Construction of corpus for learning fatigue detection from speech

Fatigue, which belongs to human body's natural response and self-regulation for protection, is a complex physiological and mental phenomena. In recent years, a large amount of researchers from both speech signal processing and machine learning domains have already proved that fatigue detection from speech can be carried out automatically. However, the main researches concentrate on driving fatigue detection which contribute for people under work force. Besides, no one pay attention to students in school regardless of the truth of that learning fatigue is becoming more and more indispensable for its positive significance in students' school life experience, the efficiency of learning, even their physical and mental health. Although there are many methods to detect fatigue, detection from speech is a more convenient assumption. So, the corpus is the foundation of researches in detecting learning fatigue from speech. While there are several corpora about fatigue detection, few of them focus on learning fatigue (which is mainly caused by brain activities) and proving the authority and reliability of these corpora. In this paper, we construct the Soochow University Speech Processing Researches-Learning Fatigue Detection (SUSP-LFD) corpus to implement learning fatigue detection from speech. In order to solve the issues among existing corpora, we use heart rate and mean arterial blood pressure to evaluate our corpus. We first describe the construction approach in detail, and then we verify and evaluate the applicability of the corpus.