This is a very interesting website which upgrades audio. It sounds very incredible, here is a sampler using a John Lennon interview!
hoops://we.tl/t-Bq3H0Z6rkW
Very interesting! Sadly it does remove some detail and is mostly used for podcasts/interviews but this is very interesting still and will be useful for some!
Here is the site for those interested!!! - https://podcast.adobe.com/enhance#
Adobe Podcast - Audio Upgrading AI
-
moptophugger
- Posts: 582
- Joined: Sun Apr 18, 2021 9:41 am
- Has thanked: 2 times
- Been thanked: 4 times
-
theboxinargentina
- Posts: 301
- Joined: Fri Oct 08, 2021 4:12 pm
- Has thanked: 187 times
- Been thanked: 97 times
-
mojofilter
- Posts: 999
- Joined: Mon Apr 19, 2021 6:31 am
- Has thanked: 46 times
- Been thanked: 193 times
Re: Adobe Podcast - Audio Upgrading AI
Well, boys, it looks like the future is here.
This is track one of the Beatles' open-end interview for A Hard Day's Night. First track unprocessed, second track with AI. As Ed Sullivan never said, "America, judge for yourself."
hxxps://we.tl/t-64NF7Ei4MM
The AI did upsample the audio to 48K, but I resampled it back to 44.1.
I've just sent it some of The Little Girl Tape on a whim, and you would not believe the difference. It changes drastically from a noisy, distant cassette recording to a you-are-there kind of thing, although I think the AI is doing some damage to some of the words. Either the girls have really thick accents, or they are getting mashed by the algorithm, I'm not sure, but mostly I can understand them fine, but some of it sounds like gibberish. I'll have to listen to it louder tomorrow, it's 4 AM here. But this tool could be an outstanding breakthrough in background noise elimination without affecting the intelligibility of speech on top of it. I'm sort of gobsmacked. I need to look into this some more.
This is track one of the Beatles' open-end interview for A Hard Day's Night. First track unprocessed, second track with AI. As Ed Sullivan never said, "America, judge for yourself."
hxxps://we.tl/t-64NF7Ei4MM
The AI did upsample the audio to 48K, but I resampled it back to 44.1.
I've just sent it some of The Little Girl Tape on a whim, and you would not believe the difference. It changes drastically from a noisy, distant cassette recording to a you-are-there kind of thing, although I think the AI is doing some damage to some of the words. Either the girls have really thick accents, or they are getting mashed by the algorithm, I'm not sure, but mostly I can understand them fine, but some of it sounds like gibberish. I'll have to listen to it louder tomorrow, it's 4 AM here. But this tool could be an outstanding breakthrough in background noise elimination without affecting the intelligibility of speech on top of it. I'm sort of gobsmacked. I need to look into this some more.
Last edited by mojofilter on Thu Dec 29, 2022 1:45 pm, edited 1 time in total.
Eye. Eyesore a girl in my dreams.
Re: Adobe Podcast - Audio Upgrading AI
Quite incredible results! Insanely good reconstruction of the lost high frequency content!
Complete ATMP remix now released-
naturalkatsup
- Posts: 41
- Joined: Mon Aug 23, 2021 7:03 pm
- Location: Canada
- Has thanked: 23 times
- Been thanked: 28 times
- Contact:
Re: Adobe Podcast - Audio Upgrading AI
interesting stuff to me, but it recreates the voice than upsampling/upscaling it. ive seen attempts where it came back with garbled messes that vaguely resembled paul and john
i think its also worth noting it isnt the best with vocals meant for music, ive tried the free as a bird demo and paul simons "me and julio down by the schoolyard"'s vocals with it and it struggled HARD
perhaps the engineers behind MAL could attempt a voice upscaling model without having to recreate the vocal tone of the speaker, but thats just wishful thinking
i think its also worth noting it isnt the best with vocals meant for music, ive tried the free as a bird demo and paul simons "me and julio down by the schoolyard"'s vocals with it and it struggled HARD
perhaps the engineers behind MAL could attempt a voice upscaling model without having to recreate the vocal tone of the speaker, but thats just wishful thinking