Estimating Respiratory Rate From Breath Audio Obtained Through Wearable Microphones.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Respiratory rate (RR) is a key indicator of health, fitness, and disease progression.
• Remote monitoring of RR offers a cost-effective solution for continuous health tracking.
• Existing methods for RR monitoring can be intrusive or expensive.

Purpose of the Study:

• To investigate a model-driven approach for estimating respiratory rate (RR) from audio signals.
• To assess the feasibility of using wearable headphones for remote RR estimation.
• To develop a system capable of tracking cardio-respiratory fitness and disease progression.

Main Methods:

• Collected audio data from 21 healthy adults using microphone-enabled headphones during and after strenuous exercise.
• Utilized a multi-task Long-Short Term Memory (LSTM) network with convolutional layers to process audio features (mel-filterbank energies).
• Implemented a model for both regression (estimating RR) and classification (detecting heavy breathing, RR > 25 bpm).

Main Results:

• Achieved a concordance correlation coefficient (CCC) of 0.76 for RR estimation.
• Obtained a mean squared error (MSE) of 0.2, indicating accurate RR approximation.
• Demonstrated the viability of audio signals for estimating RR, even in noisy conditions.

Conclusions:

• Audio-based estimation of respiratory rate using wearable headphones is a feasible and cost-effective approach.
• This technology can facilitate remote tracking of cardio-respiratory fitness and disease progression.
• The developed multi-task model effectively estimates RR and detects heavy breathing from audio data.