"Can it run Doom?" is an old engineering joke, and people have answered it on calculators, printers and pregnancy tests. A small satellite in low Earth orbit has now answered it too, with a twist. Nobody uploaded a script or sent a binary telecommand. A person spoke the request into a radio, and the spacecraft worked out what he said on its own.
On July 27, amateur radio operators in Oslo transmitted the phrase "PRETTY play Doom Doom Doom" to OPS-SAT PRETTY, a cubesat in the European Space Agency's OPS-SAT Space Lab. Speech-recognition software running on the satellite turned the audio into text, matched the command and opened the game. The spacecraft then sent back a "postcard": a frame of gameplay, the transcription and diagnostics on the received signal. The result was reported by Space.com on Nov. 9.
Estonian company Tanagra Space, which supplied the software, says it is the first time a satellite has recognized and carried out a spoken command on its own, according to Gizmodo. That is Tanagra's own claim, and it is a narrow one. Satellites have long relayed voice, and ground systems can turn speech into commands. The claim here is that the listening and the decision both happened in orbit.
The spacecraft
The hardware has had more than one job. ESA's OPS-SAT mission page describes PRETTY as a 3U CubeSat, about 10 by 10 by 34 centimeters and 4.6 kilograms. Its name stands for Passive REfleTometry and DosimeTrY. Beyond Gravity Austria, Graz University of Technology and Seibersdorf Laboratories built it under an ESA General Support Technology Programme (GSTP) contract. Gizmodo reports that it launched in 2023. It finished its original mission in October 2024 and joined the OPS-SAT Space Lab in 2025.
Two features made it a good fit for this experiment: a reconfigurable processing platform and a software-defined radio. TU Graz operates the satellite, and ESA's European Space Operations Centre (ESOC) in Darmstadt coordinates the experiments. The Space Lab gives outside teams a real spacecraft to test software on, and here it managed the experiment for ESA.
Because the radio is defined in software, its receive chain can be changed after launch. The team needed that. Space.com reports that the receiver channel was too wide to pick up the voice signal and had to be narrowed. The toolchain included GNU Radio for signal processing and Sherpa-ONNX for speech recognition.
Six misses, then Oslo
It did not work the first time. Space.com reports that radio amateurs made six unsuccessful attempts on passes over Poland and Norway between April and July. The seventh, run with the Oslo Group of the Norwegian Radio Relay League, succeeded.
The effort took 59 days and cost about 7,200 Norwegian kroner, roughly $775, including pizza. The figure says something about the experiment. It was built from a reprogrammable satellite, radio software and volunteers on a roof, not a dedicated flight program.
Gizmodo reports that the team ran two tests on July 27. In the morning they looped a synthesized voice. In the evening they spoke into a live microphone. Space.com names the operator who spoke the command from an Oslo rooftop: Jon Bergli Heier, call sign LB9BJ. The two tests answer different questions. A looped synthetic voice is clean and repeatable, which is a fair way to check that the signal chain works. A live voice brings in the variation that speech recognition finds hard: different pacing, stress and microphone handling, plus whatever the radio path adds. Space.com notes the satellite also had to contend with Doppler shift and an antenna not designed to relay voices.
Repeating "Doom" three times was probably more than a joke. A satellite passing overhead gets a short, noisy, Doppler-shifted signal, and saying the key word more than once gives the recognizer more chances to catch it. The sources do not explain the phrasing, though, so treat that as a reasonable guess, not a documented design choice.
What "on board" means
The key detail is where the transcription happened. In a normal operation, a ground station encodes a command as structured data, sends it up, and the spacecraft checks and runs it. Here the uplink was audio. The satellite had to receive it, recover the voice from the noise, run a speech model on board, decide whether the text was a command it knew, and then act. No computer on the ground turned the voice into a command first.
The postcard is how the team proved it. It carried what the satellite heard (the transcription), what it did (a gameplay frame) and how good the signal was (the diagnostics). That combination shows the decision was made in orbit, and it gives the team data to work out why the first six attempts failed.
Why It Matters
Doom is the hook. The use case Space.com points to is more serious: transcribing "mayday" distress calls. A satellite that can recognize speech on board could, in principle, pick out an emergency call in the radio traffic it hears and flag it, rather than sending raw audio to the ground and waiting for a person or a ground computer to listen.
Turning a novelty demo into a working service would take a lot more. One successful pass after six failures shows the idea can work, not that it is reliable. A distress-call system would have to work with weak signals, different languages and accents, and noisy channels, and its false alarms would matter. A missed "play Doom" costs nothing. A missed mayday could cost a life.
The experiment also shows what retired small satellites can still do. PRETTY's original reflectometry and dosimetry mission ended in October 2024. Because the satellite has a reprogrammable computer and radio, it could be turned into a test platform for software its builders never planned for. ESA's Space Lab approach of reusing a spacecraft that already works and opening it to outside teams got Tanagra and a Norwegian radio club to an orbital demonstration for about $775 in the club's own spending.
The "first" label should still be read carefully. It comes from Tanagra Space, the company whose software was being tested, and it describes something specific: a satellite that recognized a spoken command and acted on it by itself. Within that definition, the July 27 pass is a real milestone. It is also a reminder that how capable a spacecraft is now depends more and more on what software someone uploads to it after launch.
