Hi folks,
I think similar questions have been already asked, but I'm still confused about the down-sampling process.
In essence, I would like to achieve 2 goals:
1. Reduce the size of my music by converting hi-rez tracks into 16/44 flac or even Apple AAC
2. Perform my own personal ABX tests between 24/96 and 16/44 FLAC's
I will probably start separate thread in the listening tests section regarding ABX tests, so for the moment I just would like to know what would be the highest quality re-sampler to use?
From what I've read here, people seem to generally prefer SoX resampler.
Which one is the best option?
Foobar plugin?
Audacity SoX (according to this link (http://wiki.audacityteam.org/wiki/Libsoxr) they use SoX, but it might be different to standalone SoX)
Standalone SoX via command line?
Any other options?
I would greatly appreciate any links / suggestions on how to use command line SoX as well.
Thanks in advance.
I wouldn't loose any sleep over downsampling... Some resampers "measure" better than others, but I've never noticed any artifacts/aliasing when re-sampling no matter what software I happen to be using (always 16/44.1 or better). And, since I always had a good reason for re-sampling I never bothered with ABX tests.
Of course AAC is lossy so you might want to do some ABX tests to see if you can hear a difference with any of your files. I'd recommend keeping a lossless archive just so you'll "feel better" or in case you change your mind at some later date. It's up to you if you want to keep the high resolution originals and if you've got the disc space, why not?
If you use Audacity, there are some quality-setting options so you might as well use the highest quality setting.
https://hydrogenaud.io/index.php/topic,110058.0.html
Thanks, I will have a look. I will difinetely keep 16/44 files and most probably delete hirez ones. I just don't see any practicality of keeping these enormously sized tracks if I will fail ABX :)
Question regarding ABX tests of 24/96 vs 16/44:
I have seen Archimago's tests and read few threads here on this subject. I understand the logic of downsampling the hirez original and then upsampling it again to the same level - is to avoid any audible hardware induced changes during the playback of files with different sample rates.
However, aren't we introducing other variables (possibly audible) by upsampling it again from 16/44 to 24/96? I guess all these previously cut off frequencies won't magically reappear, but what happens to the audible part of the range?
I mean at glance this process looks similar to 2 step FLAC → AAC →FLAC conversion when the second step AAC → FLAC will give as the exact same AAC but simply different container.
In other word, aren't we inducing audible changes in 16/44 to 24/96 upsampling?
Modern DACs are often delta-sigma and they resample (or oversample) no matter what signal you feed to them. The exception is when the DAC allows you to disable that function. If you have read Archimago's articles enough you probably knew that already, for example his TEAC UD-501 has this function.
SoX's resampler is a free and popular one. It can provide transparent resampling out of the box and when in doubt, people can also specify the parameters being used to perform the resampling, that's why it is recommended to upsample the downconverted files again with SoX instead of using potentially non-transparent (or sample rate dependent) algorithms from some incompetent or dishonest DAC manufacturers.
My DAC is Audioengine D3. I don't think I can disable resampling, and honestly I'm not sure if it's delta sigma. Anyway, when I'm ready I will try to make 2 tests - with and without upsampling.
Archimago reviewed AudioEngine D3 already:
http://archimago.blogspot.hk/2014/03/measurements-audioengine-d3-usb-dac.html
...and be careful of the clipping problem he mentioned with the D3:
http://archimago.blogspot.hk/2014/03/measurements-another-look-audioengine.html
...as well as the intersample clipping I mentioned in this thread:
https://hydrogenaud.io/index.php/topic,110058.0.html
Such issues can affect ABX test accuracy.
The conversion to 16 bits (if you decide to keep lossless; it's not needed for any reasonable lossy encoder) is also a dilemma: simple rounding can introduce audible harmonic distortion where the signal is very quiet and almost noise-free (often happens on fade-outs, for example), but there are a lot of dithering methods to choose, with different trade-offs. (and dithering will also make losslessly compressed files a little bit larger, all else being equal)
The difference will be hard to hear indeed, but possible if you really turn it up, or even more possible if you later use a compressor/dynamic gain effect while listening for some reason (that is, extremely turn up quiet parts).
Also if gapless playback is required some care should be taken or you can get audible clicks (https://hydrogenaud.io/index.php/topic,114960.msg947760.html#msg947760).
Also if gapless playback is required some care should be taken or you can get audible clicks (https://hydrogenaud.io/index.php/topic,114960.msg947760.html#msg947760).
Resampling to 16/44 often results in sector boundary errors which can create gaps/clicks. Traders Little Helper can be used to fix SBEs:
http://tlh.easytree.org/