I have never even tried do use ST for handling the input signal, I just simply use the gain control of the editor. So, I don't know about your problem, but it is of minor importance. If ST makes you trouble, just reduce the gain in the editor first.
Also I use ST modules one by one, which allows me to use the maximum settings, starting with Configuration/CPU&Latency (for instance 'Audio Quality' set at 150%), and then Extreme Tweaker user interface to set each module max. quality/optimal settings (for instance 'Dequantizer' both 'precision' and 'quality' set at 100%) etc.

 

Yes, that makes sense, good advice. I suppose I wanted to make ST as much of a 'one stop shop' as possible but doing some preparatory work beforehand to reduce gain by about 3db is no big deal.

Thanks for taking the time to reply, much appreciated.

 

excellent thread! pratyahara, if u don't mind me asking, what's the software u are using that appears in ur screenshots?

 

Platinum Notes, which uses Izotope Exciter and IRC Limiter Technology can do the mp3 job (in most cases)

 

Here's how to drastically improve 128kb mp3…

1. Select all the low quality mp3 files you want to enhance
2. Press "Del" key
3. Clear recycle bin
4. Go to ebay and buy the CDs dirt cheep
5. Rip them as .wav

 

Actually @pratyahara has managed to improve the audio quality significantly. Well done.
I found the song on amazon or somewhere similar where you can listen to a few seconds and I'm not sure if it was just intentionally lowered quality mp3 by the site or the album has been recorded that badly, I prefer keeping and listening to the new version by pratyahara

 

@pratyahara hey there, been reading this thread, I'm really impressed by your generosity..

Could you please explain how do we discover "double filtering" issues? I'm using RX software to compare mp3 files downloaded from different places (mostly YTube and SoundCloud); how do I discover and distinguish "right" mp3's from "wrong" ones? I'll be very grateful if you explain it based on this or any other image...

 

This is very simple. Try to use only True 320kbps MP3 with True frequency range for 320kbps which is between 20Hz - 20kHz.
You can easly check any mp3 is it true 320kbps bitrate on the diagram like this one on the picture on the right side.
Like you can see. This is not 320kbps mp3 file on the picture.

Pattern:
MP3 320 kbps Cut-off at 20 kHz
MP3 192 kbps Cut-off at 19 kHz
MP3 160 kbps Cut-off at 17 kHz
MP3 128 kbps Cut-off at 16 kHz

Conclusion - The worse file bitrate is then it loosing more highs and is more muddy and often at 128 kbps bitrate and below have artifacts when there are sounds below their frequency range.

 

If you mean 're-encoded' you can see that in Fakin' the Funk by difference between the nominal bit-rate and the actual one. Actual one is almost always lower /except the case when someone re-encodes a file to the same or lower bit-rate/.
You can judge the mp3 quality of non-fake files of the same bit-rate by the white line in Fakin' the Funk -the less irregular and less tilted it is, the better the encoding.

Re-encoded

Low quality encoded


For FLAC files to check if they stem from genuine WAVs you should use any spectral viewer (like Spek 0.8.2) and see if it was filtered below 20/22 kHz.

 

Last edited by a moderator: Jul 17, 2023

Those are good examples. Guys remember also that FLACs and WAVs can be faked too. And True Flac or Wav files are not cutted. They can be below or around 22kHz and also below 20 kHz, but they shouldn't be cutted just at 20 kHz and for sure under. In most cases there should be some spikes below 20 kHz. It will be cutted at 20 kHz only if artist decided to do it but it is uncommon.


@GeekedGlitch Looking at those diagrams try to find something like solid block. Like Ignore spikes and artifacts.

For me first example isn't True 320 kbps MP3. It is faked. It is 192 kbps.
It is 192 kbps file converted or recorded as 320 kbps. Waste of space on the hard drive. Artifically bumped size.
Like you can see this is not just solid block cutted at 192 kbps.
It is cut under 192 kbps range and there are 2 smaller blocks with higher range around 192 kbps.
You can save file like this not to constant bit rate but various bit rate to save even more space on the hard drive.
So you can save the example from the picture as 192 kbps vbr to save even more space.
It is maybe not so much but if for example you burning mp3 cd with tracks for listening in car if you do this to all possible tracks it can give you few more extra tracks!





Second example isn't that obvious and is harder to distinguish. But if you look closely you can distinguish where artifacts started.
You can see that around 160 kbps sound started to be stretched and weaker more and more and solid blocks transform into spikes.
For me this file is 160kbps because below it there is partially lack of information and there are many artifacts.
I guess it will be better to cut this and save file as 160kbps than have some ugly artifacts like this.

 

No, it would be not better at all. You suggest re-encoding a mp3 file in order to improve it... It means yet another bout of compression, filtering, pre-ringing, stereo image distortion and quantization errors.
In fact filtering of harmonic range is least important because 90% of people can not hear anything above 16kHz. Sound degradation due to mp3 encoding is due to (in order of degree of detriment):
1. quantization (harsh and edgy sound)
2. high compression (flat, narrow, cold and dry sound)
3. loss of transients (less punch)
4. psychoacoustic filtering (removal of concurrent or masked sounds)
5. loss of detail
6. pre-ringing
7. stereo image degradation
8. increasing DC offset.

 

Last edited by a moderator: Jul 17, 2023

But what if I open this file in some DAW and then cut it or even didn't touch and just save as 160 kbps MP3? Is it avoid all those compression, filtering etc. further degradation or no and result will be the same as conversion with some small random program?

 

What I have found most useful with stereo tool that it can return some semblance of dynamic range to stuff that has been rammed thru the "modern mastering wringer". No, it's not going to return bits to you that were lost from MP3 (or other lossy formats) compression however it does a decent job (when used correctly) counter acting the artifacts that are a result of the lossy compression process. I have been using it to restore the dynamic range and some fidelity from higher rez streams recorded from 24/96 and 16/44 streaming services. This is sort of a dicey affair seeing as a lot the content these services offer (even with older content that was not originally hosed up by "modern" bogus mastering process) have been processed twice or three times over using for-shit modern mastering practice. The resultant content can end up being completely munged from all the crap that was done to it to the point where is just sounds like your being beat in the head with a lead pipe, over and over again and you just want to make it stop. This isn't the fault of the streaming service itself per se but never the less I have found this to be true even of content that is offered as fee based downloadable files. In fact I have found very little to no differences in streams of the exact same content recorded in real time from the service(s) (in the proper format, at no additional cost to the regular service fee), to that same content acquired as a fee based downloadable file.

Enter stereo tool... I have used stereo tool successfully so far on some lesser hosed content that did return a great deal of dynamic range along with some return of mix clarity where original data was not lost just buried under a mound of modern processing sludge. As long as there is still data present in a piece of content that wasn't completely wrung out of it by garbage present day mastering practice, stereo tool does seem to have the capability of working with what is there to produce a better sounding result in a variety of cases.

example - waveform before;

https://audiosex.pro/attachments/sample-before-wav.40971/?temp_hash=8c748b9a2269624abc4848f8d58c9e27

example waveform after;

https://audiosex.pro/attachments/sample-after-wav.40972/?temp_hash=8c748b9a2269624abc4848f8d58c9e27

A lot of the dynamic range has returned and no more flattened peaks. I stopped using MP3 a few years ago and went to FLAC instead for archives since there is now access to higher rez source files.

 

Last edited: Jul 18, 2023

Any "save" to mp3 format means mp3 encoding. So if you open mp3 file in any program it gets decoded to WAV. No matter if you do something to it or not afterwards, saving that WAV to mp3 is done by mp3 encoding, so you are re-encoding the original mp3.

 

Interesting thread, very useful information. I'd like to throw in ApUnmask by Apulsoft. It probably won't remove/hide artifacts from mp3 , but it might counteract the "burying" of certain frequencies. Not sure if this would work though, since, for example, adding dithering before video encoding will just make the encoder be more aggressive to achieve the target bitrate and compress it MORE. Either way it's a really good, actually useful, psychoacoustic enhancer using tech found in codecs.

 

I am curious what to tell have AI technology about it.
Isn't yet any trained models for this area?

 

I've read post on audiosciencereview where user tried to create something like this. I couldn't find this post right now, but I remember the thing.

The thing is that audio is much more complex than image or even video, and something related to AI audio enhancing-recreating would require crazy amount of computer "brainpower". So that guy created a model, but he couldn't experiment further because of required GPU (or CPU, I don't remember) horse-power that he was lacking

 

Yeah those Nvidia gpu's are expensive beasts. Prices like for a super sports car.

 

I've been extensively testing Stereo Tool and holy hell it does magic to a lossy audio file...the UI takes a bit to get used to but once you do it's definitely worth it. The CPU usage is so low for what it's doing too...definitely recommend
Edit: Some of the effects aren't properly latency compensated...

 

Last edited by a moderator: Jul 19, 2023

If I remember it right, he had 2060 or 2070, and he needed something like 4x GPU rig to make his model work. So we're talking about +- supercomputer. But I will try to find that thread and give a link