SpeechSpiro: Lung Function Assessment from Speech Pattern as an Alternative to Spirometry for Mobile Health Tracking

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Biomedical Engineering

Background:

• Respiratory illnesses like asthma and COPD degrade lung function.
• Traditional spirometry requires specialized equipment and clinic visits.
• Voice characteristics can indicate airway obstruction and restriction.

Purpose of the Study:

• To develop a novel, non-invasive method for assessing lung function using speech.
• To create a hybrid deep learning model (CNN-LSTM) for predicting lung function parameters from audio.
• To validate the accuracy and generalizability of the speech-based spirometry method.

Main Methods:

• Collected 60-second phone audio recordings from 201 subjects.
• Utilized a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) deep learning model.
• Analyzed spatiotemporal patterns in speech to predict lung function parameters.

Main Results:

• SpeechSpiro achieved accuracy comparable to conventional spirometry devices.
• Demonstrated a mean normalized Root Mean Square Error (RMSE) of 12%.
• Achieved an R-squared score of up to 76% for predicting lung function.

Conclusions:

• Speech-based spirometry offers a convenient, mobile alternative for lung function assessment.
• This technology can enable continuous monitoring of respiratory health outside clinical settings.
• SpeechSpiro has the potential to improve accessibility and management of respiratory conditions.