Publication details
BESST : Brno Extended Speech and Stress Test
| Authors | |
|---|---|
| Year of publication | 2022 |
| Type | Article in Proceedings |
| Conference | KOGNICE A UMĚLÝ ŽIVOT XX : Sborník z 20. ročníku konference |
| MU Faculty or unit | |
| Citation | |
| Web | https://kuz2022.ciirc.cvut.cz/KUZ_2022_sbornik.pdf |
| Doi | http://dx.doi.org/10.14311/BK.9788001070079 |
| Keywords | machine learning; stress; automatic speech processing; Maastricht Acute Stress Test |
| Description | Stress detection is a traditional topic in the field of automatic speech processing. A historically problematic area within the so-called deep learning approach is the lack of quality reference data for the training of artificial systems. In current research, we have applied psychological methodology to IT in order to gather the necessary empirical data suitable for effective training of deep neural networks in the context of speech load. Neural network models, driven by adequate data inputs, can significantly support the classification and detection of stress in automatic speech processing. For this purpose, the Brno Extended Speech and Stress Test (BESST) was developed, which is an extended adaptation of the original protocol of the Maastricht Acute Stress Test (MAST). The modified BESST methodology aims to maximize the collection of speech outputs from participants in various stressful contexts. The proposed methodology presented in this paper is a functional and scalable tool for collecting the key datasets necessary for stress detection using deep learning techniques. |