Speech Workload Estimation for Human-Machine Interaction

Abstract

Performing tasks quickly and accurately in dynamic and intense environments is critical, such as supervising a remotely piloted aircraft; however, these environments contain periods of low and high workload, which can decrease task performance. A system capable of intelligently adapting its interaction modality based on the human’s workload state may mitigate these undesirable workload states: underload and overload. Such a system requires mechanisms to determine accurately the human’s overall workload state and each workload component state (i.e., cognitive, physical, visual, speech, and auditory) in order to understand the current workload state’s underlying cause effectively. Existing work estimates multiple workload components, but no method estimates speech workload. This manuscript presents an algorithm for accurately estimating a human’s speech workload level using methods suitable for real-time workload assessment. The algorithm is an essential component to future adaptive human-machine interfaces.