Echo Delta — Case study — Marco Middeldorff

01

Starting point

Echo Delta is an air traffic control simulation for Windows. You handle the radio traffic with the pilots yourself, by voice, in real German airspace. I started it because no existing controller game convinced me.

Two things were missing. Speech recognition in comparable simulators was poor, sometimes you had to repeat sentences several times until they were understood. And each game covered only one role, either the tower or the center. I wanted to be able to play every position in German airspace, from the delivery controller to the center and radar controller.

02

Problem

2.1Speech recognition that does not understand sentences. Existing controller simulators recognized spoken radio traffic poorly. Sentences had to be repeated several times until they got through, and that destroys the feeling of actually working on the radio.
2.2Only one role per game. Existing games were designed either for the tower or for the center. A simulation covering every position in German airspace, from delivery to center and radar, did not exist.
2.3Even good recognition is not complete. Even with Whisper, not every spoken sentence is recognized. Without a second opinion, the radio would fall flat at exactly those points.

03

Decisions

3.1

Whisper for speech recognition, an external AI service as a fallback

Speech recognition runs on Whisper. Every recognition gets a confidence value. If it is below the adjustable confidence threshold (default: 0.85), a request goes to an external AI service that evaluates the sentence again. Claude, ChatGPT and Gemini are supported, and further services can be added later.

It was important to me that the player can always see what arrived: at the bottom center of the screen it shows what was recognized, or what exactly was not understood, for example a callsign that was not recognized. F9 also opens a debug overlay with the recognition in detail. That way you repeat precisely what is missing instead of the whole transmission.

3.2

One base, several roles: the rest of the airspace is simulated

All positions share the same base. When the player picks a position, the other positions around them are simulated. Aircraft always spawn outside the German borders and fly in from there. This way every role, from delivery to center and radar, feels like part of the same airspace, and no separate game per position is needed.

3.3

Official DFS data

German airspace is based on data from the DFS. I obtained and had confirmed the permission to use it in advance.

3.4

Unity and C#, offline first

Unity was the obvious choice. I already know C#, and the learning curve is much flatter than with C++ and Unreal Engine. On top of that came a requirement that was fixed from the start: Echo Delta should work without an internet connection, without an account and without a server. A multiplayer mode comes later, in which several players take different positions and control the airspace together.

04

Result

4.1Fast recognition in the normal case. Whisper evaluates a radio call in about 200 milliseconds.
4.2A safety net for uncertain recognition. If the confidence is below the threshold (default: 0.85), an external AI service evaluates the sentence. That takes 2.3 to 4 seconds, and 6.4 seconds for the first approach contact with four instructions in one transmission. In return, the player does not have to repeat anything.
4.3Five positions playable. Delivery, Ground, Tower, Arrival and Departure are implemented, and the rest of the airspace is simulated around the player. Center and Radar will follow.
4.4Free and testable. The game is free and stays free. It is currently 2.5 GB and still in development; on request you can already play and test it now.