SRC and Dithering settings (in general)*

We have often discussed dithering in this forum and by now I guess that most everybody is up-to-date on what are the best dithering algorithms out there. But there is something that has not been discussed previously here and that, upon reading up on dithering, I find of utmost importance: the settings used.

As I see it now, dithering settings can mean the difference between a normal to bad job and a great one. For example, I imagine that given Barry Diament's sensitivity to phase distorsion due to delays in the different frequencies and the havoc they can create he might be inclined to use linear phase (http://en.wikipedia.org/wiki/Linear_phase) settings as his prefered dithering choice.

What say you?

 

Hi Metoo,

"Sensitivity to phase distortion due to delays in the different frequencies" is only one factor of many. In fact, there are lots of instances where I much prefer the steeper settings.

Goes back to what I've said many times: The only rule is there are no rules.
There are no simple recipes or formulas and in fact I would say looking for these is guaranteed to take you further from (not closer to) your goal.

For dither settings, I suggest comparing results of different approaches with the undithered original. Listen for timbral differences on instruments and voices and listen for spatial differences, particularly how well (or not) the size and dimensions of the soundstage are preserved.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Thanks for your input, Barry. As it often is the case with your posts, it provides great insight.

I have tried several different dither settings lately and have arrived at some conclusions, but have not yet recurred to direct A/B comparisons of the original to the dithered results, but to my memory of previous dithering instances of the same content.

I have sometimes used very steep personally created settings, but while they tend to result in higher frequencies reproduced in the final file, I currently have certain doubts as to their complete lack of distortion.

I will certainly do some testing with the criteria you have pointed out.

Again, thanks Barry.

 

Hi Metoo,

With MBIT+, I very often find the default settings (Normal intensity) with "Ultra" for the curve, to be just what I want.

But in my view every separate situation merits that comparison against the undithered original.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

You are right, Barry.

I just realized that I made a mistake in my original post in this thread, something I can trace back to having slept very little last night. Sorry.

I was really referring to the sample rate conversion process, not the dithering per se. This is why I mentioned steep and phase distorsion instead of a noise scheme such as MBIT+ or triangular. My question must have originally misled you because you replied in the same sense.

I just asked a gort to please change the name of this thread to "SRC and Dithering settings (in general)" to better reflect the discussion.

 

Hi Metoo,

I just saw the title change. Well, I'd still re-state the "rule" about there being no rules.
I'd also still suggest comparison with the unconverted original as a means of assessing how well the SRC worked.

The defaults in iZotope's 64-bit SRC are a good starting point, especially with the Quality slider maximized. But... are you using their SRC? I thought you are running Windows? (There IS a 32-bit version in RX Advanced. Don't know why their web site calls it "64" unless they mean the same algorithm is incorporated into a 32-bit plug-in.)

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com


Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

This is how I do it!

But, for right now, I use the lone dither setting in RX (the cheaper version).

I am happy with using SoX, with the steep phase setting, for the SRC, and it sounds quite transparent, but it also somehow exposes the deficiencies in my dither afterward. Either that, or I am setting my expectations unrealistically high for going from 96k/32-bit to redbook. But, if I just dither a 44.1/32-bit to redbook, it is very transparent. If I just SRC from 96k to 44.1 it is very transparent. Following me here?

 

Hi Grant,

I'm following... I think.
Two processes are not as transparent as one.

Have you tried this using different hosts?
It is difficult to determine is the issue is SRC exposing flaws in the dither, dither exposing flaws in the SRC, the host itself causing an alteration in the Save or a combination of these.

You are doing the dither last, right? With absolutely NO changes after dithering?

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

There are several SRCs that use the linear phase setting, SoX comes to mind, as do R8brain Pro and iZotope's SRC. What I see in the graphs I have checked out at http://src.infinitewave.ca/ is that intermediate phase doesn't seem like a good choice in several of them.

IIRC, Alexey said here that RX Advanced is not 64-bit, but rather 32 bits, which is why the noise floor is slightly higher as you can see in the 1kHz tone graph.

On another note, I am currently also using the SoX plugin for Foobar with the following settings: Steep Filter, Linear Phase, Very High quality, not allowing aliasing... Sweet.

 

Again, as you say, it depends.

I have noticed nice results using a no alias option for downsampling my laserdrops from SACD from 192/24 to 96/24 as part of my testing (the frequency fall curve tends to do away with most of more notable the high frequency noise around 48 kHz as per the frequency graph in Audition) and then use steep linear phase to take the resulting file down to 44.1/16.

For now, I am experimenting with SoX that seems quite close, if not the same quality, as iZotope's SRC and is free.

 

Yes, in fact as part of my confusion yesterday this is exactly what I undertook with the SRC after reading your post.

 

Hi Metoo,

At this point in time, unless and until I hear something that outperforms the combination, I have no complaints at all after using the 64-bit SRC and MBIT+ for dither/noise shaping.

The results compare with the 24/96 (or 24/192) original quite well, considering the limitations of 16/44. I've not yet heard better outside of leaving the file at 24/96 (or 24/192), which of course, won't work when CD is the target.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Believe me, were I able to use the iZotope SRC it would be my main choice, but I don't have the US$1,100 to spend on it. This is why I have mentioned SoX as a feasible and free alternative.

This is why I created this generic thread so that we can share information about settings that perhaps can be applied or tested in other algorithms to see if they improve results.

This said, generally I downsample my 192/24 files solely to 96/24, unless - for example - I want a copy on CD for the car. This is where my mention of the 16/44.1 files enters the picture.

As to the 96/24 files derived from 192/24 ones, they seem to keep most of the soundstage and a lot of the timbre and details of the original. When going down from 24 to 16 bits I notice, as always, a 'flattening' of the sound, which is more or less apparent depending on the source.

 

Correct. I know better! Always dither last.
I'm wondering if it is due to the way the two different hosts are handling the data. Both programs are of high quality.

 

That's exactly what I use now. I love the results.

I no longer use Audition for the conversion process. It is woefully lacking! Like Sound Forge, they haven't kept up with the curve. I'll have to jump on them about it!

 

Hi Grant,

In my experience, if the hosts are clean, the results will be.
If the results between two hosts are not clean, one or both of those hosts is questionable.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Yes, plus SoX is cross-platform.

The only thing missing in SoX, at least for the moment, is a GUI.

 

Well the Foobar component actually serves as a kind of GUI for the SoX routines. I haven't compared SoX itself to the Foobar implementation, but I assume it should be the same.

 

How can SoX results come out clean, the RX dither come out clean, but both results together don't? Either that, or, again, I set my expectations unrealistically high for the conversion results.

 

Why would someone go through all that trouble to write such great code, but not the GUI? Some company needs to jump on it!

 

Hi Grant,

Are either of these standalone or do they run in a host?
If a host is used, that is one place to address the question.

Personally, I'm not so sure about SoX either.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Well, if so, how do I go about downsampling a file with it? I am able to play back files at different resolutions, but haven't found a way to resample any from inside Foobar.

 

Grant, what SRC settings are you using?

 

RX, as you know, is standalone. That is what I use for the dither only.

SoX is only used for the SRC. The host is Foobar. But, again, both results alone sound very transparent.

I am talking about extremely tiny differences between the original and the final redbook result. maybe i'm too anal for wanting a virtually perfect match?

 

Right-click the file in Foobar. That brings up a whole new world of options.

From what I can tell, Foobar was originally created by computer geeks who aren't that anal about music. That's why the GUI is simple. But, it is very light and fast. Maybe some seriously music-oriented people need to chip in.