Denoising Cough Sound Recordings Using Neural Networks

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Cough sound analysis aids respiratory disease diagnosis and management.
• Background noise significantly degrades the performance of cough sound analysis models.
• Effective cough sound denoising is crucial for reliable diagnostic tools.

Purpose of the Study:

• To develop and validate a novel method for denoising cough sound recordings.
• To improve the quality and intelligibility of cough sounds corrupted by background noise.
• To investigate the impact of signal processing and neural network parameters on denoising efficacy.

Main Methods:

• Utilized signal processing and machine learning techniques, inspired by speech enhancement research.
• Employed a fully connected feed-forward neural network for supervised learning.
• Trained the model to map noisy cough sound spectra to clean spectra.
• Validated the method on 300 manually annotated cough recordings with babble noise.

Main Results:

• The proposed method demonstrated significant improvements in cough sound quality and intelligibility.
• The denoising performance was evaluated by investigating various signal processing and neural network parameters.
• The developed technique outperformed conventional denoising methods in subjective and objective evaluations.
• The study confirmed the effectiveness of the machine learning approach for cough sound enhancement.

Conclusions:

• The developed cough sound denoising method effectively enhances audio quality and intelligibility.
• This approach holds promise for improving the accuracy of automated cough sound analysis in real-world conditions.
• The findings support the integration of advanced signal processing and machine learning in respiratory diagnostics.