# OPS-SAT PRETTY: how a spoken command made a satellite play Doom

> A radio amateur spoke "PRETTY PLAY DOOM" to an ESA satellite and it obeyed. Here is the pipeline, the kit and what a voice-command test needs.

- Published: 2026-11-05
- Topic: [Space](https://futuretechnologyhq.com/topic/space/)
- URL: https://futuretechnologyhq.com/article/ops-sat-pretty-how-a-spoken-command-made-a-satellite-play/
- Publisher: [Future Technology HQ](https://futuretechnologyhq.com/), free to read. Quote as "Future Technology HQ" with a link to the article.

## Key points

- The satellite did the whole job onboard: capture, denoise, transcribe, match and launch, with no human in the loop
- A noise-tolerant matcher matters because the radio link mangles the transcript
- Two amateur teams, in Poland and Norway, formed the ground segment
- Voice recognition in orbit points to distress-call relay and onboard event detection

## Article

After reading this, you will understand how an amateur radio team got an ESA satellite to launch Doom from a spoken sentence, and what you would need to attempt a comparable voice-to-orbit experiment yourself.

On 27 July 2026, a radio amateur group in Oslo transmitted "PRETTY PLAY DOOM" to OPS-SAT PRETTY, an experimental European Space Agency satellite. The spacecraft captured the signal, recovered the voice, transcribed it, recognised the command and launched the 1993 first-person shooter, then sent back a "postcard" image of the moment. According to Tanagra Space, which ran the experiment, it is the first time a satellite has acted on a plain spoken command on its own in orbit.

The stunt is light-hearted. The engineering is not. Satellites are normally commanded with telecommands, structured binary packets sent during scheduled passes under strict protocols such as the CCSDS TC Space Data Link Protocol and the European Packet Utilisation Standard (PUS). Here, ordinary speech crossed a noisy amateur radio link and an onboard AI model made sense of it.

This is a reporting-based guide to the method. It is not a recipe to commandeer a spacecraft: the experiment depended on scheduled passes run by the OPS-SAT mission operations team.

## What you need before starting

**Difficulty:** advanced. **Realistic timeline:** the team's link tests ran from April to June 2026 before the successful July campaign, so budget months, not an afternoon.

- **A software-defined radio on board.** The satellite's software radio records the signal as the spacecraft passes over the ground station.
- **Mission operations access.** Passes were scheduled and operated by the OPS-SAT mission operations team.
- **A ground segment of radio amateurs.** Two volunteer teams did the transmitting: the Oslo Group of the Norwegian Radio Relay League (NRRL), operating from a rooftop under the club call sign LA4O, and radio amateurs in Legnica, Poland (SQ6RDP and SQ6QV), who ran the earlier link tests.
- **A speech-to-text model and a command matcher** that run on the satellite's small onboard computer.

## Step-by-step process

### Step 1: Capture the signal

As the satellite flies over the ground station, its software radio records the incoming radio signal. The ground team transmits on the 23 cm amateur band during a scheduled pass.

### Step 2: Narrow in on the voice

The recording is mostly noise. The onboard software finds the voice buried in it and tunes to just that sliver of the spectrum, discarding everything else.

### Step 3: Transcribe onboard

A speech-to-text AI model, which turns recovered audio into words, runs on the satellite itself. The source describes the processing as running on the satellite's own computer with no human in the loop.

### Step 4: Recognise the command

A matcher checks the transcript for the command phrase. Crucially, it is tolerant of the mangling a noisy radio link causes. The source says the matcher is tolerant of the mangling that a noisy radio link causes.

### Step 5: Act and report back

On a match, the satellite plays Doom, renders a postcard from the gameplay and the signal, and downlinks it. That image is the proof that the chain worked end to end.

### Step 6: Escalate from recording to live voice

The team ran the test twice on 27 July. On the morning pass (Run 6), they transmitted a synthesised voice message from a keyer, on a loop. That evening (Run 7), an operator spoke the command live into a microphone, with no keyer and no recording. Both were detected. The source describes the second as the first detection of a live human voice from orbit.

## Common mistakes to avoid

- **Demanding an exact transcript.** Radio noise corrupts speech. A matcher that is not tolerant of mangled words will miss genuine commands.
- **Skipping the link tests.** The Polish team's April to June runs came before the successful campaign. Those link tests ran from April to June 2026.
- **Starting with a live voice.** The team began with a synthesised voice on a loop, and moved to a live operator that evening.

## Tips that make a difference

- **Do all the processing onboard.** The whole point is autonomy: capture, denoise, transcribe, match and act, with no human in the loop.
- **Make the result visible.** Having the satellite downlink a postcard image of the moment turns an abstract detection into evidence anyone can check.
- **Use volunteers wisely.** The amateur community supplied the ground segment, and the project credits them by call sign.

## Why it matters beyond Doom

Georges Labrèche, the principal investigator and founder of Tanagra Space, told SpaceNews that voice recognition in orbit could help capture distress calls globally and act on them immediately, rather than waiting for a Coast Guard or another ship. He floated a service where someone calls "mayday" into the air and a network of satellites relays it at once. He also suggested the same approach could train models to recognise astronomical events, such as gamma-ray bursts or colliding black holes, directly onboard so they are detected sooner.

Those are possibilities, not products. For IT and security teams, the nearer lesson is architectural: putting a model next to the sensor and acting locally, much as the on-device approach in [Nvidia's AI factory operations agent](/article/nvidia-ai-factory-operations-agent-how-sre-agents-work/) moves decisions closer to the system being run. It also raises an obvious question for any autonomous command channel: who is allowed to speak to it? Readers deploying AI that acts without a human, in space or on the ground, can start with this [10 point check before handing an AI agent access](/article/ai-agent-security-checklist-10-points/).

## Next steps

- Read the flight runs and per-pass analysis, and the Oslo (LA4O) campaign report, linked from the project page.
- Study the source code on GitHub to see how the matcher handles noisy transcripts.
- If you are a licensed amateur, look at how the Oslo and Legnica teams structured their link tests before attempting transmissions.
- Learn the baseline: how CCSDS and PUS telecommands work, so you can see what voice would and would not replace.

## Key takeaways

- The command, recovery, transcription, matching and launch all ran on the satellite's own computer.
- The matcher's tolerance for noisy transcripts is the unglamorous part that made it work.
- Two volunteer teams, in Legnica and Oslo, supplied the ground segment.
- Practical uses, such as distress-call relay, are the team's suggestions rather than services.

## Summary

You now know how a spoken sentence became a satellite action: a software radio captured the pass, the voice was pulled from noise, an onboard model transcribed it, a tolerant matcher recognised the command, and Doom launched with a postcard sent home as proof. You also know what a comparable experiment needs: a flexible satellite, operations support, licensed ground operators and months of link testing. Watch the project's GitHub record for what the team tries next.

## Sources

- [Tanagra Space](https://tanagraspace.com/opssat-pretty-doomed/)
- [Gizmodo](https://gizmodo.com/an-amateur-radio-operator-told-a-satellite-to-play-doom-and-it-listened-2000835488)
- [spacenews.com](https://spacenews.com/satellite-on-orbit-responded-to-voice-command-in-experimental-first/)
