What it is, in plain words

It is a voice unit installed next to a machine. An operator says what they want, such as starting, stopping or adjusting the machine, and the unit passes that instruction to the machine's controller. It then answers by voice, so the operator knows the command was understood. Everything happens on the unit itself. Nothing is sent over the internet.

How it works

• The operator speaks a command in their own words.
• The unit hears it and filters out background factory noise.
• Speech recognition turns the audio into text, and a small local language model works out what the operator means.
• The unit sends the command to the machine's PLC (programmable logic controller, the industrial computer that runs the machine) over Modbus TCP, the industrial protocol PLCs use to communicate over a network.
• The unit confirms the action by voice.

What you get

• A voice unit installed beside your machine, running on local edge hardware (a computer on site, rather than a server in the cloud)
• Start, stop and adjust commands carried out through your machine's PLC
• Natural phrasing, so operators don't have to memorise fixed keywords
• Noise handling built for loud factory floors
• Spoken feedback that confirms each command
• Fully offline operation, with no internet connection needed

Who it's for

• Packaging lines, where operators' hands are full and screens are out of reach
• Production lines where walking to a control panel interrupts the work
• Factories with weak or unreliable network coverage on the floor
• Sites where production data must stay inside the building
• Machine builders who want to add voice control to their equipment

Why it runs fully offline

Noise handling, speech recognition, command understanding and voice feedback all run on the edge hardware beside the machine. Nothing is sent to the cloud. That has three practical benefits. The system does not depend on your network, so a dropped connection does not stop it. Your production data stays inside the factory. And because there is no round trip to a remote server, the response is fast.

What we learned about factory noise

Our first prototype used a microcontroller, a small low-power chip, with simple keyword matching. It struggled with factory noise and only understood fixed phrases. We moved to more capable edge computing hardware that runs a modern speech recognition model and a compact local language model. The language model understands natural phrasing, not only exact keywords, so operators can give a command the way they would normally say it.

We also found that filtering noise too hard makes recognition worse. Speech recognition models are trained on natural speech, and over-filtered audio can confuse them. Getting that balance right is where most of the engineering goes.

Where it's going

Our goal is a voice module that can be installed on any machine and adapt to any factory's noise without retraining on site. That is the direction we are working towards. Today, the best first step is to talk to us about your machine and your floor.

Safety and control

Voice control is an extra way to operate the machine, not a replacement for its existing controls. Your physical controls, emergency stops and safety systems stay in place and work as they do today.

Does it need an internet connection?

No. Everything runs on the unit beside the machine, so it keeps working without internet.

Does our data leave the factory?

No. Speech and commands are processed locally on the unit. Nothing is sent outside the factory.

What machines can it connect to?

Machines controlled by a PLC that speaks Modbus TCP. If you are not sure what your machine uses, send us the details and we will check.

Do operators have to learn special commands?

There is no fixed phrase list to memorise. The local language model understands natural phrasing, so operators can speak the way they normally would.

Interested in voice control for your line? Contact MAGO AI to arrange a demo or an assessment of your machines.