Objective. To evaluate changes in speech signal parameters in patients with chronic obstructive pulmonary disease during the course of treatment, which can be used as a basis for remote monitoring and construction of a binary classifier that distinguishes patients during periods of exacerbation and its relief. Materials and methods. Voice records of 114 patients with chronic obstructive pulmonary disease during the period of aggravation at the time of admission to hospital (group 1) and 60 patient records on the day of discharge after treatment (group 2) were analyzed. For acoustic analysis, Praat v 6.4.35 program was used to calculate spectral and time features, and Python v.3.11.4 — to determine the parameters of chaotic state and complexity of the fundamental tone frequency. Acoustic and prosodic speech parameters were calculated. To study the possibility of classification, an augmentation by randomly dividing the original records into 10 fragments of 20 seconds duration was carried out. The classification was done on four sets of input features — original spectral, time, nonlinear parameters and a combined set. Results. The results of the frequency domain analysis show that there are a significant decrease in the frequency and width of the bands of the first and third formants, as well as significant changes in eight mel-frequency cepstral coefficients. In the time domain, only the non-parametric kurtosis coefficient changed significantly. Analysis of the chaotic state and complexity parameters showed that group 1 and group 2 differ in selective, spectral entropy and Hjorth mobility. In binary classification, the best results are obtained when using mel-frequency cepstral coefficients as features. Conclusion. Changes in the parameters of acoustic voice analysis, which characterize the spectral structure of the signal and the chaotic state of the fundamental tone frequency, were noted in patients with chronic obstructive pulmonary disease. Observation of these changes can be used as a basis for remote monitoring of patients after discharge from the hospital. Additionally, it is possible to develop a classifier that assesses the likelihood of a patient’s condition worsening.
Garanin et al. (2026) studied this question.