What this thread is not for. Discussing feelings on current state of AI audio. There are other threads you can voice your opinion on that.

What this thread is for. Is I wanted to compile a list of tools people are using to fix the current state of AI audio. Over time these algos will sound better but currently only a good year into this this tech. A lot of the outputs are distorted, highs are cut, they have that underwater "chrip" sound. I was wondering if anyone has some suggested plugins or tools they are using to get better quality audio out of these AI algos.

Here are what I have been using with good results:
Zynaptiq - Unchirp:

This plugin was made to improved poorly encoded mp3s streaming music etc. It helps with that "chirp" sound that is in current AI music. Also helps bring back that uppper high end that has been cut off in most AI music.

Zynaptiq - Unfilter:

This helps bring out some of the filtered content that might be buried within the diffusion process of generating AI music.

A few other plugs in my chain that help clean things up are. Gullfoss, Wavesfactory - Equalizer and Spectre.

I was hoping for something like Khaelis - UNOV to remove some distortion artifacts. but that plug seems to be snake oil. Is there anything similar?
https://www.kvraudio.com/product/unov-by-khaelis

One last one I like is Ozone - Master Rebalance. It can help bring vocals more forward in the mix.

I would love to hear about any other plugins you guys might suggest to help in this process.

 

never had much success with Unchirp but I salute your ingenuity in resurrecting it! Can you get stems on these projects? EQ has worked wonders in the processes of cleaning old interview/dialogue passages using ai... not 100% 'underwater' artifact removal (yet), but pretty convincing masking. Rx11 also useful to a degree - interesting thread

 

yeah unchirp has been hit or miss. Its hard to find out how to use it online. Ive played around with some settings but one tweak one way effects the sound. Its a delicate balance.

Yeah you can use any of the stem seperation algos out there to get stems. But these arent 100% yet either. So boosting can bring out those underwater artifacts in the stems.

Its an amazing time to have all of these tools! They are just in the early stages.

 

I did want to mention there are some Algos out there to help with this. But I havent gotten the best results yet.

https://replicate.com/nateraw/audio-super-resolution

 

ive been using stem splitters, then audio to midi and replacing everything lol

 

So you first want to use ai tools, and then fix the outcome?

 

Yes per my original message above. I want to improve the audio quality of the current state of AI audio outputs.

 

This is great for fixing mixes and making alternate mixes. Obviously for remixes, too; but that is just to get starting samples. For fixing mixes it is extremely helpful. You are either almost re-tracking the parts, or just supplementing and backing them up. In Logic, you create a duplicate of the track, press E to open wave editor, analyze the audio for flex editing, switch it to Flex Pitch and then Create Midi Track from Flex Pitch Data. Load your new instrument on that new track and you are ready to go. It takes less time than it did for me to type this. (after separation). I'm not using Logic 11 yet.

 

Awesome! I use Rip-X, EZkeys, EZbass, even moises to convert to Midi from generated content.

https://huggingface.co/spaces/Audio-AGI/AudioSep
and
https://tuanziai.com/en-US

these are kind of cool for isolating just a spacific instrument in generated audio

 

I've found running ANYTHING the computer generates, especially the modern stem separated machine learning generated audio, through real tube hardware, it really adds to the audio in a very pleasing way. I've had good success with the Coil Audio 286a preamp in line mode (dialed in) and the Black Box HG-2 (the low setting with alt tubes shined) easily tame that fake digital top end and allow for the mid to be harmonically enhanced, thus becoming usable in modern song production.

There was a noticeable before and after hitting the tubes, never have tubes shined more IMHO.

 

thats genius! I bet those tube harmonics help soften and create some of that top end in a more natural way.. Ill try to sum through my dangerous D-box.. its the closest thing at my disposal right now

 

No really a solution to your problem but I find that people don't really notice the typical AI artifacts, I've experimented in some gigs and people were for the most part completely unaware.

 

The "bandmate" function they added to EzKeys2 is very good. I'm usually fixing bass with the Midi channel so I use Modo Bass 2. For more complicated detection, Melodyne is apparently the preferred method by many people. I'm just not a big fan of the program itself. But they say it will do polyphonic detection, which is not something Logic is great for. EzKeys can still get some decent results from more complicated audio files.

Melodyne seems to be preferred by people doing this stuff with guitar parts.

 

Last edited: Jul 10, 2024

After reading this I ran into the Techivation AI-Clarity Plugin..I also use Chirp with like you said hit or miss. Rob Papen UniMagic works too but probably the end result is too vivid in some cases but tends to be enough.

 

Man thanks for suggesting AI-Clarity! this really helps fix the "boxieness" of the sound. Seems like UniMagic is adding some unison reverb delay space. Its now part of my clean up chain!

 

Ive been digging around the internet and old AS forums. I wanted to update this thread and add some more tools I have found to help. Also S/O to @clone for the tips:

This helped alot! I found out about stereo tool via this:
https://audiosex.pro/threads/any-software-for-frequency-generation-spectral-recovery.56478/

Noise Removal - BCC: @clone :
https://audiosex.pro/threads/removing-noise-from-live-bootlegs-newbie-warning.72032/#post-725506

Zynaptiq - Unchirp Tips:
https://audiosex.pro/threads/help-with-zynaptiq-unchirp.71857/

 

Adding Thimeo stereo tool to this list! TBH its probably the best tool for cleaning up AI music that I have found! It can denoise, unchirp (dequantise), spectral repair. It pretty much does everything needed to bring the sound back to a quality to apply other processing after.

https://www.thimeo.com/stereo-tool/

(I also found a way to make it sister site work on mac )

 

Trying to split the stereo output into stems highlights the limits of AI music. Everything is undefined and murky. The "other" stem holds too much content.
I know udio paid for version has stem output. Not tried it yet.

 

its trash. there are better stem separation options third party.

 

udio is just using all the free stem seperation algos like ones i mentioned here:
https://audiosex.pro/threads/uvr5-the-best-ai-stem-separation-algo.69463/

you can just use that with generated music. Ive oddly had good luck using this ios app called Ripple to separate AI generated music. It uses a different algo than demucs, MDX. etc.

But if you clean up your audio before hand it stem separates better.

Yes AI music is limited today. but we are a little over a year in on some of these algos. With more compute and fine tuning things will get better with time.