Monitoring human emotions through wearable systems has become an important area of research. Electrodermal activity (EDA) has proven to be a good indicator of emotional arousal, and numerous works have focused on using EDA data to predict emotional states. However, to successfully integrate EDA data into real-time wearable emotion recognition systems, several challenges of practical real-life scenarios, need to be addressed. This paper explores the relationship between speech signals and EDA reactions and analyzes a new approach for classification of skin conductance reactions elicited by emotional arousal using speech signals as a triggering event. Results show an average improvement in skin conductance reaction classification accuracy of at least 5.6% when using speech-triggered reactions compared with traditional methods. The use of speech as a triggering event could help improve real-time emotion recognition algorithms implemented within wearable systems.
EDA signals were captured at a sampling rate of 128 Hz using the Shimmer GSR+Unit1, which provides connections and pre-amplification for one channel of EDA data acquisition. This sensor monitors EDA between two electrodes attached to the middle phalanx of the index and middle fingers on the non-dominant hand. Data collected using this device was transmitted in real time via Bluetooth connection to a laptop (Intel i7, 1.80 GHz) assisting in the study.
In addition, audio signals were captured at a sampling rate of 44.1 kHz using the integrated microphone (Realtek (R) Audio) of the laptop in use. Data recording and synchronization were managed by the iMotions software platform.
