NTSB Wants PDF Removed After It Exposed Final Cockpit Audio From UPS Crash (cnn.com) 65
The NTSB temporarily closed public access to nearly all investigation dockets after people used a spectrogram image from a PDF in the UPS flight 2976 crash file to reconstruct approximate cockpit voice recorder audio and post it online. "We show our work and we've been doing this type of thing for years. Nobody was aware that you can recreate audio from a picture," a spokesperson for the board said. "NTSB is looking to make sure there's nothing else in the docket that could compromise anybody's privacy... now that we understand the possibility of a digital recreation." CNN reports: Cockpit voice recordings, often referred to as the CVR, capture everything commercial pilots say and are valuable during NTSB investigations, but are almost never released out of respect for the victims and their families. UPS flight 2976 crashed on November 4, when an engine separated from the wing while it was taking off from Louisville, Kentucky. The three crew members onboard were killed along with 12 people on the ground. During a two-day investigative hearing this week, the board released a docket full of details about the crash. Besides thousands of pages of reports and video showing the engine separating, it included a transcript of the CVR and a PDF file showing an analysis of the spectrogram of the audio it recorded.
A spectrogram is a still image that is a visual representation of the audio, showing the ups and downs of the frequencies. Using that still image, members of the public were able to recreate the voices of the pilots in the moments before the plane crashed and post the results online. The clip, which included background noise and echoes, covered the last 30 seconds of the flight as the pilots struggled with the disabled aircraft as well as recordings of testing the NTSB did on another aircraft.
In a statement on Thursday, the board made clear it "does not release cockpit voice recordings" due to federal law and because of the highly sensitive nature of what they include, but it was "aware that advances in image recognition and computational methods have enabled individuals to reconstruct approximations of cockpit voice recorder audio from sound spectrum imagery." Investigation dockets are made public for transparency, but this week, the board took the rare step of closing public access to all dockets, including the one for the UPS crash. [...] The NTSB is urging platforms like X and Reddit to remove posts with the audio.
A spectrogram is a still image that is a visual representation of the audio, showing the ups and downs of the frequencies. Using that still image, members of the public were able to recreate the voices of the pilots in the moments before the plane crashed and post the results online. The clip, which included background noise and echoes, covered the last 30 seconds of the flight as the pilots struggled with the disabled aircraft as well as recordings of testing the NTSB did on another aircraft.
In a statement on Thursday, the board made clear it "does not release cockpit voice recordings" due to federal law and because of the highly sensitive nature of what they include, but it was "aware that advances in image recognition and computational methods have enabled individuals to reconstruct approximations of cockpit voice recorder audio from sound spectrum imagery." Investigation dockets are made public for transparency, but this week, the board took the rare step of closing public access to all dockets, including the one for the UPS crash. [...] The NTSB is urging platforms like X and Reddit to remove posts with the audio.
Re: (Score:2)
Re:Hmmmmm... (Score:5, Informative)
Nothing.
There is a 30 year old law that prohibits releasing audio from aircraft black boxes. They accidentally "released" the audio by publishing a spectrograph, which is effectively a violation of the law.
So now they're going through all their stuff making sure they aren't accidentally releasing data they are legally prohibited from releasing.
No conspiracy needed.
=Smidge=
Re: (Score:1)
No, they did not release audio.
A spectograph is not audio, it's an image of the timing, signal, frequency of a signal.
The fact that it could be reverse engineered into coherent audio is not consequential.
That's like saying that someone released public source code to a program when all they did was release the binary. Or, more accurately, released a use video of the software, which someone then reverse engineered.
Re: Hmmmmm... (Score:5, Informative)
Re: (Score:3, Informative)
It makes complete sense why they release these though as it seems pretty important to the investigation. The audio isn't even a big deal, if any part of this event triggered PTSD I think the already released video would do a better job than the audio would. Compare:
https://youtu.be/vp1RnbNoKis?t... [youtu.be]
Vs:
https://youtu.be/8ZpytycbBhM [youtu.be]
Just from this, it makes sense why the spectrograph is included at all, namely you can visually observe the sound patterns in the actual recording against known sounds already made b
Re: (Score:1)
All they had to do is to scale down the image. The embedded spectrograph has way more fidelity than was absolutely needed for communicating the point they wanted to make.
Re: (Score:2)
All they had to do is to scale down the image. The embedded spectrograph has way more fidelity than was absolutely needed for communicating the point they wanted to make.
So it's not releasing audio if they apply lossy compression?
Re: (Score:3)
Nobody was aware that you can recreate audio from a picture
I guess the NTSB is unaware of the fact that, since the development of "talkies" as sound-on-film recordings, the audio track of movies was recorded as an image of the audio waveform on the film alongside the movie images. In 1900, Ernst Ruhmer was able to record audio as an image on film and then recreate the original sound from the image. Other developments advanced the recording of sound on film, but sound recording and reproduction technology was not adequate until Lee De Forest was awarded patents in 1
Re: (Score:2)
Re: (Score:2)
It's more like someone claiming they didn't violate copyright for copying a song without permission, because they shared a video that included a sound track containing the song, and not the song itself.
Or it's like spycraft that encodes messages into the pixels of an image.
It doesn't matter how the audio was encoded, if it can be decoded as audio, it's audio.
Re: (Score:2)
then sharing sheet music is the same playing a song
Re: (Score:2)
Sharing sheet music is indeed a violation of copyright, if you do it without permission.
There is one difference, however. Sheet music rarely captures the entire audio spectrum of an arrangement, especially if there are multiple instruments involved. So sheet music can't be used to mechanically reproduce the full audio of the music.
Re: (Score:2)
A spectograph is not audio, it's an image of the timing, signal, frequency of a signal.
So ... audio. What else do you think isn't audio?
An MP3 file which is a collection of windowed discrete cosine transforms? (Like a spectrograph is a windowed representation)
A CD which contains discrete points of waveform amplitude information in time? (Like a spectrograph is an amplitude representation)
Maybe a tape isn't a recording either, after all it's just a bunch of magnets and no one even knows how they work.
Re: Hmmmmm... (Score:2)
It's pretty close to being an MP3 marked as a BMP, actually. It's the result of taking a reversable transformation of the audio signal that separates out the different perceptible components and then discarding the ones that matter least, and keeping the important ones in a convenient form for accessing them. It's the first step you'd take if you wanted a computer to identify speakers or what they were saying. The only part that's image-related is making the diagram, but getting back to the data is just tak
Re: (Score:2, Informative)
Given you can find the audio online and see it matches the official transcript which they *did* release, they aren't hiding anything.
Re: (Score:3)
The only thing they could possibly be covering up is that they have vast teams of ultra-specialised uber-gurus who have no meaningful cross-domain expertise (which is understandable, you can't be an ultra-specialised uber-guru if you do) but also that they've essentially nothing else and therefore nobody who can red-flag when a skill in one domain allows a person to exploit information that is released by another.
There is nothing wrong, at all, with having ultra-specialised uber-gurus for something like the
Re: (Score:1)
He was from West Virginia?
Re:Scott Manley knew (Score:4, Insightful)
Indeed...
Like Scott Manley says, going from a frequency domain image representation to a time domain sound file is something that is extremely old and does not and has not required AI the last 50 years. It's just that they vibe coded the extremely old, extremely normal algorithmic solution. AI did not recreate the dead pilots voice, it just made data preparation and coding a bit less work.
It's almost certain you've used software or seen/heard software output today that transformed between frequency domain and time domain. It's ubiquitous.
Re: (Score:3)
In other words, it's like saying someone violated copyright by making a MIDI version of a popular song.
Anyone gather all the past data? (Score:2)
Wonder if anyone grabbed old crash data and is running analysis for us now.
Re: (Score:2)
Not sure what you hope to gain from this. For the most part where audio was captured complete transcripts were released. In this case too.
Re: Anyone gather all the past data? (Score:2)
It would be very interesting to cross reference the data set to see if there are any out there that do NOT match the transcript.
Re: (Score:2)
Very unlikely considering the CVR audio quality typically being meh and a reconstruction from a printed spectrogram would make it even worse. Unless you are implying that the investigators falsify things.
Re: Anyone gather all the past data? (Score:2)
Well the whole point of the files getting locked down is that sometimes they edit things out of the transcript. The presumption is it's innocuous, like for keeping a name private or some such thing, but checking to confirm is a good idea.
Re: (Score:2)
They usually edit out expletives, too. But that's about it.
That should work well (Score:3)
The NTSB is urging platforms like X and Reddit to remove posts with the audio.
Yes, that should work. Barbara, what do you think?
Re: (Score:1)
Perhaps they can pay them to have it deleted. That's had great repercussions so far.
What's wrong with people (Score:2)
Like . . come on.
Hmmm. (Score:4, Interesting)
A spectrogram is basically a description of the sound and Daphne Oram pioneered technology for turning the informational sections of a spectrogram into sound back in 1958. That would be.... 68 years ago, by my reckoning.
Now, technology has moved on a great deal in 68 years. Exactly what you could do today, relative to what she did back then, is obviously significant. But this really should not have come as a shock.
The lack of understanding of this sort of stuff shows what happens when you have too many niche specialists and too few people who understand the broad technology.
Re: (Score:1)
It wasn't a spectrogram though. It was a picture of a spectrogram. I haven't had a chance to watch Scott Manley's video yet, but unless somebody screwed up and embedded the actual spectrogram, or maybe an unreasonably high resolution image of it, it is pretty impressive to reconstruct 30 seconds of recognizable audio from what should have been a random JPEG figure in a PDF.
Re: (Score:2)
Daphne Oram's Oramics machine turned hand-drawn squiggles that were the information-bearing portions of a spectrogram into recognisable audio. And, like I said, that was 1958. A spectrogram, as others have noted, is an image.
Re: (Score:2)
On what basis do you draw that conclusion?
On the basis that a woman from the Radiophonics Workshop innovated a technique?
Perhaps you are going to argue Einstein was a moron because Noether figured out the relationships between symmetry and conservation laws.
I know there are some idiots here, but frankly you are one of the worst.
Re: (Score:2)
The lack of understanding of this sort of stuff shows what happens when you have too many niche specialists and too few people who understand the broad technology.
I don't think anyone didn't understand this, I think they didn't thing it mattered until someone actually proved the edge case. Heck maybe it still doesn't matter. It could very well be that a lawyer had a look at the wording of the law and changed their mind and that reconstructed audio counted.
If you spend your life analysing every single edge case that may happen you'll never get anything done. It's not a question of understanding, it's just a question of when near enough is good enough to move on, and h
Re: (Score:2)
If you had specialists do so, I'd agree.
This is why programmers do not check every corner case, they hire QA enginers to check every corner case.
You hire generalists to see how interactions between technologies impact things.
Re: (Score:2)
Dude, a spectrogram is just a rendered version of an FFT, and the "technology" for converting it back to audio is just an inverse FFT.
For that matter trained linguists can often read speech directly from spectrograms - you just need to recognize the formants, and there are other obvious clues such as fricatives (burst of high frequency noise), plosives (sudden onsets of speech energy from closed lips to open), etc.
Re: (Score:2)
Yes, that's perfectly obvious. If you check the Oramics machine from 1958, that's basically what it was.
Re: (Score:2)
Yeah, but you said:
> Daphne Oram pioneered technology for turning the informational sections of a spectrogram into sound back in 1958
You seem to want to make it sound like some discovery rather than just IFFT or just "add the sine waves back together". For sure what's trivial to do on a modern computer would have been more of a challenge using 1958 tech, but that's because of the tech, not because of the problem requiring some breakthrough "technology".
You also said "A spectrogram is basically a descript
Re: (Score:2)
FFT didn't exist for another decade.
Re: (Score:2)
Sure, although the Fourier transform itself dates back to 1822.
Of course all you need to create a spectrogram is a filterbank - a bunch of analog filters, and all you need to invert it is to add the frequency amplitudes back together.
This is all a bit irrelevant though since what Oram built was actually an analog musical synthesizer controlled by hand drawn (then optically scanned) film strips. No spectrogram involved at all. Similar analog scanning tech COULD have been used to scan a spectrogram and contro
Re: (Score:2)
FWIW, the first machine to scan a spectrogram and turn in back into sound was "pattern playback", created about a decade before Oram's synthesizer.
https://en.wikipedia.org/wiki/... [wikipedia.org]
Please (Score:2)
Please sar remove the audio spectograph sar it's very important sar
Re: (Score:2, Informative)
Any university s
Why ... (Score:3)
On the other hand, when I worked for Boeing, we would produce thick certification documents chock full data of questionable value. To impress the FAA, I suppose. And to keep them from actually bothering to read the interesting bits.
Spoken like a true Trump-Pet (Score:4, Informative)
"Nobody ever before in the history of the world..."
Well, since about 2008, at least, anyway -> https://arss.sourceforge.net/e... [sourceforge.net]
Re: (Score:2)
This is actually old tech. (Score:4, Insightful)
Re: (Score:2)
This isn't some hi-tech research secret. A spectrogram is just a rendered version of an FFT, so it's gob-smackingly obvious you can convert it back to audio.
OUCH! Doesn't instill much confidence (Score:2)
"""Nobody was aware that you can recreate audio from a picture," a spokesperson for the board said. " """ - OUCH! Doesn't instill much confidence in their competence. Should be bleedingly obvious that the image of a spectrogram has got all the information needed ...
Really? (Score:3)
What tool was used for this? (Score:1)
What tool was used for this "reverse engineering"?