September 22, 2026
We tested voice input in two conditions at STQ: open tarmac during a ground inspection and enclosed maintenance bay during an equipment service.
Tarmac results. Voice input performed adequately. Ambient noise was a factor — the assistant misheard technical terminology in approximately 20% of queries — but crew members self-corrected naturally and the interaction pattern felt usable. Response quality was consistent with the touch-input sessions.
Maintenance bay results. Not usable. The enclosed bay produces a reverberant acoustic environment that the model’s noise filtering could not handle reliably. Error rate on equipment-specific terms was above 40%. Two crew members stopped trying after three failed attempts. One said: “It’s faster to look it up myself.” That is not what we want to hear.
Decision. We are not making voice the primary input mode. The environments where the assistant is most needed are also the environments where voice works least well. We are instead redesigning the touch interface with oversized tap targets and a reduced-input mode intended for gloved operation. Voice remains available as a secondary option for open-air conditions.
This is not a failure. It is exactly the kind of thing a field test is supposed to surface. The demo worked in the conference room. The conference room is not the job.
Revised UI mockups in review. Next field test in October.