Devlog 01: Voice Control
Early Access for MultiScope Air Traffic Control is getting close. Until launch we'll post a devlog every week, each one a closer look at a single part of the game.
MultiScope ATC is a radar simulator made by two student air traffic controllers. Early Access launches with Amsterdam Schiphol. This first devlog is about voice control.
ONE TRANSMISSION
Hold push-to-talk and say it in one go: "KLM62P, turn left heading 210, descend to 2000 feet, QNH 1010, speed 180 knots, cleared ILS approach runway 18R."
After you let go of the key, this is what happens:
- What the model heard appears at the top of the scope, so you can check it.
- The aircraft's label changes to the new clearance straight away (optional), before the pilot answers.
- The pilot reads the clearance back and starts the turn shortly after.
- If the model understood you but the game can't carry it out, for example because the runway isn't in use, the text turns red.
WHY WE BUILT OUR OWN SPEECH MODEL
Windows speech recognition capped us at three instructions and no conditional clearances in a transmission, no flexibility, so we trained our own model. A clearance like the one above wasn't possible at all. Even three instructions often didn't come through, since anything worded slightly differently from the list wasn't understood.
So we trained our own model on ATC phraseology. It recognises callsigns, flight levels, QNH, ILS clearances and waypoint names, handles different voices and microphones, and answers in about a fifth of a second. (Depending on PC performance)
WHAT YOU CAN SAY
"Reduce to minimum clean speed" and "maintain slowest practical speed" give the same instruction, the first in ICAO wording and the second in FAA wording. A speed of 220 can be "two two zero", "two twenty" or "two hundred twenty". An "uh" halfway through, or a "good morning" in front of a clearance, is understood. If you misspeak, say "correction" and repeat the whole transmission or only the part you got wrong and the parser will filter the corrected instructions.
Chained instructions work too, conditional ones included:
- "KLM1234, descend to reach flight level 70 by RIVER, speed 250 knots until passing flight level 100"
- "Transavia 6RK, proceed direct ARTIP then fly heading 270, expect ILS runway 18C"
- "Speedbird 789, cross SUGOL at flight level 100, reduce speed to two twenty knots or less"
A SHARED FREQUENCY
Every jurisdiction has its own frequency: Tower, each approach sector and each Area Control sector. When you work more than one sector, you transmit on all of their frequencies at once, as a controller does when sectors are combined.
All the aircraft on a frequency share that one channel, and only one of them talks at a time. Pilots wait for a quiet moment before they call you. If you key the mic while a pilot is talking, the two transmissions block each other: you hear a buzz and neither message gets through.
Pilots also check what you tell them. If you clear an aircraft below the transition level without a QNH, the pilot asks for it and holds the descent until you give it, and a wrong QNH gets a request to confirm. A "descend" to a level above the aircraft gets "confirm climb?" back. A handoff only works to the next controller on the aircraft's route, so you can't send one to the wrong sector by accident.
If a pilot hears their callsign but misses the instruction, they ask for the part they missed: "Say again the level, KLM4733." Calling with just a callsign gets you "go ahead", and "disregard" gets a "roger".
VOICE AND THE INPUT PANELS
You can mix voice with the other ways of giving instructions at any moment. The Basic Input Device (BID), our simplest imput device, has arrow buttons for heading, level and speed, the List Input Device (LID) shows full columns of values, and the Touch Input Device (TID) is made for tapping. The Flyout Input Device (FID) opens menus straight out of the aircraft label. Everything also works from the keyboard, with shortcuts you can rebind, so you can play the whole game without saying a word.
Voice Auto Label Update is on by default, so the label changes as soon as the game understands you. Turn it off and you update the label yourself using one of the input devices, like a real air traffic controler does.
SETTING IT UP
Speech recognition runs on your own PC, so you can play without an internet connection and nothing you say is sent anywhere. With a graphics card you get the full model. Without one, the game uses a lighter version on your processor. If that's still too heavy for your PC, you can switch to Windows speech recognition, which only understands fixed phrases, like the early versions of the game did.
The settings have a microphone test, a setup guide under Audio, and a recognition setting that runs from Strict to Lenient. If your accent or microphone gets missed a lot, set it lower.
LEARNING THE PHRASEOLOGY
The phraseology trainer is meant for players who haven't spoken on a frequency before. You'll find it under New Game > Tutorial. It shows you a task, for example "Turn TRA556 right onto heading 190", and you say the clearance on push-to-talk. The same recogniser the game uses grades it, and your best score is saved for each level. There are eight levels, and the first one is only headings.
Once you're playing, the ? button in the pause menu opens a searchable list of every command. A small toggleable cheat sheet on the scope shows the most sensible commands for the aircraft you've selected.
NEXT WEEK
Devlog 02 is about multi-sector play. You take the approach sectors you want and AI controllers work the others. When you need something from a neighbour, like an aircraft on your frequency early or an agreed transfer level, you ask for it in COSYS.
Follow and wishlist MultiScope ATC on Steam and the next devlog shows up in your activity feed. We're also on Discord:
discord.gg/BDvP6ZuY5m
See you on frequency.
- The MultiScope Team
This announcement was originally published by Scope Studios on Steam on Oct 10, 2026.
Open the original on Steam