Audio-based cough counting using independent subspace analysis

Area of Science:

• Computational acoustics
• Signal processing
• Biomedical engineering

Background:

• Manual counting of cough events in audio recordings is time-consuming and labor-intensive.
• Existing automated methods often require pre-trained models, limiting their adaptability.
• Ambulatory cough monitoring is crucial for diagnosing and managing respiratory conditions.

Purpose of the Study:

• To develop and evaluate an algorithm for automatic detection of characteristic cough events in audio recordings.
• To reduce the time and effort associated with manual cough event analysis.
• To provide a cough detection method that does not rely on pre-trained models.

Main Methods:

• Utilized time-frequency representations of audio signals.
• Employed independent subspace analysis (ISA) to identify cough-specific sound characteristics.
• Tested the algorithm on a dataset of publicly available audio recordings in synthesized scenarios.

Main Results:

• Achieved a true positive rate of 76% for cough event detection.
• Recorded an average of 2.85 false positives per minute.
• Demonstrated successful automatic event detection and summarization without pre-training.

Conclusions:

• The developed algorithm effectively detects cough events in audio recordings using ISA and time-frequency analysis.
• This automated approach significantly reduces manual counting time for ambulatory cough monitoring.
• The model's performance shows promise for real-world applications in respiratory health assessment.