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...).
