Izotope SRC

[ed71]

Yes, I had been playing with the settings on my own without knowing what I was doing and I had forgotten what the default settings were. I think I ended up with the steepness way off of where it probably should be. This way I'll know how to get back to the default settings if I do any more experimenting. The Jud 2 & Superdad settings were a big improvement over what I had been trying on my own, but I seem to like the defaults even better.

[Dyson]

As well I have been playing with the settings since I changed my DAC, but never really found something that convinced me. A few days ago I rediscovered the default settings, and for it was the second best that pleased me the most, more than all my trials. So I decided to start this thread again (yes from #1) in order to refresh my knowledge. I tested many of the settings that were discussed along the posts and finally came back to post #70, that I changed a little bit:

 

Steepness - 4

Filter max length - 1,333,333

Cutoff frequency - 1.28 x Nyquist

Anti-aliasing - 200

Pre-ringing - 0.07

Forced Oversampling - x4 on 44.1 only

I apply these only to 44.1 (and some 48) tracks that I upsample to 176.4.

My DAC runs in Bit Perfect mode (=NOS).

 

Is there something wrong using such settings (it was not much discussed after it was published) ?

Am I really hearing only the original content, or is it all the artifacts from the conversion that I'm listening to and that I find so "nice"?

[Dyson]

OK, something was definitively wrong with the previous settings: confusing soundstage and too much highs. I started to decrease the cut-off by 0.01 and ended up with:

 

Steepness - 4

Filter max length - 1,333,333

Cutoff frequency - 1.18 x Nyquist

Anti-aliasing - 200

Pre-ringing - 0.07

Forced Oversampling - x4 on 44.1 only

 

This sounds much more "normal" and natural with very good spatial instruments placement.

 

If increasing too much this cutoff setting has such a "bad" impact, why is it even possible to go up to 2.00?

(human is always tempted by "more is better"...)

[skatbelt]

Something in me says the above settings can not be optimal for speaker playback regardless of DAC type. But may be you did find these settings work well in your headphone setup.

 

For optimizing a (higher end) system in general, the use of a track with a single instrument that has a relative compact physic and can play or excite lower and higher octaves is very usefull. I personally use Christian McBride - Night Train (track) - Gettin’ To it (album) or Antonio Forcione - Night Passage (track) - Live (album). In these tracks, all energy should come from the same (virtual) projection of the instrument and its player in the soundstage. If you close your eyes you should 'see' the instrument. In not well setup or lower quality/bad systems projections become huge, not well defined and/or bass notes seem to come from an other source than higher notes.

[Dyson]

skatbelt, you're right I did those "researches" with the headphone. Your test tracks are very good ones to hear when something is wrong! Finally I'm back to the default settings in their Jud 3 version (pre-ringing 0.75 instead of 1.00), hey work well for both the headphone and speakers. Thanks for pointing me to those 2 tracks for testing.

[liteguy1]

Well, out of curiosity I tried your settings. It gives me a deep and well ordered soundstage but the presentation of instruments and voices are flat. My own settings (in my setup) are much more life like, I can hear more details (tape hiss, chair movement etc.) but - more important - instruments and voices feel like volumes in space. They do have a front and a back, are more 3D so to speak. As I mentioned above, steepness and pre-ringing are strong related so it took a while to find my optimum. My settings are:

 

Steepness: 83

Filter max length: 800,000

Cutoff freq.: 1,08

Anti-aliasing: 200.0

Pre-ringing: 0.30

Forced Upsampling: Power of 2

 

Hello Skatbelt, Thank you for what probably are endless hours experimenting with these parameters. These settings have transformed my system to pure liquid. Thank you!!!

[Dyson]

Some advice from the settings specialists (Jud, SuperDad, Skatbelt, others…) would be greatly appreciated. I’m experiencing no filtering setting and I’m very surprised to hear how much the “Filter max length” has an influence on the result! I have been using, amongst others, the 2 files Skabelt has mentioned earlier to help me better confirm the results.

 

I start with:

Steepness: 0

Cutoff freq.: 2.00

Anti-aliasing: 50.0

Pre-ringing: 1.00

Forced Upsampling: Power of 2 (meaning 705.6 from 44.1 for me)

 

Then I get for following Filter max length:

Above 500’000

Wrong stereo image

500’000

It’s the upper limit where the stereo image is somewhat acceptable, but instruments are a bit too wide open (missing consistency)

Under 450’000

Stereo image is closing down (becoming very narrow) and getting very smooth, missing clarity

450’000

Good soundstage, nice instruments consistency but a little weak on bass

455’000

Same as above with more bass, but bass missing definition

456’000

My current setting, bass is present and defined, natural clarity (not over bright), very good 3D perception with instrument consistency

 

I have tested from 450k to 460k in 0.5k steps to find out. Above 460k it’s already starting to be less realistic.

 

How can this setting have such an impact? Some said that it was more a memory optimisation setting to reduce as well the computation need.

 

Thanks for feedback, also if someone wants to lose time to test it :-)

[Superdad]

 

I start with:

Steepness: 0

Cutoff freq.: 2.00

Anti-aliasing: 50.0

Pre-ringing: 1.00

Forced Upsampling: Power of 2 (meaning 705.6 from 44.1 for me)

 

 

Hi Dyson: I hate to break it to you, but with a steepness of 0 you are not creating any filter at all! Therefore none of the other settings should make any sonic difference at all. Also, the anti-aliasing setting is an indication of the maximum depth of the filter's attenuation of frequencies above the cut-off. A setting of 50 is very little.

Oh wait, I also see that you have cut-off set to 2.0! So that means that even if you had a filter (which with a steepness of 0 you don't), it would not start until twice the Nyquist frequency--you have it set at 44.1kHZ.

 

So I have no idea what you are hearing (I personally think headphones are not good for tuning filters), but you have created a very strange filter that does not do anything. No offense intended. You might want to read a bit in the A+ manual to better understand the function of the various parameters.

[Dyson]

Hi Dyson: I hate to break it to you, but with a steepness of 0 you are not creating any filter at all! Therefore none of the other settings should make any sonic difference at all. Also, the anti-aliasing setting is an indication of the maximum depth of the filter's attenuation of frequencies above the cut-off. A setting of 50 is very little.

Oh wait, I also see that you have cut-off set to 2.0! So that means that even if you had a filter (which with a steepness of 0 you don't), it would not start until twice the Nyquist frequency--you have it set at 44.1kHZ.

 

So I have no idea what you are hearing (I personally think headphones are not good for tuning filters), but you have created a very strange filter that does not do anything. No offense intended. You might want to read a bit in the A+ manual to better understand the function of the various parameters.

 

Hi Superdad: I want to understand how each parameter influences the result (and each other), that's why I'm starting with a no filter setting. Because it is a no-filter I'm so surprised to have influence of a setting that even with a filter should only affect the quality and the CPU load, not the sound image and bass/treble response.

[Dyson]

With the Max filter length set as per above experimentation, I started to work on the filter without touching the Pre-ringing and Anti-aliasing. I ended up with following:

 

Steepness: 1

Max filter length: 456'000

Cutoff freq.: 1.01

Anti-aliasing: 50.0

Pre-ringing: 1.00

Forced Upsampling: Power of 2 (meaning 705.6 from 44.1 for me)

 

With those settings I lost a very little bit soundstage, but makes it even more natural. Clarity and body are still there. The upsampling side-effects that were sometimes audible at higher volumes are gone with the filtering. It works also very well with speakers, not only headphone.

[skatbelt]

Hi Dyson. I don't know what you exactly intend to do but I believe it is generally assumed that it is best to keep Anti-Aliasing at it's maximum value. This equals best quality and is advised unless your cpu cannot handle it.

[Jud]

With the Max filter length set as per above experimentation, I started to work on the filter without touching the Pre-ringing and Anti-aliasing. I ended up with following:

 

Steepness: 1

Max filter length: 456'000

Cutoff freq.: 1.01

Anti-aliasing: 50.0

Pre-ringing: 1.00

Forced Upsampling: Power of 2 (meaning 705.6 from 44.1 for me)

 

With those settings I lost a very little bit soundstage, but makes it even more natural. Clarity and body are still there. The upsampling side-effects that were sometimes audible at higher volumes are gone with the filtering. It works also very well with speakers, not only headphone.

 

Remind me of what DAC you have.

[Dyson]

Hi Dyson. I don't know what you exactly intend to do but I believe it is generally assumed that it is best to keep Anti-Aliasing at it's maximum value. This equals best quality and is advised unless your cpu cannot handle it.

 

That is exactly the problem: all those settings that claims to be purely "optimisation" ones, have in fact much more effects than expected on the sound quality!

 

I try to achieve a filter that keeps the result as close as possible to the original content. I don't want to listen to a filter, I want to listen to the original intention of a musician. When I test the settings I use a lot of different tracks, comparing with and without upsampling (when not up sampling i keep my DAC in NOS mode) redbook content and comparing up sampled PCM to its DSD version.

 

I rely more on my hears (and brain...) than on "mathematical" formulas :-)

[Dyson]

Remind me of what DAC you have.

 

iFi micro iDSD

[skatbelt]

That is exactly the problem: all those settings that claims to be purely "optimisation" ones, have in fact much more effects than expected on the sound quality!

 

I try to achieve a filter that keeps the result as close as possible to the original content. I don't want to listen to a filter, I want to listen to the original intention of a musician. When I test the settings I use a lot of different tracks, comparing with and without upsampling (when not up sampling i keep my DAC in NOS mode) redbook content and comparing up sampled PCM to its DSD version.

 

I rely more on my hears (and brain...) than on "mathematical" formulas :-)

 

That's the way you look at it. Assuming that no filter or up/oversampling gives a more pure result is one theory. My experience is exactly the opposite: the use of SRC with the optimal parameters for my system brought me closer to my references...

[Dyson]

That's the way you look at it. Assuming that no filter or up/oversampling gives a more pure result is one theory.

 

That's not what I said, or at least what I wanted to say...

[skatbelt]

That's not what I said, or at least what I wanted to say...

 

You're right. My misinterpretation.

Leaving Anti-Aliasing at max and out of your tests is probably a wise thing to do and makes life a bit easier...

[Dyson]

Leaving Anti-Aliasing at max and out of your tests is probably a wise thing to do and makes life a bit easier...

 

Some observations based on A+ implementation: the pre-sets do keep 3 values fixed, changing only the steepness and the anti-aliasing (AA). If you divide the steepness by the AA you get:

 

[TABLE=class: grid, width: 500]

[TR]

[TD]Steepness[/TD]

[TD]19[/TD]

[TD]53[/TD]

[TD]97[/TD]

[TD]150[/TD]

[/TR]

[TR]

[TD]AA[/TD]

[TD]125[/TD]

[TD]150[/TD]

[TD]175[/TD]

[TD]200[/TD]

[/TR]

[TR]

[TD]ratio[/TD]

[TD]0.152[/TD]

[TD]0.353[/TD]

[TD]0.554[/TD]

[TD]0.75[/TD]

[/TR]

[/TABLE]

 

I found that if you keep the AA fixed and change the steepness to have 0.35 or 0.55 ratios, you do vary the stereo image (opening or closing left-right image).

If you keep the ratio fixed and increase (x2, x3) steepness and AA together, you do vary the depth (adding more air).

Using even value for steepness, especially in the lower value, does emphasis the bass (loosing definition).

So couples that work rather well for the ratio 0.35 are 42/120 or 63/180 (as well as the pre-set of course). For 0.55, that would be 66/120, 99/180 and pre-set.

 

I just wanted to share this, as the settings are much more interlinked as what I thought.

Not intended as new settings proposal (even if I’m currently listening with 42/120…)

[skatbelt]

Triggered by a PM of Dyson I thought I'd share my latest findings with iZotope SRC.

 

I am still quite close to the settings I posted earlier.

 

Steepness: 73 dB

Max filter length: 645.600

Cutoff freq.: 1.00

Anti-aliasing: 200.0

Pre-ringing: 0.33

Power of 2

 

The biggest change for me is going back to a cutoff frequency of 1.00. I found the cutoff setting to be more or less comparable to the focus of a lens in photography. So back to front. A value higher than 1.00 leads to back focus, a lower one to front focus. It can be perceived as airiness (in some genres) but it is artificial and leads to a wrong presentation of soundstage depth.

 

The combination between steepness and pre-ringing to my ears acts more as a focus of the stereo image in the left to right plane. Not set right can lead to a stretched or compressed image, depending on the offset. I optimised my settings by striving for realness and lifelike presentation of stringed instruments (violin, cello, etc. but also piano). An offset leads to softness, less realness but can be perceived as more fluent/liquid, analogue, tube like. Not my cup of tea though. I call it distortion, he he...

 

The relationship between steepness and pre-ringing for me is somewhat constant. .01 up in pre-ringing is 3 or 4 dB down in steepness.

 

Since a few weeks and for headphone purposes I own a Chord Hugo. I was surprised to find out that with the Hugo I prefer exactly the same settings as with my Berkeley Alpha USB/Lyngdorf combo.

[Dyson]

I share skatbelt’s description of the effect of the cutoff frequency, but for me a setting of 1.00 is closing too much the soundstage depth. When I compare with natural DSD recordings from NativeDSD none sound as closed/front focused as with the 1.00 cutoff freq. setting, true also with the same SACD and PCM track comparisons. Instruments need more air to come to life.

 

My current settings come from listening with different DACs, different headphones and speakers setup. I have also compared with other players that people seems to like for their sound & soundstage presentation.

 

Steepness: 128

Max filter length: 1'310'270

Cutoff freq.: 1.08

Anti-aliasing: 192.0

Pre-ringing: 0.52

Forced Upsampling: Power of 2 (x2, x4…)

 

With those settings you can also use 1.00, 1.02… cutoff frequency to get the focus effect skatbelt describes, but at the expense of loosing details (if I keep the photography analogy : lowering the cutoff freq. is also opening the lens (aperture) so loosing the depth of field).

[RunHomeSlow]

Hello All,

 

this is so complicated tweaking measures... i just bought a used Wyred DAC 2 DSDse plugged direct in Proceed AMP2 and was pretty happy with the sound coming from my Focal Cobalt speakers...

 

Read all the pages here... So, everybody here use oversampling files?

Isn't it blurrying the music to oversampled between leaving it to NONE ?

Isn't a fake produces sound or is it really better to go at 352 and 384 with my Wyred DAC ?

 

Sorry for my english lol

[pl_svn]

briefly and very roughly:

 

every DAC but true NOS ones oversamples anyway internally

doing it yourself before feeding the DAC might lead to overriding your DAC's own filters and shape one you like better

[RunHomeSlow]

briefly and very roughly:

 

every DAC but true NOS ones oversamples anyway internally

doing it yourself before feeding the DAC might lead to overriding your DAC's own filters and shape one you like better

 

 

Hello,

 

i couldn't see in Wyred Manual if my DAC as Filters.

i don't know why, but i'd like too max out all files at 352/384 from 44/48/96/192... flac lol

if my DAC was originally able to do 32/384 why shouldn't i do it?

 

but opposite like some said it here before, i couldn't see big differences or some...

applying oversampling settings of people done here.

[Jud]

Yes, the Wyred DAC uses upsampling and filtering, first to 352.8 or 384, then to DSD. Almost every DAC does this. Only a tiny handful do not.

 

So the question isn't whether or not to upsample, it's where to do it: in the DAC or in software.

 

One reason there is such a wide variety of settings people are using is because many DACs don't allow 352.8 or 384kHz input, so you may get a combination of the software's upsampling and your DAC's.

[RunHomeSlow]

Yes, the Wyred DAC uses upsampling and filtering, first to 352.8 or 384, then to DSD. Almost every DAC does this. Only a tiny handful do not.

 

So the question isn't whether or not to upsample, it's where to do it: in the DAC or in software.

 

One reason there is such a wide variety of settings people are using is because many DACs don't allow 352.8 or 384kHz input, so you may get a combination of the software's upsampling and your DAC's.

 

 

Thank you for reply,

 

So you say even if i choose in Audirvana settings: NONE upsampling,

the DAC will do it anyway to 352 or 384, so i better choose myself the way the filters will sound at the end in izotope settings ?