A Time-Series-Based Analysis Method for Crew Task Allocation Using Cockpit Voice Recognition

In the field of aviation safety, the evaluation of flight crew time pressure based on crew function allocation is a key indicator for workload assessment, and its accuracy directly influences cockpit design safety. Traditional studies relying on voice activity detection (VAD) techniques have utilized the mere presence of voice signals as the basis for judging communication function execution. However, this coarse-grained method cannot distinguish intra-crew conversations from formal communication, making it difficult to accurately interpret actual crew task distribution. As human-centered cockpit design has received increasing attention, there is an urgent need to improve dynamic models for crew workload assessment. This study proposed a method for analyzing crew tasks based on cockpit voice recognition. By constructing a deep semantic analysis model, certain core task categories, such as communication coordination, operational monitoring, and emergency decision making, were identified. This method integrates natural language processing and speech signal processing techniques to overcome the limitations of conventional VAD methods. Leveraging the Paraformer model, semantic features in the pilot speech were extracted to enable time series-based crew function recognition. The experimental results demonstrated that compared with conventional approaches, the proposed method presented a more accurate identification of crew functions, offering an improved solution for optimizing crew task allocation.