Izotope SRC

[WRWSTD]

Lol.

Yeah, its hard just trying to find that perfect setting of all the sliders. Frustrating, yet fun at the same time.

[jhwalker]

I've settled on the following:

 

Steepness: 20

Filter length: 500,000

Cutoff freq: 0.93

Anti-aliasing: 200

Pre-ringing: 0.33

 

I feel like the relatively shallow steepness coupled with the lower cutoff frequency minimizes 1) aliasing into the audible spectrum and 2) ringing. I also tend to prefer minimum phase to linear, thus the relatively low pre-ringing setting. I've modeled these settings in Izotope RX, and it produces a pretty clean output with almost no aliasing in the audible spectrum, very little pre-ringing, and fairly short post-ringing - sounds good, too

[skatbelt]

I went big time in the opposite direction. I used a steepness of 144, and filter length of 192,000. Well, I have to say that my original settings sounded better to me.

There seems to be a somewhat smeared sound especially on the higher frequencies and a little less overall clarity with the revised settings.

 

The biggest correlation is between steepness and pre-ringing. See my earlier posts. Did you experiment with lower values of pre-ringing or kept it at 1.00 in your quest?

[WRWSTD]

The biggest correlation is between steepness and pre-ringing. See my earlier posts. Did you experiment with lower values of pre-ringing or kept it at 1.00 in your quest?

With the steepness set to 144, I mostly kept the pre-ringing at 1.0 . Any significantly lowered number (such as .5) would seem to make the music sound disconnected as a whole. Although, transients like a cymbal tap would sound clearer yet muted with .5 as a pre-ringing value.

[WRWSTD]

I've settled on the following:

 

Steepness: 20

Filter length: 500,000

Cutoff freq: 0.93

Anti-aliasing: 200

Pre-ringing: 0.33

 

I feel like the relatively shallow steepness coupled with the lower cutoff frequency minimizes 1) aliasing into the audible spectrum and 2) ringing. I also tend to prefer minimum phase to linear, thus the relatively low pre-ringing setting. I've modeled these settings in Izotope RX, and it produces a pretty clean output with almost no aliasing in the audible spectrum, very little pre-ringing, and fairly short post-ringing - sounds good, too

 

I have tried your lower Steepness and Cutoff Freq. settings and I agree, they sound good.

I also have downloaded the Izotope RX last night, (Thanks for the idea!) and started experimenting with the resample function. It looks like I will learn a lot from this.

[Dyson]

With the steepness set to 144, I mostly kept the pre-ringing at 1.0 . Any significantly lowered number (such as .5) would seem to make the music sound disconnected as a whole. Although, transients like a cymbal tap would sound clearer yet muted with .5 as a pre-ringing value.

 

Certainly true but then that's where the other settings come in as well and do interact each other (I have pre-ringing set 0.51). It's not the pre-ringing alone that disconnects the instruments or mutes them, it's always a combinaison. That's why when I played around with the settings I first looked at the couple steepness/anti-aliasing, then started with the filter length and so on. But regularly coming back on previous settings, improving (very) little step by little step.

 

For sure if could have some knowledge of the implementation (and limitations) that would help understanding how to play with the settings.

[bits]

The izotope SRC looks great and Id like to try it, but reading this thread it seems its only for apple OS player apps?

 

Is there anyway to use it for playback in windoze eg as a VST run in Foobar?

 

Thanks.

[slk350amg]

Hi to everyone.

 

I have a Metrum Acoustics Pavane nos Dac. Can you please recommend all the parameters (especially the advanced ones) for Izotope filter for the following 2 scenarios ?

 

1. Playing redbook material

2. Hi rez PCM

3. DSD (64 &1 28)

 

Thank you

[earflappin]

Not sure if this has already been covered or not, but would love to hear from someone like Alexey Lukin or others who are expert in Izotope SRC as to the significance of the closed form filter touted by Schiit Audio that appears to heavily influence the sonic performance of their Yggdrasil and other multi-bit DACs.

 

This closed form filter is said to preserve the original samples, adding new interpolated samples in between, and optimizing for both frequency and time.

 

Thanks.

[orgel]

Not sure if this has already been covered or not, but would love to hear from someone like Alexey Lukin or others who are expert in Izotope SRC as to the significance of the closed form filter touted by Schiit Audio that appears to heavily influence the sonic performance of their Yggdrasil and other multi-bit DACs.

 

@Miska has a couple of closed-form filters available in HQPlayer, so he might be willing to shed some light on what they're all about. Maybe you should post in the "HQ Player" thread so he'll be sure to see it.

 

--David

[earflappin]

Great idea David. Thanks.

[Alexey Lukin]

Not sure if this has already been covered or not, but would love to hear from someone like Alexey Lukin or others who are expert in Izotope SRC as to the significance of the closed form filter touted by Schiit Audio that appears to heavily influence the sonic performance of their Yggdrasil and other multi-bit DACs. This closed form filter is said to preserve the original samples, adding new interpolated samples in between, and optimizing for both frequency and time.

Honestly, from their description I don't see much benefit. “Closed form” means that you can describe the filter with a simple formula, as opposed to calculating it as a result of the optimization algorithm. Many competing SRC filters, including iZotope SRC (in some modes), are closed-form filters too.

 

The fact that they retain the original samples and insert (interpolate) the new samples between them may sound significant. But I believe that it doesn't make much sense in the audio world. Again, some competing filters work like that and there are good ones and bad ones among them (just think of linear interpolation).

 

What matters more is that their filter is “optimized in both time and frequency domains”. It means that the designers are conscious of ringing and frequency response. In competing filters, the trade-off between the time and frequency performance is set by the filter length, which affects the cutoff steepness and the amount of ringing. The fact that a certain filter is “optimized” in both domains simply means that the designers have chosen a certain balance between time and frequency performance. It does not automatically mean that the filter is better than what you can achieve with competing products, esp. when you have access to the steepness control.

[audiventory]

Here desription of close form filter (if I right understood that is it) http://www.ee.cityu.edu.hk/~hcso/ifac03_1.pdf

 

The principle can be applied for multiply upsampling only.

 

In my opinion, it have sense for integer numbers calculations that have less precision than float point. As example into hardware applications.

 

With increasing of bit-depth using source samples also lose sense.

[earflappin]

Honestly, from their description I don't see much benefit. “Closed form” means that you can describe the filter with a simple formula, as opposed to calculating it as a result of the optimization algorithm. Many competing SRC filters, including iZotope SRC (in some modes), are closed-form filters too.

 

The fact that they retain the original samples and insert (interpolate) the new samples between them may sound significant. But I believe that it doesn't make much sense in the audio world. Again, some competing filters work like that and there are good ones and bad ones among them (just think of linear interpolation).

 

What matters more is that their filter is “optimized in both time and frequency domains”. It means that the designers are conscious of ringing and frequency response. In competing filters, the trade-off between the time and frequency performance is set by the filter length, which affects the cutoff steepness and the amount of ringing. The fact that a certain filter is “optimized” in both domains simply means that the designers have chosen a certain balance between time and frequency performance. It does not automatically mean that the filter is better than what you can achieve with competing products, esp. when you have access to the steepness control.

 

Alexey, thanks very much for your informative response. Very helpful.

[Alexey Lukin]

Here desription of close form filter (if I right understood that is it) http://www.ee.cityu.edu.hk/~hcso/ifac03_1.pdf

I think that it's unrelated. This paper describes closed-form notch filters that are completely different from oversampling filters.

[gianco]

After having tried many settings, in my system, the ones that work best are the following:

Steepness: 32 / Filter max lengh: 2,000,000 / Cutoff freq.: 1,00 / Anti-aliasing: 200.0 / Pre-ringing: 1.00

 

With these settings the sound of my system is very natural, clear and lively especially with classical music and acoustic jazz.

 

 

 

 

Mac Mini / Wireworld Starlight 7 USB 2.0 / Gold Note Dac 7 + Gold Note external PSU 7 AC / Klimo Merlino Gold / Kilmo Kent Gold / Dynaudio Contour S 3.4

[fbczar]

I have an Emotiva XMC-1 processor. The XMC-1 uses Burr Brown DSD1796 24bit DACS. Given that the iZotope settings in Audirvana 2.3.3 are somewhat DAC dependent I would appreciate input relative to the settings that might work well with my equipment. Thanks.

[amber]

New to this forum and the audiophile world. A bit of context first. I am building my first audio system and I am researching a musical / not fatiguing system versus analytical. I listen essentially to jazz / blues / acoustic music but not just. My experience is very limited with audiophile grade equipment. I sold recently a Burson audio headphone amp (didn't like the DAC/sabre signature). I am also using Audiravana for a while (love it !).

 

I am looking forward to purchase a DAC which is a complex decision. After countless hours of reading this forum and head-fi I narrowed down my choice to 2 R2R DACs (actually 3 but the last one is out of budget for now..):

- Audio-GD Dac-19 Standard version (8x oversampling by default - possible to use NOS mode but NOS mode is not performing at the level of the new DAC below)

- Audio-DG Dac-10 NOS version (ºÍ§Ó­µ响)

- Metrum Musette

 

My readings lead me to believe I should prefer Audirvana/iZotope (software DSP) versus an on-board DAC DSP (even if I understand high-end DACS have excellent DSP/Filters but feel under 1K USD I will be much better served by iZotope)

 

I am trying to confirm if it is fair to assume Audirvana with iZotope oversampling disabled combined with the standard DAC-19 (x8) and Audirvana 8x iZotope combined with the NOS version of the same DAC sonic signatures should be closer than a pure NOS approach (no oversampling in Audirvana combined with the NOS DAC). If I am correct I understand it should be possible to reproduce the standard DAC-19 oversampling filter with iZotope settings (this how the thread started with the Ayre filter no ?)

 

NOS DACs sound signatures seem specific and I read many times not everyone like it (read most don't like it). Overall I wonder if by oversampling with the software and using a NOS DAC I am taking a risk not liking the NOS sound.

 

This is very theoretical for me now but can't wait to finish building my first system and start using my ears. But until then I rely on my reading and experience of people like you on this forum to make the right choice. Thanks in advance for your comment..

 

Overall I am leading toward the NOS version of the DAC

 

I recently read a quote I found appropriate in the context:

"At a time when we're having to take such difficult decisions about how to cut back without damaging the things that matter the most, we should strain every sinew to cut error, waste and fraud"

[Dyson]

Since A+ 2.5.1 and up is now capable of PCM -> DSD conversion, but that the conversion also uses the iZotope up sampling, I went back here and started trying to achieve again the most neutral iZotope settings. Based on Jud's remark that he prefers low values of the steepness in order to keep the pre-/post-ringing as little as possible, I finally came to the following settings for my iFi micro iDSD:

 

Steepness: 4

Max filter length: 524'288

Cutoff freq.: 1.03

Anti-aliasing: 50.0

Pre-ringing: 0.97

Forced Upsampling: Power of 2

 

Maybe that helps someone as well... I'm now listening with those settings since a few days and each time that I try again without up sampling, I quickly switch it on again as, at least on redbook content, it is much better with (even if not converting to DSD).

[orgel]

Steepness: 4

Max filter length: 524'288

Cutoff freq.: 1.03

Anti-aliasing: 50.0

Pre-ringing: 0.97

Forced Upsampling: Power of 2

 

I'm curious about how you came up with the value for max filter length. Could you explain your methodology?

 

Thanks.

 

--David

[Dyson]

I'm curious about how you came up with the value for max filter length. Could you explain your methodology?

 

Thanks.

 

--David

 

I started with the idea that in computer science for integers computing it is based on power of 2 values and tested those, to me the one I'm currently using sounded better than "just" 500'000 so I kept it. I tried the same approach for steepness (4 was better than 5) and anti-aliasing (but here 50 was better than 64). So it is just a little idea to start with and then trying & listening (the famous trial & error method with lots (really lots) of fails!).

 

PS: for the cutoff values very close to 1 (1.01-1.03), I kept a sum of 2 with the pre-ringing (i.e. 0.99-0.97), some how those seems to be linked has well

[WRWSTD]

Dyson,

I agree with what you are saying and I will add to this idea.

Please try:

Steepness: 12

Max filter length:1,806,336

Cutoff freq.: 0,96

Anti-aliasing: 72.0

Pre-ringing: 1.00

Forced Upsampling: Power of 2

Memory buffer: 3072

The reasoning for this is that ringing will be kept to a minimum without much aliasing. The sample rates, buffer size and various filter settings should all fit within the max filter size with no remainder.

[WRWSTD]

The setting for steepness and antialias was based on the premise that there seems to be a phase relationship between the two.

[Dyson]

Dyson,

I agree with what you are saying and I will add to this idea.

Please try:

Steepness: 12

Max filter length:1,806,336

Cutoff freq.: 0,96

Anti-aliasing: 72.0

Pre-ringing: 1.00

Forced Upsampling: Power of 2

Memory buffer: 3072

The reasoning for this is that ringing will be kept to a minimum without much aliasing. The sample rates, buffer size and various filter settings should all fit within the max filter size with no remainder.

 

So my basic assumption is not that crazy :-)

Your settings sound pleasing and relaxing, on my system they lack some bass body and the mids are a bit too forwarded. They remind me my trials of last year, when I couldn't really be convinced that up sampling was better than not.

It has lots to do with personal taste and I'm personally searching for settings that are as close as possible to the original (redbook vs. up sampled or up sampled vs. hi-res version).

thanks for sharing

[copy_of_a]

I’ve investigated how the relation of „Steepness“ and the „Anti-Aliasing“-Filter in Audirvana Plus’ iZotope SRC plugin works.

 

Why?

- First… the original iZotope software suite does not contain an „Anti-Aliasing“ option in its resample component. Anti-Aliasing is handled by the steepness of the resampling filter alone. So I was curious...

- I favor low „Steepness“ for (overall) low ringing.

- I also favor lower pre-ringing (most of the music I like sounds a bit lifeless with linear upsampling).

 

Generally lower „Steepness“ produces lower ringing.

However, lower „Steepness“ is also less effective in removing high frequencies above the Nyquist limit.

 

Here’s a screenshot of the „Fabfilter Pro Q2“ (for previewing purpsoes set to „zero latency“) plugin running in realtime-mode when upsampling in A+ with an often posted low steepness SRC-setting for A+ („Jud 1“: steepness: 5db / Filter: 10.000 / Cutoff: 1 / Anti-Aliasing: 200 / pre-ringing: 1).

 

As you can see, there’s steep rise of high frequencies right above the Nyquist limit (20.050 here, since the source is 44.1kHz).

The Anti-Aliasing Filter cuts everything above the Nyquist limit but is shifting frequencies beyond the Nyquist limit (acting like a negative „peaking“/„bell“ filter with a very high Q-factor).

Maybe no reason to worry about when your system (hardware…: DAC, Amp, speakers) does not reproduce frequencies above 20kHz.

However, the steep rise of frequencies right above the Nyquist limit certainly does not look healthy… and depending on your hardware will introduce some blemishes to the sound („harsh“ highs for example).

 

Increasing the steepness of the filter reduces the amount (gain) of frequencies that passes the „Anti-Aliasing“ filter. However, again, this will also produce increased ringing.

 

What I finally decided to do to get low ringing, low pre-ringing and excellent filtering is the following…:

- Resample in iZotope RX (Version 4 in my case…) without any „Anti-Aliasing“-Filter applied (again, aliasing in the original iZotope suite is controlled by „steepness“ alone)

- my SRC settings finally are: steepness: 16db (for low ringing) / cutoff: 1 / pre-ringing: 0.72 (for low pre-ringing — at 16db steepness 0.72 produces literally no pre-ringing at all) and I only resample by power of 2 (so 44.1kHz to 176.4kHz for example).

- apply the Fabfilter Pro Q2 plugin set to high cut everything above the Nyquist limit with a 96db/octave slope in linear phase mode at the highest quality setting … 3 times in a row! … to effectively remove the unwanted rise of frequencies above the Nyquist limit

- save as 24bit TPDF-dithered file

(- re-tag the files for instances with „Yate“ if required...)

…

- I upsample and render only 44.1Khz and 48kHz files to 176.4kHz/192kHz … 88.2khZ and 96khZ get upsampled on the fly in A+ with the same SRC-settings (and Anti-Aliasing set to the lowest factor, so „50“ in A+) since a rise of frequencies above the Nyquist limit of such high res files certainly do not blemish the sound (in my system)…

 

Converting/filtering a simple CD may take 15-20 minutes or so on my 8-Core / 24MB RAM MacPro utilzing all the Core power available. And though it takes some time I am extremely pleased with the results.

 

What I also take from this… how can any software can produce such nice (precise) results „on the fly“ in realtime?

I do own A+, Fidelia and HQPlayer (and Amarra)… but I have to say I am really done with „on the fly“ upsampling/filtering.

Offline upsampling and high precision offline filtering is sooo much better …

 

Only drawback is disk space. But that’s very well handable nowadays…

 

… IMHO…