Izotope SRC

[Bob Stern]

Note that this discussion relates only to 64-bit floating point arithmetic, not 64-bit addressing of memory.

[bdiament]

Hi Bob,

 

According to the first 10/27/2012 post in this thread from the Audiofile Engineering forum, Audiofile Engineering stated that they use only 32-bit floating point arithmetic in Sample Manager:

Audiofile Engineering - Forums

 

In contrast, they use 64-bit floating point arithmetic in Triumph and Fidelia.

 

Because SRC requires millions of arithmetic steps (note the option to increase the filter length to 2 million taps), higher precision arithmetic reduces rounding errors. That's why Metric Halo DSP uses 80-bit arithmetic even though its A/D converters are "only" 24 bit precision. Do you think BJ would say that 32-bit DSP would have been just as good?

 

I believe there is some confusion between the word length employed for addressing and that employed for internal processing of the audio. These are not at all the same thing. One deals with operating efficiency and the other deals with audio precision.

 

Sample Manager does its internal processing at 32-bits but saves its internal temp files at the word length of the original source, so if you feed it a 16-bit source, that is the word length of its temp files. Triumph, according to what AE tells me, is 64-bit in terms of the *addressing* it uses to operate but *not* in terms of the internal processing, which is at 32-bit. (Triumph automatically converts to 32-bit float upon import, so you can feed it a shorter word length file and still process fully at 32-bits.)

 

Since we're talking about 64-bit addressing and not 64-bit internal math, while I can see an increase in operating efficiency, I don't see how there is any increase in precision. What you are referring to with Metric Halo's DSP is 80-bit *internal processing*. That is where there will be an effect on the audio.

 

I remain open to learning something new of course. What I've said above is based on conversations I've had with folks at AE, asking these very questions. If I have misrepresented what they told me, the fault is my own. But as I keep looking over the communications, (unless I hear differently from them) I believe I have correctly understood them.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[goldsdad]

I was wrong. Apologies Orgel. The advanced version of Fidelia will do offline upsampling if you check "create preferred formet when adding to library" in the output prefs.

 

 

geoff

 

 

No apologies necessary. I was able to do a clean install today and verified that the offline upsampling works, just as you say. It's a little fiddly, so if anyone wants, I can post detailed instructions.

--David

 

 

Last time I checked (over a year ago), Fidelia employed Apple's resampling for offline resampling, whereas there was a choice of Apple or iZotope for on-the-fly use. How have you determined that iZotope is being used for offline resampling?

[Geoffrey Armstrong]

Interesting: Based on the passband and transition plots (on the SRC comparison page) of the Sonic HD filters, they are all "apodizing" filters--in that they cut off quite a bit before Nyquist. Unless I am still mis-understanding the properties of an apodizing filter.

 

And if anyone here has Amarra with the Sonic HD SRC engine, can you tell us what parameters it allows you to control?

 

You have to buy Sonic Studio Process to get the "Classic" SRC. I wonder if this isn't iZotope?

 

Within Amarra the "Classic" SRC is greyed out (non-selectable).

 

It seems they are claiming that Sonic HD is superior; but as per the Amarra manual, within Amarra it is also limited.

 

SRC is limited to 192khz max, and you get additional curves and dither types in Sonic Studio Process.

 

This is an additional $249 license on-top of Amarra. Use of the more limited version within Amarra should determine whether the additional features for Sonic HD are worth it; but you can't evaluate the effects of the "Classic SRC" at all without the additional license.

 

The following is taken from the Sonic Studio Process page:

 

http://www.sonicstudio.com/sonic/products/sonicstudio_process

 

"Classic" SRC

 

 

Painstakingly created conversion tables

Superb natural sound consistency

Original ground-breaking SRC technology

 

 

"Sonic HD" SRC

 

 

Newly created to Sonic's exacting standards

Completely transparent

Best-in-class SRC design

 

 

This video shows you which parameters you can control in Sonic Studio Process

http://www.sonicstudio.com/tutorials_pro/Process/Sonic_Studio_Process_Overview02_3.mov

 

In Amarra I get two filters: Steep and Gentle. An additional Gentlest is greyed out.

 

For dither I have: none, Trangular PDF, Rectangular PDF. Noise Shaped follows; but is greyed out.

 

Hope it helps,

 

geoff

[bdiament]

The "Classic" Sonic SRC is not iZotope. It is the SRC from the old Sonic Solutions application (which over the years, has morphed into soundBlade, from which Amarra is derived).

The HD is something altogether new.

 

There was a while when iZotope was included but this has since been replaced in their software by their own algorithms.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[Geoffrey Armstrong]

The "Classic" Sonic SRC is not iZotope. It is the SRC from the old Sonic Solutions application (which over the years, has morphed into soundBlade, from which Amarra is derived).

The HD is something altogether new.

 

There was a while when iZotope was included but this has since been replaced in their software by their own algorithms.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

 

Thanks for the clarification Barry. While we're talking I do realise from previous posts that you are using re-sampling to down-sample High-Def recordings, whereas most of us here are using it to up-sample and trying to squeeze more quality out of plain Redbook.

 

Sometimes though I also need to down-sample and was wondering if there are particular settings that you favour for this within iZotope? Apologies in advance if you've already posted those and I missed them. I'm also interested if you believe the same settings would apply equally well to up or down-sampling within iZotope. For someone like myself who has found their preferred up-sampling settings, would they be likely to hold true for down-sampling as well?

 

Thanks,

geoff

[bdiament]

Hi Geoff,

 

Thanks for the clarification Barry. While we're talking I do realise from previous posts that you are using re-sampling to down-sample High-Def recordings, whereas most of us here are using it to up-sample and trying to squeeze more quality out of plain Redbook.

 

Sometimes though I also need to down-sample and was wondering if there are particular settings that you favour for this within iZotope? Apologies in advance if you've already posted those and I missed them. I'm also interested if you believe the same settings would apply equally well to up or down-sampling within iZotope. For someone like myself who has found their preferred up-sampling settings, would they be likely to hold true for down-sampling as well?

 

Thanks,

geoff

 

At this point in time, all I can say is that the settings I use depend on the program material I am converting.

Beyond that, I'm not comfortable revealing details of the recipes I use in my work and I hope you understand.

 

I will say that to my ears, as the filter moves toward what has been called an "apodizing" type, while preringing might be minimized or avoided, I find there are other, perhaps higher sonic prices to pay. But that's just me.

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[Geoffrey Armstrong]

Hi Barry,

Thanks again and, of course I understand. You've given me a basis to work from, which I'm sure will prove useful.

Best,

geoff

[Marino Marco]

Currently using NOS Metrum Octave DAC & BADA

Steepness: 4

Filter Length: 2,000,000

Cut-Off: 1.03

Anti-Aliasing: 200

Pre-ring: 1.00

 

I don't think I'm at my final resting place with these settings but they are dialed in fairly well for my set-up. Having an NOS dac it's great to have this, I can get a traditional over-sampling sound, very bold and full with these settings. Or, switch back to NOS and get more delicacy and nuance. It's like having 2 Dacs...nothing to complain about there.

 

The only setting I haven't spent time with is the Filter Length Max. For this, I've just done what Jud has done

excuse me but I have tried your settings. It's unbelievable how it sharpened the sound of my audio technica ath m40fs headphones. And I'm not talking about lossless, i'm talking about AAC 256 files. hahaha

 

Upsampling with your settings improved the sound in a way thats pleasing as hell to my ears. I can hear every instrument in details, and i'm talking heavy music here. Well, sorry but i'm enthusiastic right now.

 

My blabbing may sound amateurish but I'm into this Audirvana thing for at least a year now and have purchased a couple of high definition audio files in the past, Own a pretty good DAC/Recording interface and a decent set of cans.

 

Thanks.

[Geoffrey Armstrong]

These may be the perfect settings for you; but you may also want to try some others that Ken and I came up with between us. If you do first make a note of your current settings, then change Steepness to 80, Cut-off to 0.95 and pre-ringing to 0.99.

 

You might like these even better. If not just go back to the settings in your post above.

geoff

[wwaldmanfan]

These may be the perfect settings for you; but you may also want to try some others that Ken and I came up with between us. If you do first make a note of your current settings, then change Steepness to 80, Cut-off to 0.95 and pre-ringing to 0.99.

 

You might like these even better. If not just go back to the settings in your post above.

geoff

 

These settings suggestions are interesting and fun to read, but very DAC-dependent, so quite subjective.

I dislike the sound I get with the steepness set at more than single digits; On my rig, the instruments and vocal separation become increasingly muddy.

I also remember previous discussions when the pre-ringing was suggested as low as .65 to .75, which is pretty extreme, but, I suppose, appropriate with higher steepness.

I don't hear much difference with the Nyquist cut-off one way or the other, so I leave it at default.

Another member did suggest lowering the pre-ring from default to .99 to clean up sibiliance, which worked well for me, in conjunction with my low steepness setting. So, right now, for 4x upsampling from Redbook, and 2x upsampling from 24/88.2 and 24/96 downloads, I'm at:

 

Steepness: 5

Filter ML: 1M

Cutoff: 1.00

AA: 200

P-R: .99

 

Basically, just a little bump up from "native" playback, which is how I like it. YMMV.

[Savage]

These may be the perfect settings for you; but you may also want to try some others that Ken and I came up with between us. If you do first make a note of your current settings, then change Steepness to 80, Cut-off to 0.95 and pre-ringing to 0.99.

 

You might like these even better. If not just go back to the settings in your post above.

geoff

 

Yes, both Geoff and I liked these. Really nice for vocals & a bit of presence with a smoother sound.

 

Steepness: 80

Filter Length: 2,000,000

Cut-off: 0.95

Anti-Aliasing: 200

Pre-ring: 0.99

 

As Geoff mentioned before it would be great to be able to name & save different settings and quickly select the one you're in the mood for on playback.

[Geoffrey Armstrong]

I should have made clear that both Ken and I are using Metrum DAC's and the same applies to Marino. Yes they are system dependant and taste dependant. To my surprise though, I've also found them to work very well through a pair of Grimm LS1 speakers. The Grimm's have a great combination of naturalness with accuracy in my view and so perhaps not so far removed from the results of some systems which employ Metrum DAC's, even though the DAC in the Grimm is obviously quite different.

[Marino Marco]

These settings suggestions are interesting and fun to read, but very DAC-dependent, so quite subjective.

Fun to read? I think its fun to try them out! haha

I think we should let the technical talk aside for a while and post more settings.

 

Thanks for the settings, guys. Will definitely try these later today. awesome.

[Geoffrey Armstrong]

Last time I checked (over a year ago), Fidelia employed Apple's resampling for offline resampling, whereas there was a choice of Apple or iZotope for on-the-fly use. How have you determined that iZotope is being used for offline resampling?

 

This is correct. I verified it with AE today. when I asked: "If I check Create Preferred version when adding to library and the preferred version involves up-sampling to a higher sample rate, which sample rate converter does Fidelia use?" The response was: "It currently uses Apple's standard SRC."

 

They confirmed that Sample Manager allows you to choose the SRC. So only Sample Manager will up/down sample using your Izotope settings. Apart from, of course, RX from iZotope themselves.

[Superdad]

Geoffrey: Thank you for clarifying that for all of us. Seems like if I want to do batch off-line upsampling with custom iZotope settings, then Sample Manager (plus the Action pack) will be the way to go. Just need to set aside $100 for the license.

[goldsdad]

Thanks for the confirmation, Geoffrey.

[Geoffrey Armstrong]

You're both welcome. I did some offline up-sampling today with Sonic Studio Process as provided with my version of Amarra and I was very impressed with the results.

 

In the version bundled with Amarra one curve is left out (gentlest) and one dithering paramater (Noise Shape). The classic SRC is also missing. To get these features would cost an additional $249, yikes! I'm so encouraged by the results though, that I'd really like to hear from anyone else who has both Amarra and Sample Manager or RX.

 

It would be interesting to get opinions on how people feel Sonic HD compares with iZotope. There are less settings to fiddle with than iZotope.

 

Using Sonic HD, my impression is that one of the dithering options is required (I also tried the no dithering option). Both dithering options with both slopes gave excellent results. I think my favourite is the steep slope with Triangular PDF dithering.

[jamesg11]

Back on topic ... Spent an hour or 2 reading thru suggestions from beginning of thread, & trying them out on ripped Beatles USB stick @ 24/44.1 upsampling to 24/176.4, handful of Rubber Soul tracks. So a macmini/ Mytek FW here gets to optimum with Alex's 7/1,150,000/1.04/200/0.65. Tried the sigma delta idea of 24/1,550,000 but happier with former. Some big changes coming up with mini, ext psu etc so will go around again & see where we are then!

[wwaldmanfan]

Having steadfastly clung to the notion of "powers of two" upsampling, I got crazy and experimented with maximum upsampling (24/192) in Audirvana Plus, and, on redbook 16/44.1 rips, it sounds better...

 

Steepness 7

Filter ML 1.4M

Nyquist 1.03

AA 200

P-R .97

[bdiament]

Hi wwaldmanfan,

 

Having steadfastly clung to the notion of "powers of two" upsampling, I got crazy and experimented with maximum upsampling (24/192) in Audirvana Plus, and, on redbook 16/44.1 rips, it sounds better...

 

Steepness 7

Filter ML 1.4M

Nyquist 1.03

AA 200

P-R .97

 

;-}

As I mentioned way back in post #290, "In my experience, the better algorithms (and I personally rank iZotope's 64-bit SRC by Alexey Lukin as the most transparent I've yet heard), can handle the more complex math involved in finding the (higher) *common* multiple. They will, to my ears, create far more transparent results, even at non-integer conversion (e.g. 192k to 44.1k or vice versa), than other algorithms can with the simpler math involved in integer conversion."

 

Best regards,

Barry

Soundkeeper Recordings

The Soundkeeper | Audio, Music, Recording, Playback

Barry Diament Audio

[jamesg11]

Yes ... I've got a J Kenny Ciunas usb-spdif converter on the way, & I'm hoping to resurrect my Cyrus DacX by going max 24/192 iZotope into it (avoids DacX internal upsampling). This will be tres interesting.

[Bob Stern]

I believe there is some confusion between the word length employed for addressing and that employed for internal processing of the audio. These are not at all the same thing. One deals with operating efficiency and the other deals with audio precision.

 

Sample Manager does its internal processing at 32-bits but saves its internal temp files at the word length of the original source, so if you feed it a 16-bit source, that is the word length of its temp files. Triumph, according to what AE tells me, is 64-bit in terms of the *addressing* it uses to operate but *not* in terms of the internal processing, which is at 32-bit. (Triumph automatically converts to 32-bit float upon import, so you can feed it a shorter word length file and still process fully at 32-bits.)

 

Since we're talking about 64-bit addressing and not 64-bit internal math, while I can see an increase in operating efficiency, I don't see how there is any increase in precision. What you are referring to with Metric Halo's DSP is 80-bit *internal processing*. That is where there will be an effect on the audio.

 

I remain open to learning something new of course. What I've said above is based on conversations I've had with folks at AE, asking these very questions. If I have misrepresented what they told me, the fault is my own. But as I keep looking over the communications, (unless I hear differently from them) I believe I have correctly understood them.

 

 

Barry,

 

Thanks for your advice, and I apologize for my delay in following up.

 

I just exchanged emails with "TJ" in iZotope customer support. He says you are correct that there is no difference in the precision of the arithmetic used internally by "iZotope SRC" and "iZotope 64-bit SRC". Furthermore, he says they both use 64-bit (double precision) floating point arithmetic! (Reasonably close to Metric Halo's 80-bit.)

 

Bob

[Geoffrey Armstrong]

Apologies if I seemed to stray off top. I thought it would be interesting though, to try some offline up-sampling and specifically compare offline up-sampling using Amarra with iZotope using Audiophile Engineering's Sample Rate Manager.

 

For those who believe up-sampling using a general purpose computer is a better way of doing it than up/over-sampling in a DAC, it must be interesting to also dertermine if this is better done offline than on the fly, as certain people, such as Sonic Solutions and the guys behind jPlay seem to believe.

 

Up-sampling in Amarra has limited options compared to iZotope using Sample Rate manager. It has the advantage of being faster though. It also produced what I felt was a very good result. Izotope through Sample Rate Manager has many more options. Specifically the various permutations for dithering. Note: These are not available via on the fly resampling in Audirvana + or Fidelia. I took one track and up-sampled with each of the permutations.

 

Admittedly I was definitely in danger of becoming obsessive.

 

Nevertheless, I found significant differences in the results depending on the parameters chosen and I had some clear preferences.

 

Playing the resulting files back through the three different players; A+, Fidelia and Amarra, of course influenced the results significantly as well. I could still hear the characteristics of the different parameters chosen though, for each offline re-sampled version of the track.

 

A+ is my favourite player, in terms of SQ, design, features and stability. It's interesting to compare players though. Fidelia is perhaps more accurate and neutral sounding and Amarra is at the other extreme, being warmer and more romantic. A+ is pitched right in the middle. Perhaps it strays from absolute neutrality; but in very pleasing ways.

 

The end result, will obviously depend on the rest of the system. The software player chosen is now, obviously a very important component in our systems.

[Jud]

Over at SRC Comparisons have a look at the "Passband" or "Transition" test results for iZotope Rx Adv 2 (High Steepness) compared to Sonic HD for the same test. In fact, compare iZotope Rx Adv 2 (High Steepness) on those tests to any other SRC listed. The results aren't shoddy for iZotope Rx Adv 2 (High Steepness) on any of the other listed tests either, but those two test results are just visually impressive even to a layperson like me.