remove noise with A.I. in pre recorded audio tracks ? Help

[schmutzbrust]

Hello,

don't mind my maybe bad english. It's not my native language. I'm planning to dub some old movies with another audio track, but the tracks come from old tapes and you can hear a lot of noise in the background, exspecially on the start of the vhs, where there are no sounds just the noise. :/ I read so much about A.I. noise reduction, but in all cases, they talk only about live recordings (like youtubers streaming outside and want to get rid of background noises). The results are quite good and couple days ago i saw a video to even split an audio track in speech, effects and more. So if this all is possible, why i can't find anything about just to remove noise properly. :/

My audio tracks also have a few seconds at the start, you can only hear the noise, so i assumed thats a good example for a tool to put on.

I already checked programs like audition, but my results were not satisfiying. It removes the noise in silent scenes but you can still here it well in all other areas. :/

I must confess i'm not that deep into audio tech to understand everything and i don't have the time (work, kids other stuff) to spend hours and hours in it. :/ So i was loking for an easy solution and this A.I. stuff sounded to good to be ture.

So maybe i just overlooked something, that fits to my needs or is it really that difficult? :/


thanks a lot for your attention

[Cornucopia]

It really is that difficult.

Let's put it this way: Describe the noise. The more accurately and thoroughly detailed you can describe it to differentiate it from the signal (or vice-versa) the more likely you can isolate one from the other. Conversely, the less exactly and thoroughly, the less likely you will have success.

It is not impossible, but it is nearly so. One of the biggest factors is that "noise" doesn't just mean one thing, or even a handful. No, it can mean any of a hundred thousand things or more, or some combination of them. That's why AI might work with a few example special cases, but doesn't really work well with any given general, random recording. AI has to have those accurate, thoroughly detailed examples with which to learn from.


Scott

[davexnet]

Perhaps post some samples

[lordsmurf]

"AI" is still just marketing nonsense,
My toaster has AI.

[schmutzbrust]

Thanks for your answers, i'll upload some samples tomorrow...

[Hoser Rob]

"ai" is still just marketing nonsense,
my toaster has ai.

thank you!

[schmutzbrust]

So, here are 2 samples (I'm sure, i have some worse ones, but can't find them atm. :/


https://www.mediafire.com/file/3e96te4399nihb7/Arm2003.wav/file

https://www.mediafire.com/file/1my4wf8306ukfw2/IYH17.wav/file

[davexnet]

Are you talking about the background hiss? Audacity NR seems to help, here's a couple of short clips,
I started out by making it a little louder and that became my "original" here

[schmutzbrust]

Thank you. That actually sounds a bit better now. So i assume, much more is not possible? Seems my expectations were just too high at this? :/

[davexnet]

Thank you. That actually sounds a bit better now. So i assume, much more is not possible? Seems my expectations were just too high at this? :/

There are all kinds of noises going on in those clips. Audience noises, general distortion, etc, etc.
Probably reducing the background hiss is about the best you can hope for and it takes a bit of work/time to do it
properly. There is no magic "AI" solution.