Huge audio improvement for free?

So you're obviously also running linux by now, and you have opted for a gtk-based desktop environment, whether Gnome, or Unity, or Mate, or Cinnamon, or XFCE.


And you feel the urge for really good quality hi-fi audio, but you want real audio quality, not fake “pay this much for better audio quality” crap.




Here's what to do:



You install “audacious” and all it's plugins and it's lib, and you install “sox”, which doesn't refer to the Sarbanes-Oxley Act, but stands for “Sound Exchange”, and “soxr” and the sox-plugins.



You start up audacious, you indicate the path to your music library and let it import it, after which it will be ready to go.



Before playing music, you go to the “Output” menu, “effects”, and select “crystalizer”. Open the Ouput-effects menu again and now you'll see an item “settings” under “crystalizer”. Select this and set the crystalizer to 1.2 and close the setting. This setting will undo the exaggerated compression of MP3 files. Then go to the Output-effects menu again and select “Extra Stereo”. Go to the Output-effects menu again and click on the item “settings” that has now appeared under Extra Stereo. Set it to 1.3 and close the setting. That will compensate for the stereophonic effect loss due to MP3 compression. Then go to the Output-effects menu again and select “SoX Resampler”. Then go to the Output-effects menu again and select the item “settings” that has now appeared under Sox Resampler. Set this to 96000 Hz. This will upsample the audio source from 44100 Hz to 96000 Hz. Then go to the “File” menu, then “Preferences”, and in the “Audio” section, set the bitrate to 24 bits. This is so that the processed sound channels through your audio hardware in a higher definition and keeps the quality intact.



Now listen to the MP3 file of your choice and enjoy on your standard audio hardware.


Feel free to compare with standard MP3 playback, for instance by opening the same MP3 in another player, or by playing it back on a Windows system with a very expensive bling audio card, and feel free to laugh at people that just spent 150 USD on a sound card that doesn't do half of what some intelligent real-time bit-shoving in open source software can do.

Feel free to share your experiences with this on this thread in order to convince others.


Will this affect the natural buttery goodness of FLAC files? I have a mixed library.

Upscaling to 96000 from 44100hz and 16-24bits will do absolutely nothing once a track is finalized lol. The main reason why mix engineers would choose 32bit floating point or 24bit in production is mostly for headroom and not sonic quality. And how is a crystalizer anything but artificial enhancement? Post this on head-fi and see what they say. Hell, a lot of the posters on there would argue that 320 and flac are very very hard to tell apart in a true A/B test. 

Use a software EQ?

That's what I was thinking. Like converting MP3 192 into 2kbps FLAC; stilll sounds like shitty 192kbps MP3, but is just in a different format. Upscaling from 44100 to 3 million will still sound like 44100, just with a lot of duplicate signals (aaaa, bbbb, ccccc, dddd, etc.)

Not if it's interpolating.

Then it goes from:

1  5  7  6


1  2  3  4  5 5.5 6 6.5 7 7.75 7.5 7.25 6  (but this is linear interpolation, it would probably use cubic or something else I can't do in my head on the fly)

Still doesn't sound like a 192 original mix, but it's much better than duplicating signals.

Kinda of like anti-aliasing for music, I guess...

FXAA for music. Except it doesn't make your ass-sounding music sound even more like ass.

They both make up stuff they don't know, but FXAA subtracts data. Interpolation adds data that may or may not be accurate, but it will be close.

All criticism is technically true, but still it works, because of the golden rule of science: a clock is precise if you know how much imprecise it runs. MP3 compression is not a secret formula, it is known exactly how much the compression impacts the original signal, so the crystalizer can actually undo the compression, and the SoX upscaler can actually interpolate what is most probably deleted with a high degree of probability, the 24-bit output is just to output the corrected signal without affecting it too much. End result: the weakest link of the audio reproduction has been reinforced. MP3 compression, even from 320, is still the weakest link, much weaker than cheap integrated circuits of built-in soundcards. And the filters used in the Audacious plug-ins are much more powerful and use much more advanced algorithms than tiny DSP chips in so-called "audiophile" audio cards, which use fixed dumbed down algorithms to do similar corrections, but without the upscaling and 24-bit output, just to reproduce them through an underpowered analogue audio path just the same. The difference between the audio path of a 150 USD consumer "audiophile" or "gaming" soundcard and standard motherboard audio, is much smaller than the difference between straight MP3 file rendering and processed MP3 rendering, and the difference between audacious plug-in processing and standard small DSP-chip rendering is much bigger still.

So fact remains that the biggest improvement for MP3 music can be made through the audacious plug-ins as described, whereas adding a better audio circuit inside a PC and powered by 3.3 V (PCIe cards), 5V (USB cards), or 12V (PCIe cards that use an extra floppy connector or externally powered USB cards), is only a small improvement that still doesn't work like it should, because in order for an integrated circuit to provide anywhere near the headroom an original analogue signal would require for anywhere near fidelity reproduction, would be around 15V, which is actually what some professional audio interfaces use by doubling the power from 12 to 18-24 V.

What's more, the DSP processing of some commonly used DSPs in cards like Creative, Asus, etc... are also known, and can be reproduced exactly by plugins, that are also available for free. So even for those that don't trust the advanced algorithms used in the Audacious plug-ins (which are an improved version of Creatives more powerful DSP processing rather than Asus' very basic DSP processing), there is a simple free solution that is still more powerful than even a 1000 USD external Asus "audiophile" sound card.

And for those that still get ear cancer from that (and I must say I understand that, I still like to listen to analogue recordings on a huge traditional Hi-Fi chain without integrated circuits that I spent more than 10 years putting together to get everything just right for my taste, I'm not saying MP3 or even high definition digital recorded music is the best thing ever, because it's not, there is no substitute for analogue recording and reproduction without any digital medium), they don't care about digital music enough to even listen to music on a PC anyway. And it's not because you know that analogue is just better in every aspect, that basic improvements of digital are not useful, I know the huge difference between playing guitar through my tube amps and between playing through some - often far more expensive - digital processing kit. There is no comparison. But that doesn't mean that the digital processing stuff isn't useful. An Axe-FX or Kemper is just good enough for some fills and fast fixes, and saves a lot of time in an audio production, but the real stuff still comes from full analogue amps and mixing consoles with analogue compressors and EQs, and preferably the recording is done on large tape trackers instead of on HDDs, but sadly that just costs too much money in production.

And to those discussing the difference in precision between 44100 Hz sampling and 96000 Hz interpolated, think of this for a second: when you record even in 192000 Hz samples, you still throw away 99% of the original signal, because every one of those 192000 samples per second are almost infinitesimally small, and even 192000 times almost zero per second is still almost zero. So yeah, there is nothing you can do to the heavily mutilated signal with intelligent interpolation that hasn't been done to it already by the recording technique itself, and if interpolation serves a better reproduction, then that's a bonus. An 1:1 intelligent interpolation is acceptable in most interpolation technologies, from upscaling of picture to upscaling of video streams, larger than than, artefacts will occur, but still nothing to really worry about.

Allow me to demonstrate this further by an example from photography: do you really need a 60 Megapixel medium format to take pictures for large billboard graphics? Or can you even upscale from a 12 Megapixel source with bicubic upscaling interpolation and hardly even tell the difference? Can you tell which billboards where made from a 60 megapixel source or from a 12 megapixel source? Exactly... it's nice to know that there are problems, but it's not wise to look for problems in solutions that solve the bigger problem and invest in solutions that don't really solve the bigger problem!

I didn't announce this trick as an audiophile solution, because digital music never is an audiophile solution, just because by AD conversion, 99% of the original signal is just thrashed and lost forever, unlike with full analogue audio processes, but at least it reinforces the weakest link so that there is a huge improvement in audio quality, an improvement that cannot be made by adding a static algorithm based small-chip DSP or a bigger-by-still-too-small integrated circuit analogue audio output path.

EQ-ing the signal is a far inferior solution to intelligent dynamic reconstruction and intelligent interpolation, and digital EQ-ing in a computer should be avoided at all costs, because it degrades the signal heavily and static digital processing causes all kinds of artefacts (intersample distortion, etc...).

I find this thread interesting but way over my head. :(

what's linux

Holy crap man. Everything you write is long winded and usually incredibly accurate. So in short, if you start with a crap source you get crap output?

I've wired up a couple guitar tube amps and am considering a HIFI amp, but seriously, unless I'm spinning vinyl will I even notice it?

I wonder if your can do that with Viper4android . Well , you know since its android . If someone can figure that out . I will learn how to use linux .

Would this have any impact on lose less audio, such as FLAC?

SoX resampling will, crystalizer and extra stereo are MP3 "repairs", they are counterproductive on FLAC files.

Holy crap man. Now that's a pretty useless post of yours, saying something is inaccurate without giving any arguments... hey, you're entitled to your own opinion, just like anyone, but either share your arguments in a useful contribution, or don't even bother posting, cause you know, it's a forum, a place to share opinions and arguments...

Holy crap man I wrote the word "accurate". Learn to take a compliment.

Will this is awkward.......

(off topic, what tube amps have you made?) 

I would like to know as well; I want to make a tube amp and tube DAC soon using some old Russian tubes a friend of mine found at a yard sale.