Advanced Speech Processing Interface for Autonomous Robots (ASPIAR)

Foreign object debris (FOD) on airfields causes aircraft damage and mission delays. Removal is currently performed by walked visual inspection and sweeper/blower-equipped vehicles, which is laborious and slow. To increase efficiency and effectiveness, the Navy is developing a system for automated FOD removal. To enable a robot to negotiate airfield access with air traffic control (ATC), Charles River Analytics proposes to design and demonstrate Advanced Speech Processing Interface for Autonomous Robots (ASPIAR), a generalizable language understanding module and speech interface for autonomous FOD-removal systems. To understand relevant ATC operations, ASPIAR has a contextually aware computational ATC and FOD domain model encoded in a systemic functional grammar (SFG). To reliably understand speech inputs, the SFG is used to generatively fine-tune an existing automated speech recognition (ASR) model to understand the idiosyncratic ATC communication language as spoken by air traffic controllers, while accommodating variation in accent, enunciation, and background noise. To generate speech responses, the ASPIAR-tailored ASR model generates text, with Synthetic Speech Markup Language metadata to enable expressive voice characteristics, then uses a text-to-speech processor to generate ATC-adherent speech. ASPIAR enables ATC to communicate with the robot using exactly the same verbal protocols they use with human debris-removal teams.