion depth prediction

Area of Science:

• Biomedical Engineering
• Acoustics
• Machine Learning

Background:

• Traditional diagnostic systems can be costly and complex.
• There is a need for low-cost, portable diagnostic solutions.

Purpose of the Study:

• To investigate earphones as sensors for acoustic load detection.
• To develop a machine learning model for predicting insertion depth using electrical impedance.
• To establish a foundation for smart, portable diagnostic systems.

Main Methods:

• Utilized electroacoustic coupling principles.
• Analyzed variations in electrical impedance as sensor data.
• Employed dense neural networks for classification.
• Identified optimal frequency-domain regions for improved sensitivity.

Main Results:

• Achieved 87% overall accuracy in predicting insertion depth.
• Improved accuracy to 91% with speaker-specific data and 99% for individual units.
• Demonstrated enhanced classification performance by focusing on sensitive frequency regions.
• Regression models showed promise despite higher error rates.

Conclusions:

• Earphones can function as effective sensors for acoustic load detection.
• Machine learning analysis of electrical impedance provides accurate insertion depth prediction.
• This approach lays the groundwork for affordable, portable biomedical diagnostic tools.