Owen Lucas
$100 Site Donor 2023
- Joined
- Sep 5, 2021
- Messages
- 4,804
The NTSB report for UPS flight 2976 included a spectrogram of the cockpit voice recorder audio.
Someone was able to run the spectrogram through an AI model and turn the image back to audio.
It's not an impossible task, likely any university lab could have accomplished this previously on the 40 year old voice visualizer algorithm with some custom coding. What made this news worthy is someone got an off the shelf script, ran the hi-res image through AI, and recreated the voice recording with 10 minutes of compute.
The NTSB took their docket offline supposedly to scrub spectrograms.
Someone was able to run the spectrogram through an AI model and turn the image back to audio.
It's not an impossible task, likely any university lab could have accomplished this previously on the 40 year old voice visualizer algorithm with some custom coding. What made this news worthy is someone got an off the shelf script, ran the hi-res image through AI, and recreated the voice recording with 10 minutes of compute.
The NTSB took their docket offline supposedly to scrub spectrograms.