Purpose. The study is intended to explore the usability and effectiveness of the acoustic environment safety estimation method in relation to potentially triggering stimuli for people with post-traumatic stress disorder. The proposed method merges power spectral density (PSD) and mel-frequency cepstral coefficient (MFCC) feature extraction with statistical testing and lightweight machine-learning classification and aims to minimize distress responses to salient sound events. The method is designed for portable embedded platforms with optional cloud-side analytics, ensuring analysis efficiency while allowing application by technical and medical personnel within a structured acquisition protocol. Material & Methods. Captured audio recordings were grouped into three categories: non-harmful everyday sounds, harm-related (danger-associated) acoustic stimuli relevant to populated environments and elevated threat salience. The signals were analyzed using Welch-based PSD across 20 Hz–20 kHz (100 Hz bands; Hamming window) and MFCC features (13 cepstral coefficients plus log frame energy). Aggregated feature vectors were computed for each file. The observed data points were divided into three groups of 252 each (N = 756 total). To assess normality, in addition to graphical review, the Shapiro–Wilk test (null hypothesis rejected when p < 0.05) and the Kolmogorov–Smirnov test (null hypothesis rejected when p < 0.20) were used. Parametric (Student’s t) or nonparametric (Mann–Whitney U) pairwise procedures were applied where appropriate. To evaluate differences across the groups, an omnibus Kruskal–Wallis test was used. Classification experiments employed two-layer neural networks, with 10% of the data used for testing. Results. The difference was statistically significant between the three groups of signals for the aggregated PSD- and MFCC-derived descriptors as a result of the proposed method implementation. The use of selected parametric tests, rank-based methods, and conditional nonparametric comparisons was justified by the non-normal distribution of per-file results for both feature families under the study’s testing framework. As illustrated by the study’s compliance matrices for PSD and MFCC inputs, a neural network based on engineered features generated classification consistent with those suggested by spectral and cepstral summaries. Conclusion. The proposed method is internally consistent within the conducted statistical analysis framework for the three-group setup and demonstrates the ability to differentiate between the analyzed acoustic contexts. Moreover, it provides easy implementation within a portable measurement concept and higher analysis speed compared to manual waveforms review. Additionally, integration of audio descriptors with formal hypothesis testing and learned models offers a reliable practical foundation for further testing on independent cohorts and hardware prototype development.
This work is openly licensed — share and adapt freely with attribution to the authors and the journal. View license terms ↗
Copyright (c) 2026 Анастасія Володимирівна Паренюк, PhD Дмитро Володимирович Паренюк, PhD, дослідник Катерина Л. Руденька (Автор)
Abstract–The paper presents the results of an experimental study of the acoustic characteristics of modular MEMS microphones INMP441 and SPH0645, aimed at their…
This paper considers the task of improving road traffic safety through the development of a bidirectional warning system for road users. The proposed system provides…
The active use of aerial objects creates a need for passive detection methods. This paper presents an experimental study of an automated acoustic hardware and software…
You may also start an advanced similarity search for this article.
