AI Fixed the Audio, But Should We Tell People?

Are any be history. Welcome Hi. The Pro Audio Sweet those guys a professional and motivated. With text The LEO Stars, George Whittam, founder of Source Element, Robert Marshall, International audio engineer, Darren, Robbo Robertson Global Voice Andrew Peters. Thanks to Tryboo, Austrian Audio Making Passion, Herd Sauce Elements, George the Tech Wisdom and Robbo and AP's International Demo. To find out more about us, check the Pro Audio Suite dot com and welcome to another pro audio suite thanks to Austrian Audio Making Passion heard and hopefully we'll be wearing their T shirts very soon. We must emailed and say give me an OC eight one a T shirt. That look cool at least a pair of socks, right. That's really funny the fact that Sennhi's were handing out socks for people with very long, skinny feet. Yeah, I was going to say maybe they maybe they kind of thought, well, if you forgot your pop guard, it might be that's right, smell. Your toes while you do get the cheesy raids. Speaking of cheesy reads, we've got Ai squaed the full Circle of Aim. What happened? Robert I had. A gig to mix a little corporate video, and funny enough, the corporate video was about the corporation's adoption and rules of adoption of AI and what they were going to do and all this, and they very good job recording it, and so it needed a lot of cleanup. And yeah, I first did the usual thing, and I thought, you know, this is pretty far gone. I'll try voice reagen. It kind of struck me as funny in a sort of meta kind of way because the video about how we're going to embrace AI and what the rules of engagement for AI are and here's AI fixing. It's like, yeah, yeah, of it. I used it recently too. I did a corporate video for a massive shall we call it international client, right who should really have the money. But they didn't have a soundo the on location, so it was recording two guys talking. It was a corporate video recording two guys talking, and they put lapels on them, but for some reason, rather than recording the lapels, they recorded off the camera mic and. It was it was awful. It was terrible, and like you, they rang me and said, look, we don't really have a budget for sound on this, but we're going to have to find some money. Can we pay you to fix this? And I went, we'll send it over and I'll have a listen. And I stuck it into pro Tools and sort of went, oh shit. And then as I was as I was about to draft the email, going look, I don't think I can do much more than you guys can do, it struck me I should chuck this in region. So I did, and sure enough they found the money to pay me to fix their audio. But it was, it was. And it keeps the videos think pretty good it does. It's not too bad. I had to. I had to sort of play with it a little bit. Yeah, yeah, a few things to fit it in. It's actually that was the other thing I was going to say was it's actually quite good. I can't because I was going, oh, that'll be the caveat on this is. I'll stick it in there and it'll come back it'll be all out of wat. Well, you'd have to use a little piece of AA to put it back, and. That's right, to squeeze it back in the time. But it actually it wasn't bad. And so using a few little tricks I was able to sort of get it back into place. But yes, you're right, though it didn't sound it. Makesn't too perfect, Yes, essentially. Yeah, yeah, it's it's it's like it's like nicely close on mic and sounding beautiful when they're in a room and clearly not on a mic and all that. It's very handy because I often get things sent to me from my daughter of course, saying oh can you run this through Regen for an audition because she's recorded in a bedroom or something in Melbourne. So it does come in very very handy. You know what I reckon? It's going to be interesting, right is I've noticed now when I get AI to generate a picture for me for my social media, it's watermarking it on especially on LinkedIn. It puts a little circle thing up in the top left hand corner of the photo and when you click on it, it says this image may be generated by AI. And I kind of wonder, what's the audio version of that? How do we you know, how do we do that? How do we sort of go I. Have gone through this rabbit hole because we've I don't know, we might have to scrub this whole thing. But basically, you know, people doing meetings review and approval, and there is a possibility that something leaks, and so when something leaks, they want to know who leaked it. Oh yeah, yeah, And so I've gone through the full rabbit hole of how to watermark eye And the reality is that it's in the audio band, right, because in this version of things, we wanted to survive being played out as speaker and back into a microphone. So yeah, it's it's playing with psychoacoustics and what you know, people can and can't hear and embedding, you know, And it's it's also a really tough game because it's like how little audio do you want to be able to identify your watermarks? Less audio you have, the more obviously watermark is but the more if the more time you have, you can bury it more subtly and still give enough redundancy for the data of your watermark to be still retrievable. So it's going to be more like a like a dog whistle something like that. Well, yeah, but that's really easily defeated, right, So if you stick it in the very very very high end. You just stick easily, that's right, exactly right. And then in the very very low end, the problem is the frequencies are so low that it takes so long to put any data in there because the low frequencies technically don't carry a lot of data. Yeah, and the cycles are so yeah. Right exactly, and also you can lop off the low end. So really it's like, if you wanted to survive as speaker, you got to put it smack in the middle. So you know, you can do all kinds of things I was playing with, but you know, follow the envelope and figure out how to hide in plane sight essentially is what it is. Do you think do you think do you think and this is really radical, don't get me wrong, but do you think the whole audio standard has to change? So instead of just left right channel, do you think we need a left right center channel and the center channel contains the metadata? Well, I think I think a couple of things. First of all, the watermark that you guys were seeing, it was a visible watermark everyone knew was watermarked. You can just watermark it being like this was made by AI and say that every ten seconds, and that's a blatant obvious watermark that people can hear. The other aspect of watermarking something is to make it so that you cannot see it usually, but you know, in the easiest sense, the way they watermark your picture and ai just there it is. You can see it with the naked eye, just like you could just put in this has been watermarked over the top of your your audio. But if you want it to be stealthy and unknown, that's a whole other thing. There's ways of putting data inside the picture that again you know, like you you see the picture for what it is, but if you run it through whatever decoder, it can find the watermark in the pixels. And same thing. You can stuff an audio code in audio that you as a person would not pick up on it, but if you have the right decoder, it can find that little bit of data that you've stuffed in it. So yeah, but funny enough, like like like little so I think, do you know where watermark in the paper? Someone realized that before the paper dried, if anything was sitting on the wet paper, then when the paper dried, if you put the paper in the light, you could still see the outline of the thing that was sitting on the paper when it was wet. And so that's how paper makers began to stamp their paper with their brand. They would let it dry and you put up the light and say, oh, this is came from whatever mill, this mill of that mill. Yeah, right, what about this is another thing if you had a one off password, so if you were received a voice but it was unusable, and then you have one off password that actually converts that file into proper audio. And now that I had the. Proper audio, I can send it off as a way file to everybody. Right, Yeah, how do you put something in there so embedded in it that it can't be gotten rid of without destroying the audio itself. That's the problem basically, Yeah, let's check. With ION and see what we can come up with. You know what, that's probably what it will take it ironically, that's where it will come from. Digital rights management. It's one of a digital version of that. I mean, I remember back in the day. To track commercials, we used to have to watermark a commercial and then some company would be running. They would just be scanning all the airwaves to track if the commercial's played, and there was just like so the commercial itself was watermarked, and then they would find their watermark and go like, yeah, it played or it didn't play. It's been going on for a while. I think this has been getting more sophisticated because if you feel that it's going into environment works something and you you know, you think about like the Apple iTunes things, here's your file and they really can't stop some from you know, ripping it, especially through an analog pass. Then you know, and if they do anything too much, they're going to harm the music. It will be it will be audible. It's and that's that was more my point, right, more than the sort of stopping theft of voice and all that sort of stuff. My interesting point with my point not saying my point was necessarily interesting. But the point I found interesting was that AI that they're now on LinkedIn, they're sort of going, well, they're just flagging stuff as AI, right, you know, they're sort of going, listen, just say, you know, this isn't a real photo, this is this is AI. And it's interesting because we had an example here of a couple of radio stations who have been using an AI voice to do their read their news without telling anybody, right yep. And the backlash has been huge, as it should be right and fair enough too, but I kind of that's my point is you know, how do we do that? You know, And it's sort of probably a similar system that what's going to take to protect voice. But interestingly, with those AI voices on radio, it wasn't the fact that they were AI voices that people that offended by. It was the fact that there was no disclosure that they were voices. And that's the thing. I think everybody would be okay if AI voices were disclosed, and there's one of two ways to disclose them. A make it, you know, just like sound like it, and then no one has to wonder. But yeah, I think that you know, when when you make images. I made a funny image with AI, you know, you know, the the the fish, the bass yeaheah, And then I was thinking about one of our politicians here, and then I was thinking about a thing that comes out your butt and I can the bass heard yep. One day we have to do an episode on how your brain works, Robert, Seriously. The scary thing is I think I know how it works. Quite disgusting, but but my brain works well enough that AI followed it perfectly. But when I got my image back, boom, it was watermarked. Ah wow. The AI company I used to generate the image, and I think that's that's fine. I don't care. But did you, but Robert, did you remain undeterred by that? Yeah, that's funny. Yes, anyway, I'm glad you the region worked because I do actually like it. I think it's get out of jail if nothing else works. Well, I guess he's now got Robert and me as well. But I mean I would, I would put my name to it. Yeah, it's it's it's not my first method and certainly probably not my preferred method. Sorry Gomez, but as an absolute last line of we must do something. Well, it's so far gone, at least in my situation. When it's so far gone that the you know, it's like if you're going through something that's forensics and you have to take there pick the words out of it, but it doesn't sound very good. That's one thing where you have to use the noise reduction. But at a certain point, when it's really far gone, you know that they screwed up by not recording the right camera. It's not noise reduction, it's it's all of it reduction. It's room reduction. And that's why Regen is such a good you know, term for it, because literally, back in the day it would have been let's redo this, let's ad R reperform it. Yeah, and so it's it's really it's like it's for those cases where it's not like a little bit of hiss or because the AI noise reduction clarity, et cetera has certainly gotten good enough that it can knock a lot of stuff straight really easily, and when it's too far gone, yeah, you have to just in a sense scrap it and have it be reperformed. And then you have this and it's yeah, it's it's interesting if there's a middle ground between the two where it could be sold. But I think, you know, for for like voice talent, I think it's very handy if you're for auditions. I wouldn't use it for an actual session. Of course, oh god no, please don't. Yeah, but if you're doing an audition, and it's like, I've got to get this thing in now, because I get people sending me stuff saying I don't care to do it on your phone, just send it in. It's like, yeah, but the trouble is you do it on your phone, you get sent in and that immediately the person listening to it. Whether they consciously make a decision or not, they will hear it and go that sounds like subconsciously. Subconsciously it puts you in a different place and you just. Go that shit and someone's recorded properly, and then you have your phone next to it. Who's going to win? That's right? I know. I don't disagree with you at all. And particularly when they're doing like you know, a couple of seconds, a couple of seconds. I think for auditions it's totally legit. Like I mean, you can do You could be in the middle of the zoo with chimpanzees and everything and you could basically record that and it could clean it up. Yeah, this sounds like a normal session to me. Maybe maybe I should sell myself as as Ai voice cleaning up because God knows, I've cleaned you up from a train and wherever else King Vodkas Robo Region I'll call it or something we call. It rob. Back in the day. It really did, like you know, I took hours after your life. Has given you your life back, Robin, Thank you man. The pro Audio Suite and Austrian audio recorded using Source Connect, edited by Andrew Peters and mixed by robo got your own audio issues, Just ask robo dot com. Text record from George the Tech Window. Don't forget to subscribe to the show and join the conversation on our Facebook group to leave a comment, suggest a topic, or just say it. Drop us a note at our website our audio site dot com. Hean