ISCA Archive Interspeech 2024
ISCA Archive Interspeech 2024

BESST Dataset: A Multimodal Resource for Speech-based Stress Detection and Analysis

Jan Pešán, Vojtěch Juřík, Martin Karafiát, Jan Černocký

The Brno Extended Stress and Speech Test (BESST) dataset is a new resource for the speech research community, offering multimodal audiovisual, physiological and psychological data that enable investigations into the interplay between stress and speech. In this paper, we introduce the BESST dataset and provide a details of its design, collection protocols, and technical aspects. The dataset comprises speech samples, physiological signals (including electrocardiogram, electrodermal activity, skin temperature, and acceleration data), and video recordings from 90 subjects performing stress-inducing tasks. It comprises 16.9 hours of clean Czech speech data, averaging 15 minutes of clean speech per participant. The data collection procedure involves the induction of cognitive and physical stress induced by Reading Span task (RSPAN) and Hand Immersion (HIT) task respectively. The BESST dataset was collected under stringent ethical standards and is accessible for research and development.