exporting 44.1 mixes when recordings were at double or quad rates

Page: < 12 Showing page 2 of 2
Author
Post
Essentials Only Full Version
gswitz
Max Output Level: -18.5 dBFS
  • Total Posts : 5694
  • Joined: 2007/06/16 07:17:14
  • Location: Richmond Virginia USA
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/06 22:42:34 (permalink)
Nika to me
I agree with you.
 
When taking a high sample rate project and doing a sample rate conversion/bounce to disk to a lower bit depth, where no further processing will take place, a noise-shaped dither algorithm, such as POW-r, is best.


StudioCat > I use Windows 10 and Sonar Platinum. I have a touch screen.
I make some videos. This one shows how to do a physical loopback on the RME UCX to get many more equalizer nodes.
#31
John
Forum Host
  • Total Posts : 30467
  • Joined: 2003/11/06 11:53:17
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/06 22:56:14 (permalink)
gswitz
Nika to me
I agree with you. When taking a high sample rate project and doing a sample rate conversion/bounce to disk to a lower bit depth, where no further processing will take place, a noise-shaped dither algorithm, such as POW-r, is best.


That is a change in the parameters. The dither is due to bit depth reduction not because of the sample rate change.  Its a little disingenuous to add this when before I was talking only about sample rates. 

Best
John
#32
gswitz
Max Output Level: -18.5 dBFS
  • Total Posts : 5694
  • Joined: 2007/06/16 07:17:14
  • Location: Richmond Virginia USA
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/06 23:36:33 (permalink)
John,
 
My reading of what Nika said was that when doing a sample rate conversion that will end at the same bit depth that it started at, a filter is required if you are reducing the sample rate (even if you are going from 88.2 to 44.1 because you have to filter out all the frequencies above the new Nyquist Freq). So, most programs will up-sample to the lowest common denominator Sample Rate and apply 64 bit processing when applying the filter. Then the Dither will be applied when getting back to the original bit depth.
 
So, this means that dither will be applied when not intentionally changing the bit depth, but only changing the sample rate down. If you were to change the sample rate UP then dither would not be required.
 
You may disagree, but this is my understanding of it.
 
Remember, years ago we had a long discussion about this on someone else's thread where I posted all kinds of quotes from the book trying to defend the position. I didn't understand it then as I think I do now.
 
One of the important take-aways for me from this is that avoiding Pow-r algorithms when bouncing or mixing internally is a good idea when original sample rates are 44.1 or 48 and Pow-r noise shaping will fall in an audible range. You don't want to build on this focused noise. So use Rectangular or Triangular in those cases. Then use Pow-r when exporting. If you are using Double or Quad rates, use Pow-r for both internal bouncing and exporting because the focused noise will not fall in the audible range when using double or quad rates and will be filtered out all-together when exporting to 44.1.
 

StudioCat > I use Windows 10 and Sonar Platinum. I have a touch screen.
I make some videos. This one shows how to do a physical loopback on the RME UCX to get many more equalizer nodes.
#33
John
Forum Host
  • Total Posts : 30467
  • Joined: 2003/11/06 11:53:17
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/06 23:59:36 (permalink)
If you use a filter to cut frequencies why would one need to dither. Also those frequencies are beyond normal hearing anyway. It makes no sense to me. 

Best
John
#34
gswitz
Max Output Level: -18.5 dBFS
  • Total Posts : 5694
  • Joined: 2007/06/16 07:17:14
  • Location: Richmond Virginia USA
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/07 11:52:41 (permalink)
John, frequencies beyond the nyquist frequency need to be filtered out or they will cause aliasing in the audible range. When you reduce sample frequency, you also reduce the nyquist frequency. To avoid aliasing you must apply a filter before doing this. Filters are better if you upsample. Then Dithering is applied when you down sample again.

Nika advises to avoid shaped dither in a all bounces except final export to avoid stacking shaped dither into the audible in his book. But when mixing at double or quad rates, the stacked dither piles up above the audible range and will be removed by the sample rate conversion filter at the time of export, so using shaped dither for bounces before the final export makes sense in these cases.
post edited by gswitz - 2014/01/07 16:03:59

StudioCat > I use Windows 10 and Sonar Platinum. I have a touch screen.
I make some videos. This one shows how to do a physical loopback on the RME UCX to get many more equalizer nodes.
#35
gswitz
Max Output Level: -18.5 dBFS
  • Total Posts : 5694
  • Joined: 2007/06/16 07:17:14
  • Location: Richmond Virginia USA
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/07 23:07:08 (permalink)
John
Its a little disingenuous to add this when before I was talking only about sample rates.

I am definitely not being disingenuous, John - not intentionally, and I think not at all. I'm trying hard to understand the details of the sample-rate stuff, SRC, and dither.
 
I brought it up b/c I could see the higher frequencies in the analyst I was using.
 
Nika was nice enough to chime in. Frankly, I think I may still have misunderstood him. At one point I thought I understood him to say that at double rates to use Pow-r for internal bouncing, but re-reading I think I just misunderstood that. He clearly is talking about exporting the final bounce. And he does say to use POW-r.
 
I think the order of events would matter a lot.
Here's a guess at order of events...
- upsample to lowest common denominator for sample rates
- filter out frequencies above the new nyquist frequency - increasing bit depth to 64 float
- reduce sample rate (now safe to do because there is no information in the wave that can't be represented within the new sample rate)
- apply dither
- truncate bits from 64 to 24 (or other)
 
John
I don't know about dithering for sample rate reduction. I believe there is no artifacts due to truncation when changing the sample rate. I could be wrong though. 

I think there can be aliasing if you change the sample frequency for audio that contains sound information above the new nyquist frequency. By this, I mean that if you have a 88.2 stereo wave with reverb tails that go above the human audible range say to 32kHz, if you down-sample without filtering out the content above 22.05 kHz you will get aliasing in the audible range as the DA convertor tries to recreate the wave form from the samples.
 
John
Heck plugins upsample and downsample. Its called oversampling and I doubt dither is used for that. All high sample rates do is increase bandwidth.

I'm sure I don't know. It appears Nika thinks that oversampling does justify dither. The linked Wikipedia article on Oversampling also mentions dither.
 
So, whether it's worth making adjustments to your dither settings in Sonar as you work your way through a project seems like a fair question. Personally, I rarely make changes and usually leave it on Pow-r 3. If I should change it to Rectangular or to None at various times, I need it spelled out. The only time I leave it at none is when exporting a 24 bit 44.1 recording at 24 bit 44.1. All other cases I leave it on.
 
It is clear in Nika's book that for internal bouncing, he recommends non-shaped dither like rectangular or triangular and for exporting Pow-r. For a second I thought he suggested Pow-r for internal bouncing when the sample rate was double or quad, but I don't think he actually said that... I think I just heard it. :-)
 
The Wikipedia article on Sample Rate Conversion (SRC) gives the same reason for possibly using dither or noise shaping algs when doing SRC.
 
bitflipper
That's the first I've heard of dither being required for (or having anything to do with) SRC, but it's been quite awhile since I read Nika's book. If you can cite a page from the book, I'd like to read what he has to say about that.

Man, I just can't find it. I've been flipping and flipping through the pages. I'm sure I've quoted it in a Thread with John back in 2008 or so, but I can't find the thread and I can't find the quote now. You can see in this thread what Nika Aldrich says about it in an email to me, but I can't find the quote in the book. Sorry.
 
bitflipper
Of course, the question is moot in most cases, because downsampling during export is usually done in conjunction with wordlength reduction, which is where dither comes into the picture. That may be the source of confusion.

 
So, it's possible that every time Sonar uses the 64 bit audio engine it applies a anti-aliasing filter then dither then truncates to the required bit depth when bouncing to tracks, assuming you aren't using 64 bit as your render bit depth size. If it does apply dither, does it apply your default dither set in Preferences > Audio > Playback and Recording? I have no idea. It could just truncate, but I would think that could have bad aliasing issues into the audible ranges.
post edited by gswitz - 2014/01/08 00:26:10

StudioCat > I use Windows 10 and Sonar Platinum. I have a touch screen.
I make some videos. This one shows how to do a physical loopback on the RME UCX to get many more equalizer nodes.
#36
drewfx1
Max Output Level: -9.5 dBFS
  • Total Posts : 6585
  • Joined: 2008/08/04 16:19:11
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/08 12:36:10 (permalink)
A few (possibly overly technical) clarifications:
gswitzHere's a guess at order of events...
- upsample to lowest common denominator for sample rates
- filter out frequencies above the new nyquist frequency - increasing bit depth to 64 float

When doing an SRC, you filter out everything above the lowest Nyquist frequency involved in the SRC.
 

- reduce sample rate (now safe to do because there is no information in the wave that can't be represented within the new sample rate)
- apply dither
- truncate bits from 64 to 24 (or other)

This is correct, but it's more complicated if noise shaping is involved.

I'm sure I don't know. It appears Nika thinks that oversampling does justify dither. The linked Wikipedia article on Oversampling also mentions dither.

 
I haven't seen anything that states or implies oversampling somehow in itself "justifies" dither. 
 

The Wikipedia article on Sample Rate Conversion (SRC) gives the same reason for possibly using dither or noise shaping algs when doing SRC.

What it says is that if SRC computations are performed at a higher bit depth, dither may be used to reduce to the original bit depth.
 

 So, it's possible that every time Sonar uses the 64 bit audio engine it applies a anti-aliasing filter then dither then truncates to the required bit depth when bouncing to tracks, assuming you aren't using 64 bit as your render bit depth size. If it does apply dither, does it apply your default dither set in Preferences > Audio > Playback and Recording? I have no idea. It could just truncate, but I would think that could have bad aliasing issues into the audible ranges.




Aliasing is from frequencies above Nyquist. It has nothing to do with bit depth.
 
Truncation/dither/noise_shaping have to do with bit depth reduction and have to do with quantization error (AKA quantization noise AKA quantization distortion).
 
But for bit depth reductions to 24bit (or higher), quantization error will never ever be audible in the real world. So it's completely academic what type of dither you use or if you use any at all.
 
Much of the confusion is because people confuse general "best practices" that you should always use dither when reducing bit depth (and what type of dither to use where) with the reality that at higher bit depths the quantization error is both buried in other noise and far below your threshold of hearing anyway.

 In order, then, to discover the limit of deepest tones, it is necessary not only to produce very violent agitations in the air but to give these the form of simple pendular vibrations. - Hermann von Helmholtz, predicting the role of the electric bassist in 1877.
#37
gswitz
Max Output Level: -18.5 dBFS
  • Total Posts : 5694
  • Joined: 2007/06/16 07:17:14
  • Location: Richmond Virginia USA
  • Status: offline
Re: exporting 44.1 mixes when recordings were at double or quad rates 2014/01/08 20:44:16 (permalink)
All good points, Drewfx1.
 
And a logical conclusion might be that applying dither when reducing from 64 bit float back to 24 bit would be a useless waste of processing if will never be heard anyway.

StudioCat > I use Windows 10 and Sonar Platinum. I have a touch screen.
I make some videos. This one shows how to do a physical loopback on the RME UCX to get many more equalizer nodes.
#38
Page: < 12 Showing page 2 of 2
Jump to:
© 2025 APG vNext Commercial Version 5.1