Fuzzy Analysis of Speech Metrics to Estimate Conductor Alertness

A Fuzzy Logic-based algorithm has been developed for processing a series of speech metrics with the ultimate goal of estimating train conductor alertness. The output is a single metric, which directly quantifies the alertness level of the conductor. The metrics were selected based on their correlation to alertness through processed speech, but without any interpretation of the spoken words or phrases. Metrics that are used include: speech duration, silence duration, word production rate and word intensity. The assessment of these metrics is an experience and human knowledge based task, which generates the need for a mathematical model to accommodate this special circumstance. The algorithm developed here uses Fuzzy Logic to cast the human knowledge base into a mathematical framework for the alertness estimation analysis. The core of this fuzzy system is a rule base consisting of fuzzy IF-THEN rules, which are derived from the existing knowledge about the effects of sleep deprivation on alertness such as Furthermore, the rules were inferred from actual voice recordings that were taken on board a train. This data was then used to create a classification scheme to determine which pattern in the speech indicates different levels of alertness from anxiety to fatigue. The simplicity of the underlying mathematical model in this approach enables this system to compute and output an alertness metric in real-time. The nature of this algorithm allows for the use of an arbitrary number of rules to classify the alertness level and therefore provides the ability to continuously develop and extend the rule base as new knowledge emerges. The resulting algorithm is a fast, multi-input, single-output system that is able to quantify the train conductor’s alertness level anytime speech is produced.