Jump to content
IGNORED

iZotope vs SoX Upsampling in Audirvana Plus 3.2


buonassi

Recommended Posts

I wanted to start a topic that focuses on the parameters of these two upsamplers in A+ and perhaps even get some objective measurements of Fourier transforms (impulse response graphs) from those people who can capture them.  My experience comes from these software upsamplers as they are implemented in Audirvana Plus 3.2 - But I don't think there's anything keeping others with different players from chiming in here.

 

To kick things off, I'm using @copy_of_a 's extreme settings for iZotope as shown below.  In order to avoid aliasing with his settings, you have to use a software plugin to further lowpass the signal.  In my case, I'm using FabFilter ProQ2

 

steepness: 3

length: 500,000

Cutoff: 1.25

Anti-Aliasing: 50

Prering: 0.36

 

For SoX, I've chosen the below based solely by ear.  No LP filter is needed with an EQ plugin since the filter is much steeper it seems.  Admittedly I haven't put near a much time into SoX as I have with iZotope:

 

Bandwidth: 90 

samples: 500,000

Anti-Aliasing: 85

Prering: 41

 

Overall, the biggest difference between these is the inability to finely tune the filter steepness in SoX.  I've read some anecdotal things and tend to agree with most of them.  For instance, Sox seems to present a great image/stage/separation, but comes at the expense of some more ringing that can be heard in the "shimmering" highs on cymbals.  What's interesting though, is that the preringing doesn't seem too bad when I scrutinize the timing of a kick drum - hearing first its batter head which should be followed by its resonate sub bass thump.  I find that my setting in iZotope still perform best here - but the SoX isn't nearly as bad in the lows as with the high frequency "smearing/shimmering" I hear.  But this could just be my settings, more time will tell which is the better upsampler for me.

 

What else have other's experienced? 

Link to comment
11 hours ago, copy_of_a said:

my experience with SoX in A+ 3.2.x is it's pretty much unusable:

 

I noticed similar aliasing and should've taken a screen shot of it.  It's weird as the aliasing pulses up to -45db but quickly falls.  This even happens with a bandwidth setting of 90.  But using fab filter, I can presumably remove some of that aliasing with a LP filter at 96db/octave.  Doing this, I still prefer your "extreme" izotope settings to any I could produce in Sox.  

 

Sox is initially appealing because it sounds "euphonic".  Perhaps this pulsing aliasing is causing intermodulation distortion that sounds good to some people?  Still I'm not sure I can hear the aliasing.  I'm more concerned that it could be damaging my electronics.  Is there any validity in this concern?  Also, what does aliasing sound like to you?  What cues should I be listening for? 

 

Link to comment
53 minutes ago, matthias said:

 

Maybe SoX adjusts the steepness according to the bandwidth? So you have a much higher steepness with a bandwith of 99% than with one of 90%.

Can you chose 500,000 samples with SoX?

 

Matt

perhaps it does just that, good conjecture.  I'm able to select 500K on mine, yes.  

500k.jpeg

Link to comment

well, I think this topic is on hold until SoX is better understood, or at least the aliasing is better controlled.  

 

In terms of development and maintenance, does anyone have an opinion on which of these upsampling software products is leading the pack here?  SoX is open source, but that doesn't mean it has any less potential to go up against the "pro audio" products.

Link to comment
  • 2 weeks later...
On 1/14/2018 at 2:48 PM, matthias said:

 

Maybe SoX adjusts the steepness according to the bandwidth? So you have a much higher steepness with a bandwith of 99% than with one of 90%.

Can you chose 500,000 samples with SoX?

 

Matt

@matthias, you were right.  I was able to observe the rolloff steepness of the SoX filters at various settings using white noise and zooming in on FabFilter's spectrum analysis between 10 and 22khz.  At the highest setting of 99.5 it is a "brick wall", whereas at the lowest setting of 74, there is a definite rolloff.  What's interesting is that even at 74, the rolloff isn't very severe.  It kicks in at 18K, much higher than I'd expect since 74% of nyquist at 44.1 is 16K approx.  This is also what I'm hearing, zero loss of high frequency air, and I have verified up to 18K hearing by an audiologist (kinda freaky for my 38 year old hearing system).  

 

Also, the ringing doesn't seem to be reduced by very much even at the 74 setting.  Sometimes I think I like SoX, other times I'm convinced that I'm getting the better staging and separation of instruments at the expense of a sloppier kick drum attack.  I guess we can't have our cake and eat it too, huh?

sox74.jpeg

sox99-5.jpeg

Link to comment
  • 1 month later...

I've given SoX some more time and I like its staging (width) and liveliness, for lack of a better term.  There's more shimmer in the highs and it's not as dry as iZotope.  I agree with @Noodle that SoX can sound euphonic, reminds me of the "exciters" on DAWs that add some sparkle by increasing the overtones/harmonics of cymbals.  It seems to spread a bigger canvas over my head (I'm a headphone listener).  

 

I've actually been enjoying it a bit more lately.  While I acknowledge that it may not be as "pure", I do seem to become more immersed in the performance.  

 

I've been using these settings - which is odd for me because I tend to like min phase better.  But something really opens up with the lin phase:

 

 

sox.jpeg

Link to comment
2 hours ago, RunHomeSlow said:

I think your safe volume reduction should be set at -3 when upsamplinh to DSD... don’t ask me why ?

yes, according to the Audirvana manual, this is to prevent the DAC from clipping.  But my DAC has analogue filters that activate when DSD is sent to it.  It has 3 positions:  0,-3,-6 db approx.  I have it set to -3 currently.  

 

Also, in my EQ software, I leave a little headroom to ensure no clipping, so that's an added layer of protection. 

 

According to A+ manual, unless you get channel dropouts or hear distortion, you should be ok to leave at 0.

Link to comment
  • 1 month later...
12 hours ago, Hoshi said:

I've come to the conclusion that SOX is a better fit for DSD.  Do you agree with this assessment?

Well I’m not sure. Sox is a resampler that performs interpolation. If it’s better for dsd I can’t say. It does sound a bit more shimmery and glaring composed to izotope. 

 

12 hours ago, Hoshi said:

noticed the I can no longer adjust filter setting by directly inputting settings into the box to the right of the sliders in A+ but must us the slider to make changes.  Have you noticed this as well?

Yes I have. I reported it, but it persists after some incremental updates. It is a pain, but not a dealbreaker for me. 

 

11 hours ago, Hoshi said:

was previously using:

steepness: 22

length: 500,000

Cutoff: 0.95

Anti-Aliasing: 100

Prering: 0.86

 

This isn’t a bad setting if you’re concerned with aliasing and want to suppress it. I have actually gone back to using this setting (or very similar) and do like it. 

 

Welcome  to the club. Glad you’ve joined us here on CA!

Link to comment
  • 8 months later...

any new comments on Sox vs iZotope sonic characteristics?  I've recently gotten back into upsampling and have decided to give linear phase / fast rolloff a good month of listening before switching back to NOS/Min phase options.

 

I like both Sox and iZotope for different reasons, and not really sure one algorithm is better than another - they both perform interpolation well. 

 

I wonder if @mansr would be willing to do some measurements for similar settings of these two converters?  Bandpass and transition zoom FR I'm not that interested in.  And since we'd keep it linear phase, no sense in measuring that.  But, which yields the lowest distortion:  Harmonic, odd order, IMD.    

Link to comment

Sorry, what I meant was I wasn't interested in seeing the FR as an indicator of bandpass response (IE any ripple) or how sharply it rolls off (transition).  That's what I meant.  Should've been more explicit.  

 

I am of course interested in how the low level distortion measures in the frequency domain.  Or are you saying there's none of that that can be attributed to the FIR filter?

Link to comment
  • 3 weeks later...

the 'hack' doesn't jailbreak the OS - and it is a reversible hack.  Actually, it self corrects with each iterative OS update you do.  The only threat is if you use a audio settings kext file that is corrupt.  You can copy your current el capitan kext file and use it (knowing it's not laden with malware) when you perform the hack.  

 

It took me about 25 minutes to do and wasn't really difficult.  

 

And, yes, I believe the direct mode is worth it if you plan on transporting via USB to a bridge converter or a DAC.  Of course, just my opinion. I haven't updated my mac in over a year because I enjoy the sound so much after achieving direct mode. 

 

In terms of consequences of not updating you OS?  I'm not sure if el capitan is still supported - probably not.  There's probably not a lot of active development for patching exploits caused by holes in other software that resides on the machine. 

Link to comment
15 hours ago, Endoxx Omega said:

Apple isn’t forcing me to update El Capitan yet. I assume that if there’s a major leakage they would?

I have to believe they'd warn you about it.  I think you can check this article at minimum - it's written by folks who know a lot more about it than I do

 

https://www.howtogeek.com/350901/which-releases-of-macos-are-supported-with-security-updates/

 

hope this helps and welcome to the forum.  we need more 'signal to noise' ratio here from good posters.  be one of them PLEASE!  

Link to comment

Search for content posted by @RunHomeSlow  .  He has by far the most experience and awesome detail outlining how to do direct mode hack

 

Actually I have invoked his name many times now and hate to keep bothering him. If he can supply the link to the instructions I’ll make sure to bookmark it this time I promise

Link to comment
  • 2 weeks later...

I've been playing around for settings with my 2qute.  Honestly, I like the WTA filters in the Chord DACs as they are, without oversampling done on the computer.

 

Not sure that you can add more highs, though you can certainly reduce them with a filter that rolls off sooner.  But, if you tell me the type of transient sounds you're looking for, I can help.  IE,  do you want more attack on the leading edge of the notes?  Do you want better instrument separation, more ethereal 3d sound?  You may want to read this to get a better understanding - then feel free to ask away!  Welcome to AS my friend!

 

 

Link to comment
  • 2 weeks later...

check this out.  Here, @copy_of_a outlines his most recent "set and forget".  I wasn't sure if you saw this recent post of his.  Could be helpful for you.  Personally, I've bookmarked it:

 

And just because...  if you want to try my most recent settings here they are, with some help from 'the upsampling master' himself.  Here's where he helped me out, and attached are my tweaked configs.  Note that I only have 2x only selected.  I find it doesn't sound so 'oversampled' and keeps the transients a bit thicker, more organic.  I only use this with my metrum NOS Dac, for my Chord, I'm rather enjoying its WTA filters for now.  The max filter length is arbitrary, no need to go beyond 500K - at least I can't discern any difference.

 

 

Screen Shot 2019-01-22 at 7.37.07 PM.png

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...