Spectral analysis of respiratory sounds in healthy and pathological subjects

The authors have investigated the spectral characteristics of healthy and pathological respiratory sounds. The method is based on auscultation. Respiratory sounds are recorded from both children and adults. A measurement setup is constructed and the recorded sounds are labelled as inspiration and expiration to observe the variations occurring in different phases. Power spectral densities (PSD) of the sounds are estimated using periodogram and autoregressive (AR) spectral estimation techniques. The differences between spectral energy distributions obtained from periodograms are also quantified via percentiles. Probability density functions (PDF) of the AR parameters are estimated using Parzen window estimator. The authors have observed noticeable differences between healthy and pathological spectra and PDFs, i.e. power spectral densities of sound waveforms recorded from pathological subjects are observed to contain high frequency components compared to those of healthy sound waveforms. It is concluded that spectral features can be used to develop a respiratory sound classifier.<
>