Hi,

So I am working on a source which has upper hard cutoff at 16.2 kHz, the first thing to do was to upscale it to the 16 bit CD quality i.e. 22.1 kHz. iZotope RX8 has this feature to rebuild the frequencies called "spectral recovery" but it really doesn't do anything except for writing the very low quite frequencies in the stream.

I looked around and found Perfect Declipper does what it says on the tin but the problem is, it's meant to do this on the fly in live performances which means the emphasis is on speed rather than on accuracy. It does the job but far from perfect. So is there anything else out there which could do this ?

Any help will be greatly appreciated!

 

I can only guess you can restore some upper harmonics using various analog saturation plugins and EQs,
on the other hand, many people don't hear above 14-15kHz, so depends what you actually want to reconstruct above that area

 

Problems are so...
If you start to recover those almost ultrasonic frequencies with exciters or so, you may create aliasing which is worse. So you need higher sample rate and/or oversampling.
Second, if your cutting frequencies were due to lossy codecs used in audio tracks, songs, so it can even cause detail problems, bubbling watery sounds in mids, highs...

 

There are some machine learning research ongoing on this task, take a look at this github search:
https://github.com/search?o=desc&q=Audio+Super+Resolution&s=updated&type=Repositories

 

I'm not adding anything new, but since it's related: you can search for tools developed to restore low-quality MP3 files (low-quality lossy encoded files in general).

It's not the same than your hard low-pass but similar and quite researched. Tough I doubt you can beat what you achieve with tools like iZotope.

 

I think zynaptiq got a couple of plugins that can help with that

 

Perfect Declipper synthesizes the 2nd harmonics only. So the added highs sound somewhat empty and too soft, it is less a matter of low accuracy.
You should use ACON Digital Vitalize (a DLL part of Acon Digital Acoustica 7.2.8, can be also used as an independent VST). It is highly configurable, intuitive and efficient (creates 2nd-7th harmonics).
If you feel oversampling is needed, you can use ACON Digital Vitalize through DDMF Metaplugin 3.3.6. (offers oversampling of 2x-8x).

 

Last edited: Nov 29, 2020

Try this method

 

@tzzsmk @mild pump milk

Thank you for the expert comments. It appears like it is indeed due to the usage of lossy codecs. I experimented with Perfect Clipper and it appears the newly created/restored frequencies are adding to the character of the sound. The output has better width and depth too.

But I need to keep in mind what you said to prevent it from becoming worse than it already is.

Your input it highly appreciated. Thanks again!

 

I'm no coder and I don't know how to clone a project from git, never done this before. Thanks for the valuable reaction though. Something to look for in the future.

 

iZotope is no good in this regard. But thanks for the pointers. You make me believe many such tools exist. I have more replies down below
I'm gonna try them.

Thanks again, Xupito.

 

Merci, c'est cool monsieur Jarre

 

Also you should realize, does your sound really requires full bandwidth? Some sounds don't need it - sub basses, for example, they are at lowend area. Vocals don't need subbass frequencies in most cases (only if it is a male jazz singer with deeeep bass voice).

If you produce electronic, dance, pop music, not acoustic/live music, it is ok for lossy stuff - you make sampling, so it is an electronic something. As an example, The Prodigy... Liam Howlett is obsessed by analog retro/vintage synths, processing, vinyl samples, old vintage digital sampling style, but their No Tourists album has a lot of mp3 samples, a lot of lossy stuff (Boom Boom Tap vocal sample was taken from YouTube webcam lo-fi source "sung" by a freaky voiced kid), some lossy samples can be heard/seen on The Day Is My Enemy album and maybe others... It is ok....
A lot of music made with Vengeance samples - cuts from unknown audio sources, 44100 16 bit samples... But, from the first packs up to the latest ones, there are more than 80% of stuff are just LOSSY.... Listen to side channel (mid off) - bubbling watery sounds, like they were 128 or 160 kbps upconverted to 44100 16 bit wav pcm..... Some popular libraries have lossy sounds inside too, but not too much... Way less...

I don't think your audiophile listeners will be analysing your audio.... Usually they "hear" lossy stuff only by eyes, in RX, not in real life by using golden ears.

 

and see this
http://www.zynaptiq.com/unchirp/unchirp-overview/

 

Last edited: Nov 29, 2020

One stuff to check delossifiers quality is to take a lossy file with low bitrates such as 96kbps, 112kbps, 128kbps, 64kbps....and to listen, what your delossifier does and how it does.... If it is too artificial, doesn't improve, so you don't need to recover all those useless ultrasonics....
Unchirp is not too bad, but it does improve at 40-60% sonically or less.... A lot of artificial stuff you can achieve tho..

 

Thank you from the core of my heart pratyahara. I never knew such things existed. I'm moving forward with your directions.

Okay so I got myself Acon Vitalize and Zynaptiq Chirping. Zynaptiq feels like scrubbing the bass and adding too much treble.

Did some extensive testing with Acon and Perfect declipper, here's the results.

This is the lossy source



This one is processed with Acon Vitalize:


Here's the one from perfect declipper:


Lastly, this one is processed first with perfect declipper, then with Acon Vitalize:


The last one where both plugins were utilized appears to sound closest to the original untouched lossless source. But I have no idea, given the complication of the process, if it is alright to process it with two delossifiers. I can't spot any artifacts though.

So the question is, if it is alright to do it this way, meaning processed through two plugins. And will I be getting the benefits of Acon's full spectrum harmonics elevation if the stream goes through perfect delipper first ?

Thank you for making me come this far.

 

Excellent find Roject, that's a neat little Audacity trick. Definitely a saver.

Thank you for the valuable contribution to the thread.

 

That's very true mild. But even if it doesn't need full bandwidth, I still undertook the project to learn something new and explore more possibilities.

 

Okay I did this and the processed audio definitely has more width and depth compared to the lossy source but is nowhere close to the original lossless source. I don't understand, after writing the frequencies how could it not sound like the original. However, with the ethnic sounds, it's also elevating some undesired frequencies as well but that's a given since it's mapping the audio from what is already present.

 

Yes, it's alright. Those two plugins are complementary. It is much like using the same plugin in two passes with different settings. Using the both plugins you get amplitude (Perfect declipper), as well as full frequency content (Acon VItalize). And you should render Thimeo first, then Acon (It builds on what Thimeo did).
By the way, you did not post the parameters you used for setting the plugins (if the source is lossy, you should use dequantizer also).