Audirvana + How to select Izotope settings?

[copy_of_a]

some - personal! - notes regarding "rules of thumb"...

 

This is from the manual of HQPlayer and I agree:

linear phase = better space

minimum phase (low pre-ringing respectively) = better transients ("attack")

 

"steepness" in iZotope SRC:

- higher steepness = better filtering above the nyqiust limit

- higher steepness = more ringing

- lower steepness = better instrumental seperation (especially in complex, 'full' musical arrangements)

- lower steepness = smooth roll off of high frequencies (unless you increase "cutoff scaling")

 

From these few notes alone you can see that we are always dealing with trade-offs/ compromises.

 

If you prefer low steepness (at the expense of filtering quality) you can also filter out aliasing/artefacts above the nyquist limit with an appropriate EQ in the AU Plugin section of A+.

If so, you would consequently need different EQ/Filter-settings for different source sample rates... so you would have to manually change the plugin settings whenever the sample rate of the source changes (which is pretty cumbersome).

 

If you want to effectively remove aliasing/artefacts above the nyquist limit while upsampling on the fly and without using a plugin I would say "steepness" should at least be set to '31' or higher (and "Anti-Aliasing" to '200').

 

-----

 

Meanwhile, me personally I prefer offline upsampling (of 44.1kHz an 48kHz source files) in iZotope RX4. Actually I was really happy with steepness: 16, cutoff: 1, pre-ringing: 0.72... and afterwards removed all the aliasing/artefacts above the nyquist limit with Fabfilter Pro Q2. But before I started to convert my entire library I've further fine-tuned my SRC settings and am now at steepness: 3, cutoff: 1.3 (to preserve all high frequencies), pre-ringing: 0.36. I then remove all the aliasing/artefacts above the nyquist limit with said Fabfilter plugin in linear phase mode at the highest quality setting and save the files as 24bit TPDF dithered files.

To further upsample these files to DSD in A+ I set the iZotope settings to almost the same values: steepness: 3, cutoff: 1, pre-ringing: 0.36, Anti-Aliasing: 50 (so the lowest value). Since the files are now at 176.4kHz (192kHz respectively) the nyquist limit is so high that there is no need to shift the cutoff scaling or to apply any Anti-Aliasing filtering...

[copy_of_a]

- Safe values

 

a follow up of my previous post re "safe" settings...

 

 

 

 

"Safe" settings if you want to remove aliasing/artefacts above the nyquist limit:

 

 

linear:

• this is one of iZotopes presets

- steepness: 31

- cutoff: 1

- Anti-Aliasing: 200

- pre-ringing: 1

 

 

moderate pre-ringing:

• good compromise between decent filter quality & moderate ringing & moderate pre-ringing

- steepness: 31

- cutoff: 1

- Anti-Aliasing: 200

- pre-ringing: 0.72 (0.72 - 0.78)

 

 

 

 

"proven" settings if you don't care about aliasing/artefacts above the nyquist limit (for whatever reason):

 

 

linear:

• great instrumental seperation

• soft rolloff of high frequencies (starting at 16kHz or so... IIRC)

• very weak filter quality

- steepness: 5

- cutoff: 1

- Anti-Aliasing: 200*

- pre-ringing: 1

 

 

moderate pre-ringing:

• great instrumental seperation

• pretty good transients

• very weak filter quality

- steepness: 7

- cutoff: 1.02

- Anti-Aliasing: 200*

- pre-ringing: 0.86

 

 

low pre-ringing:

• very good instrumental seperation

• very good transients

• mediocre filter quality

• for me overall a really great compromise between filter quality & low ringing & low pre-ringing

- steepness: 16

- cutoff: 1

- Anti-Aliasing: 100*

- pre-ringing: 0.72

 

 

my settings from above just to be listed again here...

 

 

low pre-ringing "extreme" setting:

• phantastic instrumental seperation

• great transients

• super weak (actualy 'no') filter quality

steepness: 3

cutoff: 1.3

Anti-Aliasing: 50

pre-ringing: 0.36

- cutoff 1.3 requires to reduce the gain by 1db to avoid too high intersample peaks. So if you set the gain by default for instance to -3db you should set it to -4db with this setting.

- Too, as outlined above, IMHO settings with such a low filter steepness are better used in conjunction with an appropriate high quality lowpass (highcut) filter in the AU plugin section of A+ if you want to upsample on the fly (and if you do so better set Anti-Aliasing to 50). But this goes for all the settings listed in this second section...

 

 

--------

 

 

personally I don't think that the filterlength really affects the sound (if so it's mostly likely related to computer-power...). I always set the max filter length 200.000. Furthermore me personally I prefer upsampling by power of 2 (with all the settings noted above).

 

 

--------

 

 

* Anti-Aliasing

The Anti-Aliasing filter in the iZotope SRC section of A+ does NOT help to remove aliasing/artefacts ABOVE the nyquist limit. It only suppresses freuquencies right AT the nyquist limit (see here for steepness: 5 / cutoff: 1 / AA: 200 / pre-ringing: 1 : http://www.computeraudiophile.com/attachments/f11-software/29056d1474067957-izotope-sample-rate-convertor-jud_5_1_200_1.jpg ). This is why I think it's pretty useless at low steepness settings and I would rather set the AA-filter to 100 or 50.

[copy_of_a]

Depends what you're upsampling to. I'm using DSD256 most of the time, so it doesn't take a very steep filter to get rid of most aliasing at that rate.

the filter acts at the nyquist limit of the _source_ file... so low steepness filters produce artefacts ("fake details" if you want so...) right above the nyquist limit of the original file. My screenshot above shows this pretty clear (source is 44.1kHz / target is 176.4kHz).

[copy_of_a]

Thanks to you and copy_of_a.

I did have a look at my filter settings in RX3 way back when, and from what little I knew (and know) they seemed to produce a reasonable graph. But obviously I could easily be wrong about that.

Here are 2 screenshots that make things a bit clearer ... I hope...

I've upsampled a track with iZotope RX4 to 96kHz with your settings (5/1/1).

The screenshots show the spectrometer while playing the file and the spectrum (in black & white; sensitivity = content from -120db and above) of the upsampled file.

The first screenshot shows a lot of "fake-details" above 22.050Hz (up to at least -50db in this section of this particular track - the horizontal white line in the spectrum is 22050kHz).

The second screenshot shows the same file filtered through "Fabfilter Pro Q2" to remove anything above the nyquist limit (the nyquist limit of the original file's resolution that is).

 

These screenshots also show the underlying intend of my previous posts: the best of 2 worlds - very high instrumental seperation facilitated through super low filter steepness in the resampler module to gain very, very low ringing (and literally no pre-ringing) but at the same time high quality filtering of aliasing/artefacts/fake-details above the nyquist limit (in my case with the Fabilter EQ plugin in an offline upsampling workflow...).

 

[copy_of_a]

I've upsampled a track ...

a 44.1kHz track ... (sorry!)

[copy_of_a]

Jud, you misunderstood Miska. The amount of ringing increases with filter steepness, but *neither* is affected by the filter length.

The filter length determines how long the "tail" of the impulse response is allowed to continue as the ringing magnitude progressively declines. There is no benefit in extending the filter length beyond the time at which the magnitude of the impulse response (the ringing) falls below the resolution of the DAC chip.

In other words, a filter length shorter than this amount truncates the impulse response so that the ringing stops abruptly instead of declining until it is buried in the inherent noise of the DAC. In the frequency domain, such truncation causes a tiny irregularity in the frequency response curve.

The steeper the filter slope, the longer the impulse response continues before it falls below the resolution of the DAC. So a longer filter does NOT cause ringing, but an impulse response with more ringing will be reproduced more accurately if its filter length is correspondingly longer.

Since the "filter length" refers to "samples" me I also thought the filter length defines how many samples are loaded into the SRC module (RAM resp.) to be upsampled at once as a "package" (simplified speaking in my bad English...). What you are saying kind of confirms this.

Thanks for your post!

[copy_of_a]

The filter length determines how long the "tail" of the impulse response is allowed to continue as the ringing magnitude progressively declines. There is no benefit in extending the filter length beyond the time at which the magnitude of the impulse response (the ringing) falls below the resolution of the DAC chip.

(...) The steeper the filter slope, the longer the impulse response continues before it falls below the resolution of the DAC. So a longer filter does NOT cause ringing, but an impulse response with more ringing will be reproduced more accurately if its filter length is correspondingly longer.

Bob,

I've checked the number of samples with my SRC settings (steepness: 3 / pre-ringing: 0.36).

With regard to the screenshot below my setting requires a filter length of ~100 samples when upsamling 44.1kHz to 176.4kHz (16bit, Mono in the pulse file shown...).

Just to be sure...: with regard to your post I assume I can easily set the filter length to the lowest value (10.000 samples in A+) when I am only upsampling (no downsampling from DSD to PCM or so).

Is that correct?

Many thanks in advance!

 

[copy_of_a]

Well, if that is true, a lot of people could set it to 10.000 and many pages in the 35page Isotope settings thread would look pointless regarding filter length - or any of "copy of a" settings above would work well so to speak.

My question has been only a technichal question to make sure that I did understand Bob correctly.

It was absolutely not a question whether or not a low filter length is "better" or sounds "better".

[copy_of_a]

Yes, that is a correct interpretation of my post.

cool - many thanks!

However, bear in mind that 16-bit resolution equals a dynamic range of 96 dB, and your graph is not sufficiently detailed to show when the ringing amplitude declines to –96 dB.

correct! When I zoom in vertically to see the full -96db the flat lines left and right of the visible waveform show more ringing (all in all around 100 samples... not only the ~50 samples the waveform shows at this magnification). However, I've only prepared the graph for a better understanding. 100 samples would be a bit low :-) ... and can't be set in A+ anyway.

Thanks again!

 

edit: for reference here's the same graph zoomed in vertically to show the scale down to -120db.

[copy_of_a]

Hello RunHomeSlow.

 

Sorry for the late reply… in fact I don’t visit the forum frequently.

 

Basically you are right: the SRC settings (intentionally) produce a large amount of artifacts („images“) due to the very low steepness of the filter. Afterwards you have to filter out the images with a lowpass filter in the AU plugin section.

Then again the steepness of Apple’s built in lowpass filter is really not sufficient.
Too, in your screenshot it’s set to around 7kHz or 8kHz… which is extremely low.

 

Fabfilter Pro Q2 for instance features a -96db/octave lowpass filter and you can stack it.
Attached a screenshot of a 44.1kHz file upsampled with the „extreme“ settings (3 - 1.25 - 50 - 0.36) with and without lowpass filtering (the two filters are set at 18kHz and 19.5kHz… both -96db/octave). There are still some images … but they do not exceed -100db… so really nothing to worry about.

 

Personally I’ve skipped the adventure of upsampling my entire library offline … because it takes ages. I still think it would be the best approach, but it’s simply too much work.
Instead I’ve tuned my „online“ settings in A+ to work with all source sample rates … if you want so, my „set it and forget it“ settings that reliably filter out images with a lowpass filter starting to set in at 18kHz (referring to 44.1kHz source files) and with pretty low ringing and moderate pre-ringing. It’s a compromise (as always) but I am happy with it…
The settings are:
Steepness: 22
Filter: 500,000
Cutoff: 0.95
Anti-Aliasing: 100
Pre-Ringing: 0.86

 

... hope that helps...

 

[copy_of_a]

11 hours ago, Piano Man said:

if I understand correctly, your latest "set it and forget" allround settings are run without further lowpass filter, right?

correct

 

Quote

Lowering the cutoff frequency by 1 kHz  or so relaxes filter steepness requirements sufficiently?

Yes. Not sure it's just 1kHz, though (can't check right now)..

 

Quote

I don't completely understand yet the other approach with a very low steepness and subsequent low-pass filtering to remove aliases. Why doesn't this low-pass filter introduce the same problems as a steep Izotope filter setting, i. e. ringing, loss of stage depth and instrument separation?

Very good question! I'll come back to you later this week when I am back home at my computer...

[copy_of_a]

On 6.6.2017 at 9:05 PM, Piano Man said:

Looking forward to your answer regarding the additional low-pass filter...

Here's a single pluse usampled in iZotope with above mentioned "extreme settings" and high cut filtered with FabFilter Pro Q2 in so called "natural phase" mode.

 

the original pulse (44.1kHz):

 

uspsampled to 176.4kHz:

 

Lowpass filtered with Fabfilter (2 filters at 18kHz and 19050kHz, both high cut -96db/octave):

 

The ringing is very low and there are absolutely no images (artifacts above the nyquist limit of the source file).

 

A settings with as low ringing as possible and a roll off of high frequencies also starting at around 18kHz is

steepness: 22 and cutoff 0,95 (and pre-ringing 0,36). Actually it still produces quite some images but they get filtered out reasonably good with Audirvana's Anti-Aliasing filter (set to 100).

The comparsion is a bit unfair, since this second setting actually would need a much steeper filter to act as effective aigainst images as FFQ2 above. However, even when I try to keep the ringing as low as possible and the lowpass filter even gentler than above (steepness only at 22) it still shows more ringing than above:

 

That comparison falls a bit short, of course. But I hope it shows the interrelationship.

 

The whole adventure also depends on 2 things: low pre-ringing settings generally produce a larger amount of images (heavy spikes above the nyquist limit of the source file that are harder to filter out). So with moderate pre-ringing settings or linear upsampling the lowpass filter does not have to be super strong.

Secondly it depends on the kind of filter used in the additional lowpass filter. Linear phase mode EQs can produce a pretty high amount of (pre- and post-) ringing (depends). But Fabfilter Pro Q2 features a filter mode that keeps the ringing generally very low. Here's a quote from Fabilter's manual:

Quote

Natural Phase
Pro-Q 2's unique Natural Phase mode (...) not only perfectly matches magnitude response of analog EQ'ing, but also closely matches the analog phase response. So it delivers the most accurate frequency response and best sound quality, even at the lowest frequencies and highest Q settings, without introducing noticable pre-ring or long latency!

 

Hope that helps ...

[copy_of_a]

19 minutes ago, Piano Man said:

Apparently the Fabfilter is not suitable as plug-in for on-the-fly processing with Audirvana, otherwise you would use it (?).

FFpQ2 works great in Audirvana Plus. But every time the sample rate of a file you are playing is changing you would have to set an approriate lowpass filter setting for FFpQ2, which is really cumbersome. Alternatively you could use the filter for 44.1kHz source files (so also when the source is 48kHz, 96kHz or whatever)... but that doesn't really make sense to me.

 

Quote

So I'll try to find the optimum with the iZotope settings.

At the beginning of this thread you, copy_of_a, included a link to another thread on that topic which does not work any more. Has this thread been moved?

 

 

->

 

 

[copy_of_a]

12 hours ago, Chipless said:

-Just to clarify, are these the "correct" "set it and forget it" settings per the above recommendation without any AudioUnits enabled (see attached)?

 

 

-I read in an earlier post that forced Power of 2 oversampling was being used: Is that the case with the aforementioned "set it and forget it" settings"?

- the settings are correct

- power of 2 or upsampling to max. rate... personally I prefer power of 2

 

Quote

If my DAC supports DSD, should I force oversample to DSD? If so, do the aforementioned settings stay the same, and which DSD filter should I use?

the iZotope SRC settings stay the same.

Whether or not it makes sense to upsample to DSD depends on your DAC, your CPU power and your personal taste.

 

Quote

-In general, is it better to force oversampling to the DAC? This article on xiph.org makes it sounds like it's good idea within the paragraph titled "Oversampling", but also makes it sound like this happens within DACs automatically. 

DACs do oversample. Upsampling in the playback software transfers the upsampling from the DAC to the software, where you can adjust the settings to your liking. Too, the quality of the software resampler might be superior over the DACs bulit-in resampler.

[copy_of_a]

Since this thread is referred to and quoted from time to time I think it's about time to update some things from above.
To keep it simple: forget the „extreme“ settings and the use of an EQ in the AU Plugin section to filter out images when upsampling.

Here are some settings that should cover more or less all needs and may serve as a reference (also for further tweaking if desired).

I’ve mainly selected settings that roughly replicate the transition / filter shape of other SRC softwares.

 

In conjunction with Audirvana’s „Anti Aliasing“ setting in the iZotope module I’d consider all the settings below as „safe“ regarding aliasing/imaging with real world music (which, of course, is already low pass filtered).

These values apply to all the settings:
filter length: 512.000
Anti Aliasing: 100
pre-ringing: „1“ for linear phase or „0“ for minimum phase

Pick your poison …

 

_____________________________

Short Filter / Slow Roll off

 

steepness: 16 / cutoff: 0.93
filter starts around 18.0kHz / attentuation @ Nyquist: ~ -27db
(transition similar to HQP poly sinc short)

steepness: 22 | cutoff: 0.95
filter starts around 19.0kHz / attentuation @ Nyquist: ~ -26db
(my favorite for a slow roll off filter in A+/iZotope)

_____________________________

Moderate (average) Filter

 

steepness: 28 | cutoff: 0.956
filter starts around 19.5kHz / attentuation @ Nyquist: ~ -30db
(transition similar to HQP poly sinc)

steepness: 31 | cutoff: 0.96
filter starts around 19.7kHz / attentuation @ Nyquist: ~ 30db
(iZotope preset „CD“)

steepness: 35 | cutoff: 0.95
filter starts around 19.6kHz | attentuation @ Nyquist: ~ -50db
(transition similar to Sonic HD preset „gentle“ = Amarra 4)

steepness: 36 | cutoff: 0.965
filter starts around 20.0kHz / attentuation @ Nyquist: ~ -30db
(my favorite for a moderate filter)

_____________________________

Steep Filter

 

steepness: 68 | cutoff: 0.96
filter starts around 20.5kHz / attentuation @ Nyquist: ~ -120db
(transition similar to SoX preset "VHQ")

steepness: 80 | cutoff: 0.96
filter starts around 20.5kHz / attentuation @ Nyquist: -180db
(iZotope preset „Steep, No Aliasing“ / transition similar to HQP poly sinc ext2)

_____________________________

Ultra Steep Filter

 

steepness: 200 | cutoff: 0.985
filter starts around 21.5kHz / attentuation @ Nyquist: ~ -180db
(transition similar to SSRC = JRiver)

 

 

 

[copy_of_a]

19 hours ago, Chipless said:

Quick question: Do you think the iZotope module is superior to the SoX one?

In generell I don't know.

Within Audirvana iZotope provides more control than SoX - most notably the steepness setting. Therefore I prefer iZotope.

[copy_of_a]

19 hours ago, johnnystar said:

I followed the recipe for your ‘extreme’ settings using the FFQ2 cutoff filters and was quite happy with the sound (with all source sample rates, not only 44.1). Your description of very low ringing and excellent separation of instruments made sense to me. However, your latest settings above leave me confused. I havent tried any of these settings yet because your accompanying descriptions go right over my head (I.e., comparing the sound signatures to other SRC software which I am not familiar with). Is it possible for you to humor folks like me and elaborate with descriptions more suited to the layman? It would be much appreciated.

I would like to refrain from descriptive speech regarding the „sound“ of SRC.
I also wasn’t referring to something like „sound signature“ of other softwares - really only to the roll off / filter transition.

In general me personally I would say: the shorter (less steep) the filter the better the instrumental separation the more „open“ the sound.
But we are talking tiny differences here! Not about differences anyone could easily tell in a blind test (unless you compare 2 extremes - so a really short filter that starts to roll off as low as 15kHz to a really steep filter that preserves everything up to 21kHz or so…).
But: with a short filter you may produce heavy aliasing/imaging artefacts. Which should be avoided IMO (at least to a reasonable degree). So it’s always a tradeoff between preserving as much as possible from the audible spectrum of the source versus avoiding aliasing.
Simplified speaking the trick with really steep filters is to shift the ringing (which is actually only audible in the transition band of the filter) to frequencies too high to be audible (above 21kHz or so). Then again I feel the filter-energy pushed back into the audible band due to the steepness of the filter still makes a difference (at least with iZotope and SoX). This is why me personally I prefer moderate filters… as yet.

 

 

[copy_of_a]

10 hours ago, johnnystar said:

extreme settings (...)

Do you still recommend those for folks that have the Fab Filter AU plugin?

no