5 ms·
With all these new text-to-speech AI models... how can we trust any recordings like this anymore?
by emehex 4y ago
With all these new text-to-speech AI models... how can we trust any recordings like this anymore?
- phailhaus 4y agoFaked audio is still easily detectable.
- maximus-decimus 4y agoI don't know, they sound exactly like the real thing to me : https://www.youtube.com/watch?v=Ep-f7uBd8Gc https://www.youtube.com/watch?v=Ep-f7uBd8Gc
- phailhaus 4y agoIt's easy to fool humans. The point is that it is still really easy to detect voice-generated AI computationally. See also: photoshop.
- bastawhiz 4y agoIt's the job of the regulators to investigate audio like this and ensure its provenance, and speak to the people involved under penalty of perjury. The ability to forge audio isn't new, it's just more accessible.
- falcolas 4y agoThis has been a potential attack since recording devices were invented. It was just done by talented people or people with mixing tools.
- zokier 4y agoyou don't need text-to-speech AI model to fake recording, you just need random guy and a microphone.