FEATURES OF RESPIRATORY SOUND REPRESENTATION BASED ON SLIDING STATISTICAL ESTIMATES OF THEIR AMPLITUDE

Abstract

This study explores common approaches to the digital representation of respiratory system noises based on moving statistical estimates of signal amplitudes, which enhances their informativeness in the diagnosis of functional disorders.

Traditional examination methods of the respiratory system are analyzed as sources of acoustic signals within the frequency range of 100–2500 Hz, along with modern trends in implementing digital techniques for the recording and processing of such signals. It is emphasized that the subjectivity of acoustic perception, external noise interference, and the physiological variability of respiratory processes remain significant challenges. Enhancing the informativeness of diagnostic signals through statistical methods can improve their robustness to stochastic distortions. The study also presents examples of modern digital instruments, including electronic stethoscopes, computer-based machine learning systems, and mobile applications that provide preliminary screening and processing of respiratory sounds.

Preliminary experimental results are presented, where acoustic realizations of various types of respiratory noises were obtained by digitizing analog stethoscope signals with an 8 kHz sampling rate and 8-bit resolution. The findings demonstrate that, even under limited computational resources, the use of moving statistical estimates—mean value, variance, standard deviation, and information entropy—enables obtaining sufficiently informative representations for further analysis.

The main frequency ranges of typical respiratory noises (vesicular and bronchial breathing, wheezes, and crackles) were identified, confirming the feasibility of statistical analysis for their classification. Furthermore, a comparative evaluation of the effectiveness of various statistical features (sliding estimates) for visual differentiation of breathing cycle phases was conducted. It was found that the moving mean does not provide adequate differentiation between pathological and normal noises, whereas the moving variance successfully highlights key amplitude variations.