Izotope SRC

This thread is for the in-depth discussion of Izotope's SRC, described by Barry Diament as "the most transparent SRC I've heard to date." Please share your preferred izotope filter settings and observations about the effects of the various filter parameters.

My first observation/question for everyone: I find that I get better results upsampling my redbook files to 88.2kHz if they're older CD's that were originally transferred to digital at 44.1khHz, but that I prefer upsampling my CD's to 96kHz when they're newer remasters that were originally transferred to digital at 24/96 and then downsampled to CD. It's subtle, but I find I get smoother sound when I do it this way. Does anyone else find this to be the case, or am I imagining things?

My first observation/question for everyone: I find that I get better results upsampling my redbook files to 88.2kHz if they're older CD's that were originally transferred to digital at 44.1khHz, but that I prefer upsampling my CD's to 96kHz when they're newer remasters that were originally transferred to digital at 24/96 and then downsampled to CD. It's subtle, but I find I get smoother sound when I do it this way. Does anyone else find this to be the case, or am I imagining things?

For CD's 176.4kHz is much better than 96kHz and in my system 88.2kHz is even better than 96kHz!

My first observation/question for everyone: I find that I get better results upsampling my redbook files to 88.2kHz if they're older CD's that were originally transferred to digital at 44.1khHz, but that I prefer upsampling my CD's to 96kHz when they're newer remasters that were originally transferred to digital at 24/96 and then downsampled to CD. It's subtle, but I find I get smoother sound when I do it this way. Does anyone else find this to be the case, or am I imagining things?

I normally up sample by 2X under Audirvana Plus, then I never tried 44.1 to 96kHz. How do you this up sample, since I guess I don't have the patience to use an external up sampler?

In some recordings I up sample 44.1 to 176.4kHz with very good results, but I guess (as you stated) that this is recording dependent (the master recording source), but also DAC dependent.

Roch

What settings would emulate the following?

http://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

These are my settings from the other thread:

I've been going between these 2 sets of settings based on how the high-frequencies are attenuated. I feel that the edge might go to the 0.90 settings since symbols etc. sound a little more crisp and present. Adjust to taste based on your taste and your system's frequency response. For me I tend to lean toward the 0.90 filter because the highs, and air might bit just a touch sharper. On very bright systems/recording, one could go with a 0.88 cut-off filter with a steepness of 23. More rolled-off but sounds very analog.

Still need to play with filter length, and to a much lesser extent, pre-ringing. 0.6 +/- 0.1 sounds about right. Only a very slight amount of pre-ringing with these settings.

Power of 2 Upsampling

Steepness: 26

Filter Max Length: 1,800,000 (still testing)

Cut-Off: 0.90

Anti-Aliasing: 200 (max)

Pre-Ringing: 0.60

-0.1: 18,570 Hz

-1: 19,140

-3: 19,530

-6: 19,850

-113: Nyquist

Steepness: 24

Filter Max Length: 1,800,000 (still testing)

Cut-Off Freq.: 0.89

Anti-Aliasing: 200 (max)

Pre-Ringing: 0.60

-0.1: 18250 Hz

-1: 18820

-3: 19270

-6: 19600

-114: Nyquist

You have so much more quality, if you are willing to accept frequency response linearity about 19k Hz. A relatively shallow "brick wall" filter, with it's lower pre and post-ringing distortion, lower phase shifting, and good attenuation at Nyquist which helps reduce a lot of the digital hash not only from the playback filters, but possibly from downsampling during mastering.

Resampled Pulse Waveform using 0.89 / 24 settings given above:

What settings would emulate the following?

 

http://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

Maybe something like:

Cutoff: 1 (because of the -6db at 22050)

Preringing: 0 (no pre-ringing)

Steepness: 3 (just an estimate)

Resampled Pulse Waveform using 0.90 / 26 settings given above:

What settings would emulate the following?

 

http://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

Which of the 4 sets of graphs Charlie is presenting are you wishing to emulate? I am assuming the last set ("Best of Both Worlds").

I can't tell you the exact settings (Levandiier is more handy with the iZotope RX2 Advanced software package trial than I am--though I bet his 10-day trial is going to expire soon on this $1200 s/w pkg.). But it is clear that you would need a cutoff freq. of less than 1, a pre-ring setting fairly low (maybe around 0.5), and still a roll-off curve much steeper than I prefer, maybe around 35? These are wild guesses on my part. I am sure Levandiier will chime in shortly and post some graphs...

Maybe something like:

 

Cutoff: 1 (because of the -6db at 22050)

Preringing: 0 (no pre-ringing)

Steepness: 3 (just an estimate)

 

 

[ATTACH=CONFIG]4643[/ATTACH]

Ha, we must have been typing at the exact same time! But with the "0" pre-ring setting I think you were attempting to replicate the 3rd set of Ayre's graphs and not the 4th set which is supposed to be their "Best of Both Worlds."

I'm curious why you estimate the steepness so low. Isn't the number supposed to represent the dB attenuation per octave? Clearly I am confused about the steepness curves!

Your cutoff guess of 1 does help me understand that parameter better, if a value of one equals 6dB down at the 1/2 Nyquist freq.

Ha, we must have been typing at the exact same time! But with the "0" pre-ring setting I think you were attempting to replicate the 3rd set of Ayre's graphs and not the 4th set which is supposed to be their "Best of Both Worlds."

 

I'm curious why you estimate the steepness so low. Isn't the number supposed to represent the dB attenuation per octave? Clearly I am confused about the steepness curves!

 

Your cutoff guess of 1 does help me understand that parameter better, if a value of one equals 6dB down at the 1/2 Nyquist freq.

Maybe I misread, but doesn't the "Best of Both Worlds" also have no pre-ringing? As for the steepness, I could be way off - a steepness of 3 was just a quick estimate to match the graph in the Ayre paper which kind of looks like -96db at 40khz give or take (plus the really low number of post-ringing cycles).

Thanks to all of you who have been sharing your settings and observations. I've been getting a lot from it.

Thus far some settings that levandier posted have worked best for me - though there were some qualities I heard with settings posted by superdad I liked a lot to. I could envision switching between them depending on the type of music.

Maybe I misread, but doesn't the "Best of Both Worlds" also have no pre-ringing? As for the steepness, I could be way off - a steepness of 3 was just a quick estimate to match the graph in the Ayre paper which kind of looks like -96db at 40khz give or take (plus the really low number of post-ringing cycles).

Yes, you are correct that Ayre's BoBW graph has no pre-ringing, thus "0" in iZotope would be required. But how the heck do they get that little amount of post-ringing?

I think we need to know for sure how the iZotope steepness setting works, and how you are getting your results. I had understood the steepness number to be how many dB per octave of attenuation is applied. So to be -96dB at 44kHz, the steepness, assuming the low pass begins at 22kHz, would have to be set at 96. Is that not correct?

Also Levandier, can you please give me your exact settings that you have been using for the downsample step in RX2 Advanced? Without knowing those I am finding it impossible to duplicate your graph results (when replicating your upsample settings). And which of the test files you posted are you using for the graphs you are posting? (Float, Int., or 24) THANKS!

ALEX

Hi, would one of you be kind enough to summarise what this is all about please?

I stumbled upon this from the Audirvana 1.4 thread. I could read back through that thread, but I'm not sure when this Izotope discussion began and from what I've seen it appears to be evolving.

Most of what is being talked about here is way over my head. But is it as simple as trying the settings you're talking about and seeing whether you like the results? Does it only apply if you're using up sampling?

Any help would be great for the probably many like me who can see this looks exciting but don't know how to begin to engage with this discussion.

Hi, would one of you be kind enough to summarise what this is all about please?

 

Most of what is being talked about here is way over my head. But is it as simple as trying the settings you're talking about and seeing whether you like the results? Does it only apply if you're using up sampling?

Yes, this discussion is only about settings experiments with the iZotope upsampling engine which Audirvana (and others) have licensed for OEM use in its software.

Actually, a well written and easy to comprehend place to start is Ayre's short white paper which discusses digital filter choices and tradeoffs: http://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf

So don't be scared off. Use your ears and experiment. And be sure to tell us what DAC, music files, and A+ settings you are using. Then we can make suggestions for you to try.

I had understood the steepness number to be how many dB per octave of attenuation is applied. So to be -96dB at 44kHz, the steepness, assuming the low pass begins at 22kHz, would have to be set at 96. Is that not correct?

I have not downloaded the iZotope RX program, but if I am interpreting the way it works by looking graphs, and then assuming that the operation of the steepness setting is analogous with Audirvana's settings, then I think levandier is providing the 1:1 match with 3db versus 96db. Also took a look at the Audirvana manual and found the following which seemingly supports levandier's (iZotope design) method. I could be totally wrong, but just wanted to give my observation if it helps.

Pre-ringing: all lowpass filters are characterized by the amount of ringing they introduce into their output. The steeper the filter, the greater the amount of ringing. There are two general types of ringing: pre-ringing and post-ringing. Pre- ringing will smear transients earlier in time, while post-ringing will smear transients later in time. This control allows users to select pre-ringing, post- ringing, or anywhere in between. A value of 1.0 produces a linear phase filter with equal pre- and post-ringing. A value of 0.0 produces a minimum phase filter, which offers no pre-ringing but has non-linear phase distortion, which can be objectionable. Setting intermediate values allows a continuous tradeoff between pre-ringing and post-ringing and allows users to linearize phase in the pass-band as much as possible.

Steepness: steepness of the transition band of the lowpass filter. Higher steepness will reject unwanted frequencies but cause more ringing in the time- domain and a higher CPU load

P.S. Thank you both, Alex and lavendier for the dialog here and special thanks to lavendier for opening the thread and being so helpful in general here. Very intriguing and educational discussion. Also, I am currently using Cutoff: 1, Preringing: 0, and Steepness: 3 and finding it enjoyable. Will do some more critical listening later.

Cheers.

Thanks for the quick reply Superdad - I'll monitor with interest and try out some of the settings being discussed.

Not that I'm asking for any suggestions now, but just to answer your questions - I use an Audiolab M-DAC, a range of music files but all apple lossless, mostly redbook, some 96 and 192kHz, the odd 88.2 (if that's what you mean?). Not sure what A+ settings are relevant, but I'm direct and integer mode enabled, not currently up sampling, not using 'optimising system for playback'.

Yes, you are correct that Ayre's BoBW graph has no pre-ringing, thus "0" in iZotope would be required. But how the heck do they get that little amount of post-ringing?

 

I think we need to know for sure how the iZotope steepness setting works, and how you are getting your results. I had understood the steepness number to be how many dB per octave of attenuation is applied. So to be -96dB at 44kHz, the steepness, assuming the low pass begins at 22kHz, would have to be set at 96. Is that not correct?

 

Also Levandier, can you please give me your exact settings that you have been using for the downsample step in RX2 Advanced? Without knowing those I am finding it impossible to duplicate your graph results (when replicating your upsample settings). And which of the test files you posted are you using for the graphs you are posting? (Float, Int., or 24) THANKS!

ALEX

I would love to know that too! The izotope help documentation on their web site doesn't really explain how the steepness works, so I just go off what the resample graph says (as you mouse over the graph it shows the frequency vs. attenuation).

For the graphs I posted before, I downsampled first using steepness 200, cutoff 1 and pre-ringing 1 (I assume that typical downsampling used in CD mastering uses a brickwall filter, but I could very well be wrong). If I then upsample, using steepness 3, cutoff 1 and preringing 0, the result doesn't look much like the Ayre graph because of the ringing introduced by downsampling; but if I downsample with say steepness 3, cutoff .9 and then upsample with the 3,1,0 settings, then it looks a lot more like the Ayre graph.

As for the files, I use the 24bit version. I would love to try 44.1khz test files to see more clearly the effect of just the upsampling filter in the absence of any pre-existing ringing, but I haven't been able to find any...

After playing around with Audacity for a while, I was able to make a 44.1khz pulse file. Here you go:

http://rapidshare.com/files/1349277383/Pulses_24%2044khz.wav

Trying out some settings similar to what Levandier suggested for the Ayre filter. This is pretty close... -6 db @ Nyquist

After playing around with Audacity for a while, I was able to make a 44.1khz pulse file. Here you go:

 

http://rapidshare.com/files/1349277383/Pulses_24%2044khz.wav

Very handy, thanks.

Yes, I think the big unknown is how the music files are treated on the mastering side. I don't know that it is fair (without further info, say from someone like Barry D.) for us to make assumptions about how the downsampling is done. Therefore I propose that, for the sake of consistency, we adopt (here in this thread anyhow) a standardized set of test tone files created at 16/44 to use in RX2 Advanced experiments.

There are a few free or trial tone generators for Mac which should allow for choice of bit-depth and sampling rate. Faber Acoustical's Signal Suite comes to mind, but there are many others.

Trying out some settings similar to what Levandier suggested for the Ayre filter. This is pretty close... -6 db @ Nyquist

 

[ATTACH=CONFIG]4650[/ATTACH]

 

[ATTACH=CONFIG]4651[/ATTACH]

Junker: Those look pretty good. Not sure how they will sound, but I will try them and report back.

However, your graph does seem to prove that iZotopes "Steepness" setting can NOT be interpreted to mean a slope of "X dB per octave." Your graph curve starts at about 16kHz and is down WAY more than 4dB at 32kHz (not even on the graph)!

Junker: Those look pretty good. Not sure how they will sound, but I will try them and report back.

 

However, your graph does seem to prove that iZotopes "Steepness" setting can NOT be interpreted to mean a slope of "X dB per octave." Your graph curve starts at about 16kHz and is down WAY more than 4dB at 32kHz (not even on the graph)!

You are very right!

At the end the final and definitive judgement will be our ear/brain system. Graphs are only a way (a guide) to measure it, but not the definitive one. I believe Ayre made a very good approach to the pre and post ringing problem (if you have it), but there is no way to adjust between Minimum and Linear phase in small increments, like we have now on A+ under iZotope SRC, since this setting are very user dependent: From your ears/brain taste, DAC, music gear and listening room frequency response, reflections, et al.

Roch

Cool. Definitely let us know what you think.

I'm a little uncomfortable with that shallow of a frequency curve ultra-sonic due to foldback distortion. IMO it sounded a little bright and brittle on the top-end...

So, I shot for -30db at Nyquist instead of -6 with: 13 slope / 0.91 cut-off / 0.30 pre-ring

Will also compare to -40db @ Nyquist with: 14 / 0.90 / 0.3

Wow. There really is more than one way to skin this cat!

Accepting a slight rolloff > 19KHz, I guess the main question is where you are comfortable ultra-sonic...

Looking forward very much to your filter tests!