Microsleep detection in a safety scenario has been a big issue that the research community is trying to solve due to its impact on vehicle safety. In this paper, we present a practical undergoing work to analyze, design, and implement a drowsiness classifier that allows through several different research phases to forecast this behavior. Normally, these systems submit the alarm after the eyes have been closed for a given number of seconds. We are considering a system that can detect and send alarms before the microsleep took place. So, our approach uses audio and video combined to pre-identify the drowsiness condition.