Voice Recognition
Type: component
PROBLEM
Construction workers need to record measurements without putting down tools or finding a place to write.
SOLUTION
The Voice Recognition module is an embedded single-chip edge AI inference unit running an on-device acoustic model trained specifically on construction measurement vocabulary including fractional inches, metric dimensions, board-foot calculations, the phonetic variations of the word and a half that occur when a person is annoyed, and the full range of ambient noise profiles present on active job sites including circular saws, nail guns, HVAC testing, and one guy who is always on the phone. The model runs entirely on-device with no cloud connection required, which means measurements are not uploaded anywhere, which was a design requirement that came from user research conducted at exactly one job site where the question was asked and everyone immediately looked up from what they were doing. Wake-on-voice activation listens for Hey Pencil or Hey Pen at low power draw and escalates to full inference mode on match. Two button taps disable the microphone entirely for power conservation or for situations the operator would prefer not to document.