Stress Classification Model Using Speech: An Ambulatory Protocol-Based Database Study

Chronic stress poses a significant risk to health, potentially leading to long-term diseases such as cancer and diabetes. Analyzing stress through speech presents a promising avenue, as it offers accessibility and scalability using only a microphone and processor. This study focuses on quantifying stress through speech analysis and its potential implications for disease prevention and treatment. A speech database was obtained from 36 subjects who participated in a stress induction protocol. Acoustic features, including Pitch and Mel-Frequency Cepstral Coefficients (MFCCs), were extracted from the audio recordings. Supervised parametric classifications were conducted using XGBoost, with feature sets defined based on correlation analysis and feature importance. The classification results were validated using leave-one-out validation. Key findings include the development of a speech database for stress detection in laboratory settings, optimization of feature sets for the model, resulting in a classification accuracy of 82%. These results highlight the feasibility of speech-based stress analysis and its potential impact on healthcare strategies.