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.
Before you advocate changing an extremely successful safety culture because you want to apply abstract principles, you might want to do some Chesterton's Fence thinking. Aviation safety depends on fearless analysis of objective data and blameless reporting, which is a very unnatural and sometimes counterintuitive framework for humans to operate in.
The NTSB releases transcripts of cockpit voice recordings, just not the literal voices. This is a human consideration that doesn't affect the quality or transparency of the analysis.
The NTSB's publishing of the transcript but not the recording is a pretty standard means for providing full privacy while increasing safety. A recording allows an incident analysis which is extremely useful in updating safety procedures to prevent incidents and plan for the ones that do occur. Publishing the raw sound recording reduces privacy with no increase in safety, but publishing an analysis of the recordings does not harm privacy, while getting the entire safety benefit.
This is different than a privacy vs liability conflict, where a recording isn't going to provide a safety benefit, it'll just move liability around, where there's far more controversy over publishing any analysis of the recording, or even creating one in the first place.
The NTSB should never have published the unredacted spectrograph, as it is effectively a raw sound recording.
On what basis should the recording of voices of identifiable flight crew (often the last words of people soon-to-perish, perhaps unknowingly) be public information?
You seem fairly sure that the public has a right to hear them and that view is not universal and I'm not even sure is a majority view.
The basis is that it's public domain data. And the public a) owns it and b) has property rights that are assertable without cause.
A long time ago (before the infantilization of the American public) this was the default, majority rule. And it's still reflected as the default position in the US Code.
Does your and my tax return data fall into that same category? Individual health records from the Veterans' Administration? Earnings records from Social Security? Census information down to the individual household? Individual FAFSA (student aid) applications? Records of border crossings or even airline flights taken by individuals?
There's a ton of data the federal government has that I consider proper for them to have but for not every detail to be released in the valid interests of privacy.
If you believe the public has ownership and unfettered access rights over all categories of those data, I understand your argument that these voice recordings should be no different. That's an entirely self-consistent line of reasoning.
If you think that some of that data should not be freely accessible to any member of the public, then it's a valid question to ask "do these voice recordings fall on the private or public side of that line?"
I'm only an ardent viewer of crash investigation stuff, not a pro, but it seems to be a good way to show specifics of warning noises, engine sounds, unusual cabin noises (if relevant) and sometimes even structural failures happening over time in a more direct way from the cockpit voice recorder without sharing the actual "audio".
A distressed airplane makes a lot of noises which are very difficult for the human ear to pick out and identify. For example, multiple closely-spaced bangs or rumbling noises will appear distinctly on a spectrogram but will be very hard to hear.
Basically they were just trying to track down weird noises on the recording. They also went and recorded audio on a similar UPS MD-11 to try to identify the source. According to the NTSB:
"This high pitch ringing sound, primarily noted at approximately 6.35 kHz, occurred at 17:13:05.5 EST, shortly after the aircraft rotated for takeoff, and continued with varying amplitude throughout the remainder of the recording. Additionally, a tone at about 2.1 kHz was present along with the ringing that could not be identified."
A spectrogram is literally the same audio, just transformed through a Fourier transform. That transform has a trivial inverse. The spectrogram isn't perfect - the visual representation is low resolution and the phase information is missing - but it's plenty enough to at least figure out what was said. There's nothing surprising that this is possible, only disappointing that whoever published the article didn't realize it.
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.