science
Release of spectrogram of cockpit recorder audio allows conversation recovery with ’emerging’ decades-old tech
The US National Transportation Safety Board, which investigates plane crashes, has a policy of not releasing cockpit audio recordings.
Nonetheless, earlier this week, the NTSB released a spectrographic image derived from the cockpit audio recording that captured the last words of two UPS pilots before their plane crashed in Louisville, Kentucky, last year.
Scott Manley, a scientist, developer, and gaming influencer, warned the agency about doing so.
“NTSB doesn’t release cockpit voice recorders from crashes, except in this case they’ve released an image of a spectrogram,” he wrote in a social media post on May 20, 2026. “I’m not sure that’s a good idea since you can probably reconstruct a lot of audio from the megabytes of data encoded in this image.”
Technically savvy individuals promptly turned the soundwave graph back into audio and posted it on the internet, prompting the NTSB to acknowledge it is now aware that advances in image processing and computation allow graphs to be turned back into approximate audio.
“Federal law prohibits such public release due to the highly sensitive nature of verbal communications inside the cockpit,” the board said on Thursday. “The NTSB takes these privacy restrictions seriously.”
The spectrogram was released on May 19, 2026, in conjunction with the NTSB investigative hearing into the November 4, 2025 crash of a United Parcel Service MD-11F cargo plane (flight 2976), which occurred shortly after takeoff from Louisville Muhammad Ali International Airport.
Three crew members on board and 12 people on the ground were killed. Twenty-three others were injured.
The accident has also been reconstructed using a flight simulator and the text transcript of the cockpit voice recorder.
In a post on social media network X, Jennifer Homendy, chairwoman of the NTSB, said, “It’s deeply troubling that emerging technology can be used to extract [cockpit voice recorder] audio from visualized data we share to help the public understand the circumstances of an accident.”
“Emerging” here means at least forty-two years ago. Relevant techniques involving a magnitude spectrum are discussed in a 1984 research paper, “Signal estimation from modified short-time Fourier transform,” by Daniel W. Griffin and Jae S. Lim. Their work builds upon a long established signal processing algorithm, the Fast Fourier Transform. But the availability of machine learning models has undoubtedly lowered the technical barriers to signal transformation.
Coincidentally, “federal science agencies lost about 20 percent of their staff in 2025 relative to the previous year,” according to Nature.
Homendy continued by noting that the laws disallowing the release of cockpit voice recorder audio exist to protect privacy, to preserve investigative integrity, to demonstrate respect for accident victims and their families.
“NTSB is taking steps to address this issue,” she said. “The public docket is offline for now, and we are urging X, Reddit, and others to take such disgusting, manipulated posts down.”
At the time this article was filed, audio reconstructions of the pilots’ last words remained available on X. ®
