Recommendations for Al Audio Restoration: Enhancing Low-Bitrate Drum 'n' Bass Sets

[Bow_Wazoo]

Hi everyone,

I'm looking for advice on how to restore or "upscale" some old Drum 'n' Bass livestream recordings from 2015.

The files are in aac+ format, and the compression is quite heavy: one file is about 70 MB for a 2-hour set, which puts it at roughly 64 kbps.

Are there any modern Al-driven tools or plugins?
I’ve heard of tools like iZotope RX (spectral recovery) or Adobe Podcast, but I’m wondering if there are newer, music-specific AI models that handle the complex transients of DnB better.
Any tips on specific workflows or software would be greatly appreciated!

 

[MaxwellsEq]

Hi everyone,

I'm looking for advice on how to restore or "upscale" some old Drum 'n' Bass livestream recordings from 2015.

The files are in mp3 format, and the compression is quite heavy: one file is about 70 MB for a 2-hour set, which puts it at roughly 80 kbps.

Are there any modern Al-driven tools or plugins?
I’ve heard of tools like iZotope RX (spectral recovery) or Adobe Podcast, but I’m wondering if there are newer, music-specific AI models that handle the complex transients of DnB better.
Any tips on specific workflows or software would be greatly appreciated!

It's an interesting problem. Ignoring AI solutions, I'd say that reversing such high levels of compression is extremely difficult, because what's discarded is content "the brain doesn't need to interpret the sounds". Putting them back requires you to know what's missing.

An AI solution might achieve something if trained heavily on the specific band. It would learn what the bass, percussion instruments normally sound like and replace the missing sonics with what "should normally be there". But it's not really recovering the missing sonics so much as pasting in alternative, similar sonics.

 

[Bow_Wazoo]

I checked the spectrum analysis, and the roll-off starts at 16 kHz. Being 46, that's likely where my hearing hits a hard limit anyway.
Maybe I should take a completely different approach: there’s still Mixcloud Pro or Premium. Perhaps the bitrate is a bit higher there.

Mixcloud

www.mixcloud.com

 

[AndyMorris]

I'm looking for advice on how to restore or "upscale" some old Drum 'n' Bass livestream recordings from 2015.

I'm interested in any solution that comes up but it also raises a philosophical point:

What are we trying to restore to? DnB records hardly have an untreated note in them. DnB DJs, much like Reggae DJs use the sound system, reverb, echo and equalization to make the sound their own, pirate radio (I'm that old) and streaming DJs do the same. And they do it with the bandwidth in mind.

So there is no 'real' sound to be recovered. what you have is the artifact that the DJ put out.

The nearest to a 'real' DnB sound would be listening in a packed club, to a speaker stack the size of a fitted kitchen. Not something one could recreate at home, unless you lived on an island, or maybe Norfolk.

Of course old 78s and old Jamaican pressings are a completely different thing and can be vastly improved. </Irony>

 

[Bow_Wazoo]

Hi everyone,

I'm looking for advice on how to restore or "upscale" some old Drum 'n' Bass livestream recordings from 2015.

The files are in aac+ format, and the compression is quite heavy: one file is about 70 MB for a 2-hour set, which puts it at roughly 64 kbps.

Are there any modern Al-driven tools or plugins?
I’ve heard of tools like iZotope RX (spectral recovery) or Adobe Podcast, but I’m wondering if there are newer, music-specific AI models that handle the complex transients of DnB better.
Any tips on specific workflows or software would be greatly appreciated!

As I have just learned, Mixcloud uses AAC+ at 64 kbps by default.

 

[Raj21p]

Unfortunately, with heavily compressed 64kbps AAC+ material, no AI tool can truly recover information that was never there — especially with fast transients and cymbal detail in DnB. That said, tools like iZotope RX, Adobe Enhance, and newer AI spectral repair plugins can still make the sets more listenable by reducing harshness and smoothing artifacts, but I’d approach it more as “restoration” than true upscaling. Also be careful not to overprocess; aggressive AI restoration can easily smear percussion and kill the energy of the mix.

 

[Curvature]

It's not restoration. This isn't an analog process.

There's certainly nothing automated.

The most important AI tool would be one capable of spectral decomposition, breaking down the music into parts, including separately recognizing the reverb, crowd noise, hum, tape hiss, vinyl clicks and pops, etc. along with all intentional components like percussion, bass and melodies. Effectiveness of isolation will vary of course. Each of those parts would then have to be deleted or exported and separately processed by you, as a mono or stereo source.

The next stage would involve mixing everything back together. It would make the most sense to use a spatial audio-capable DAW or container, specifically for the object-based panning, to avoid the issues that come with stereo mixing and standard panning. Rendering down to stereo afterwards is no problem.

I would use EQ liberally and an expander when necessary, and search for tools that can remove phasy, wavering or other corrupted components of sounds.

A big professional archivist move would be to find the originals to use as a reference for mixing and other decisions. Anyone who knows electronic music knows how difficult finding originals is.

Actually using the original music to take the place of the set recorded versions would pose another set of difficulties. Depending on the conditions of the set recording the originals might sound far too different to use outright. Hard to say what the best decision would be without getting down to the work.

This is more like translation, requiring an artful hand and interpretation, rather than something algorithmic, machine-driven and measurably accurate.