This article presents research on the detection of pathologies affecting speech through automatic analysis. Voice processing has indeed been used for evaluating several diseases such as Parkinson, Alzheimer, or depression. If some studies present results that seem sufficient for clinical applications, this is not the case for the detection of sleepiness. Even two international challenges and the recent advent of deep learning techniques have still not managed to change this situation. This article explores the hypothesis that the observed average performances of automatic processing find their cause in the design of the corpora. To this aim, we first discuss and refine the concept ofsleepinessrelated to the ground-truth labels. Second, we present an in-depth study of four corpora, bringing to light the methodological choices that have been made and the underlying biases they may have induced. Finally, in light of this information, we propose guidelines for the design of new corpora.
This paper is about automatic sleepiness state detection using speech samples. Following previous research carried out for the Interspeech 2011 challenge, we use the Sleepy Language Corpus (SLC) for our experiments. However, as we are willing to record our own subjects within a collaboration project with the Bordeaux hospital, we focus only on the read speech samples of that database. Furthermore, we are looking for understandable cues that can guide clinicians to provide a diagnostic. Hence, we devised a set of meaningful features that are close to the signal and restrict the feature selection process to methods that do not use feature combinations. Thus, using simple correlations and a grid search procedure on the training and development parts of the database, we selected a final set of 23 features, reaching a performance on par with state-of-the-art systems. A discussion is proposed on the subjective ground truth used for the boundary between sleepy and non sleepy speech in this database. Finally, we discuss on the interpretation of the features and provide hints on the physiological causes.