+- Tendo City (https://www.tendocity.net)
+-- Forum: Tendo City: Metropolitan District (https://www.tendocity.net/forumdisplay.php?fid=4)
+--- Forum: Ramble City (https://www.tendocity.net/forumdisplay.php?fid=44)
+--- Thread: 24/192 audio is pointless. (/showthread.php?tid=7244)
24/192 audio is pointless. - Dark Jaguar - 12th February 2020
So I've been hearing lots of people asking why we haven't started distributing our music using higher bitrates and sample frequencies. "HD Sound", so to speak.
In answer to this, mathematically there's no room for improvement. We reached the pinnacle of sound quality (at least in respect to bit rate and sampling frequency) when CDs arrived on the scene. There is nothing further to be gained there. The limit isn't in the tech, of course we can go further in bit rate and frequency, but in biology and physics. Our ears are the limiter. We can only hear in a range that caps at 20000. The typical 41000 is over double that, which according to a mathematical proof that cannot be challenged (literally, it's a mathematical proof, there is no overthrowing it) is all you need to perfectly capture all audio data in a band limited audio stream. The bit rate can be 2 and it will still handle that. The only thing increasing the bit rate does is reduce the noise floor. 16 bit is enough to reduce it to nearly imperceptible for the vast majority of people. 24 bit will reduce it even more, but it's a diminishing return. My own sound card even manages 32 bit sound sampling. I can't even pick up the difference using my own equipment honestly.
There are SOME ways to improve audio left to explore, but all improving the bit rate and sampling rate could ever do is make all our files take up way more space. If we're going to do that, I'd rather advocate using FLAC files instead of MP3. Not only is flac an open source codec, it's lossless and allows for more than two audio streams. That's a better way to use up more drive space. (Because it's lossless, it necessarily doesn't compress as tightly as MP3, but in this case I think it's worth the tradeoff.)
https://people.xiph.org/~xiphmont/demo/neil-young.html
RE: 24/192 audio is pointless. - Weltall - 12th February 2020
I've been using standard quality on Spotify for many years (as opposed to the high quality option) and I can't tell any difference, except in the amount of 4G data they consume when I'm away from Wifi. TBH I don't think I've downloaded an MP3 in close to a decade.
It is, however, possible to see one use for extreme sound resolution... forensic examination of audio which is alleged to be altered or edited. In the coming age of deepfakery, low resolutions can conceal or obscure artifacts that would indicate tampering.
RE: 24/192 audio is pointless. - Dark Jaguar - 12th February 2020
There's another good use for it. Studio level composition. In those scenarios having an ultra high detail audio is better, since you're relying on hardware that can spot the differences mere human ears can't.
The real problem with deep fakes isn't the deep fakes themselves though. We're still able to spot those AND train AI to spot the other AI's fakes. The real problem is that being able to directly show that someone is horrible with video evidence is going to stop working so long as someone can cry "deeeeeep faaaaaake!". It's coming. Just you wait and see. If Trump wins a second term, that's exatly what he's going to do any time someone confronts him with direct evidence he's lying. "That's a fake video. Computers made it. Get the google server, bring it to me and put it under my desk and plug it into my power strip I'll find it."
RE: 24/192 audio is pointless. - Weltall - 15th February 2020
>The real problem with deep fakes isn't the deep fakes themselves though. We're still able to spot those AND train AI to spot the other AI's fakes.
For now. I see this being the next major technological arms race. The real problem is, we consider recordings to be secure and reliable (in that tampering with a recording leaves clear traces most of the time). Deep fakes remove this certainty. Can you trust the video? Can you trust the people who insist it's real? Can you trust the ones who insist it's fake? Will the average person ever be able to independently verify?
What if someone actually does deep fake Trump? Our existing political views will largely determine whether or not we believe it.
RE: 24/192 audio is pointless. - Dark Jaguar - 16th February 2020
(15th February 2020, 3:59 PM)Weltall Wrote: >The real problem with deep fakes isn't the deep fakes themselves though. We're still able to spot those AND train AI to spot the other AI's fakes.The very point I was getting at. The issue isn't our ability to detect so much as public confidence. At that point, video evidence of say Bloomberg basically just being a smarter Trump won't matter any more. Video evidence of ANYTHING won't matter any more. It'll go back to the before times, when you needed a plurarity of first person accounts to verify things.
For now. I see this being the next major technological arms race. The real problem is, we consider recordings to be secure and reliable (in that tampering with a recording leaves clear traces most of the time). Deep fakes remove this certainty. Can you trust the video? Can you trust the people who insist it's real? Can you trust the ones who insist it's fake? Will the average person ever be able to independently verify?
What if someone actually does deep fake Trump? Our existing political views will largely determine whether or not we believe it.