On a packaging line, operators' hands are busy and screens are often out of reach. Voice control is an obvious fit, but factories are loud, internet connections are unreliable, and machine commands must never be misheard.

What we built

MAGO AI's voice control system listens for spoken commands and turns them into machine actions through Modbus TCP, the industrial protocol used by PLCs. Operators can start, stop and adjust the machine simply by speaking.

Why it runs fully offline

Everything runs on local hardware beside the machine: noise handling, speech recognition, command understanding and voice feedback. Nothing leaves the factory. That means no network dependency, no data privacy concerns and fast response times.

What we learned

Our first prototype used a microcontroller with keyword matching, which struggled with factory noise and rigid phrasing. We moved to a more capable edge computing setup that runs a modern speech recognition model and a compact local language model. The language model understands natural phrasing rather than only exact keywords.

We also found that aggressive noise suppression can hurt recognition, because a model trained on natural speech can struggle with over-filtered audio. Getting that balance right is where most of the engineering goes.

Where it's going

Our goal is a voice module that installs on any machine and works in any factory's noise conditions, without on-site retraining.

Interested in voice control for your production line? Talk to MAGO AI.