Automated Spoken-to-Coded Pilot Weather Reports in General Aviation

Abstract

A pilot weather report (PIREP) is a firsthand report of actual meteorological phenomena encountered by an aircraft in flight or on the ground. In general aviation (GA), PIREPs are a critical method of communicating weather information with pilots, controllers, and forecasters. However, GA pilots may find submitting PIREPs a cumbersome, manually intensive, and latent process. Automation of the radio PIREP submission process could lead to an increase in the quantity and quality (i.e., accuracy, timeliness, and usefulness) of GA PIREPs. This paper discusses the development and evaluation of a three-tier system to automatically convert spoken PIREPs to coded PIREPs using speech recognition, machine learning (ML), and natural-language-processing techniques. Specifically, this paper describes 1) the development, evaluation, and limitations of a trained commercial-of-the-shelf speech recognition system to generate text transcriptions of the spoken PIREPs, 2) an ML-based Named Entity Recognition model for segregating and labeling information from PIREP transcriptions, and 3) a final deterministic encoder to generate PIREP codes. If implemented in practice, this approach could significantly reduce interferences associated with submitting PIREPs, increase the number of PIREP submissions, improve the quality of PIREPs, and ultimately make flying safer for GA pilots.