Researchers from Brno created a multimodal dataset of speech under stress with 79 participants using the BESST protocol, to improve machine learning models