The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Good dither practices, what are yours? Dynamics Plugins
Old 9th February 2017
  #271
Quote:
Originally Posted by pentagon View Post
If you are going to continue processing/mixing the signal at a higher sample rate until the final stereo file which ends up at 44.1, then there's a benefit at doing higher sample rates.
If you are going to capture with the Burl at 192k and then immediately downsample to 44.1 before continuing to process/mix, you are not doing a good thing. Better to let the Burl capture at 44.1 than have a "live" SRC to 44.1. Much to the confusion of many, many people, SRC done today is not a simple division if things happen to multiples. There is oversampling and filters applied (and resulting pre-ringing and cut-off artifacts.)
And all of this has nothing to do with dither unless for some reason you are not working at 24 bit at this point (and then possibly doing SRC at 32 bit float which also doesn't need dither since you are going up not down.)
Yeah I didn't originally understand that dither is only relevant when bit reduction happens, not sample rate, so thank you for clarifying that. As far as the sample rate goes, I guess I'll just use my ears. While I understand that theoretically it shouldn't matter since we don't hear past 20khz and the nyquist of 44.1 is 22khz, if my brain tells me it likes it better I may continue to do it this way. I mean in the end my ears are going to have the final say, regardless of how wrong I am scientifically. I'm gonna try a full mix this way and see if I run into any problems. Thanks for your input, it was very helpful.
Old 9th February 2017
  #272
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Obe1 View Post
While I understand that theoretically it shouldn't matter since we don't hear past 20khz and the nyquist of 44.1 is 22khz, if my brain tells me it likes it better I may continue to do it this way.
High sample rates used not for ultrasound playback. But for proper work of too gentle analog filters in ADC nad DAC.
Old 9th February 2017
  #273
Motown legend
 
Bob Olhsson's Avatar
 

Quote:
Originally Posted by pentagon View Post
Sample rate conversion is not truncation or bit-depth change. You are confusing the two. Dither is for bit-depth change.
Sample rate conversion causes a bit-depth change so I think the two still need to be considered together.
Old 9th February 2017
  #274
Lives for gear
 

Quote:
Originally Posted by Bob Olhsson View Post
Sample rate conversion causes a bit-depth change so I think the two still need to be considered together.
If you are going from 24 bit to 32 bit float in conversion, no dither is necessary since there is no bit truncation in that process (all 24 bits can be fully expressed as 32 bit float.)
If you are going from 24 bit to 24 bit, there will be no intermediary step where one would be less than 24 bit (even with oversampling, etc.) If there is any depth down-conversion/truncation in the process, dither would need to be applied by the sample rate converter and not the user because that step is outside the reach of the user. Dithering the initial signal has no relevance to when the dither actually needs to be applied.
Old 9th February 2017
  #275
SRC conversion routines typically include a very long FIR anti-alias / anti-imaging filter. There's a lot of math being done, and it's best to do it in extended precision or floating point. If the output format is shorter-length fixed-point, then dither is needed, and a well-defined SRC routine should include it. If the SRC output format is floating point, then dithering can be delayed until the project is exported to a fixed-point format.

David L Rick
Seventh String Recording
Old 9th February 2017
  #276
Gear Guru
 
Karloff70's Avatar
 

Quote:
Originally Posted by pentagon View Post
If you are going from 24 bit to 32 bit float in conversion, no dither is necessary since there is no bit truncation in that process (all 24 bits can be fully expressed as 32 bit float.)
If you are going from 24 bit to 24 bit, there will be no intermediary step where one would be less than 24 bit (even with oversampling, etc.) If there is any depth down-conversion/truncation in the process, dither would need to be applied by the sample rate converter and not the user because that step is outside the reach of the user. Dithering the initial signal has no relevance to when the dither actually needs to be applied.
Ok, been watching the thread and thought I pretty much had it down, but this one upsets my equilibrium. This would surely mean that putting a dither plug before an output would also be 'in the wrong place'? And if it doesn't, what does this actually imply?
Old 9th February 2017
  #277
Lives for gear
 

Quote:
Originally Posted by Karloff70 View Post
Ok, been watching the thread and thought I pretty much had it down, but this one upsets my equilibrium. This would surely mean that putting a dither plug before an output would also be 'in the wrong place'? And if it doesn't, what does this actually imply?
Because even if your session (using Pro Tools terms) is set at 24 bit, modern DAW engines operate at a higher bit depth internally (and most likely floating point) so when going to the output, bit depth reduction is happening.
Old 9th February 2017
  #278
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
Sample rate conversion causes a bit-depth change so I think the two still need to be considered together.
Yes. If hardware (including computer) capable to float point operations, need use it for resampling.
Old 9th February 2017
  #279
I'm not sure if the Apollo live SRC function works in floating point or 24bit. I guess I could email their tech dept.
Old 9th February 2017
  #280
Gear Guru
 
Karloff70's Avatar
 

Quote:
Originally Posted by pentagon View Post
Because even if your session (using Pro Tools terms) is set at 24 bit, modern DAW engines operate at a higher bit depth internally (and most likely floating point) so when going to the output, bit depth reduction is happening.
Sure, I get that, and that was part of my understanding built out of this thread. It was what you added on the end about dithering being applied by a sample rate converter that confused me.
Old 9th February 2017
  #281
Lives for gear
 

Quote:
Originally Posted by Karloff70 View Post
Sure, I get that, and that was part of my understanding built out of this thread. It was what you added on the end about dithering being applied by a sample rate converter that confused me.
You aren't doing sample rate conversion when going out your hardware.

Obe1 was asking about live src on incoming audio. Dithering being applied to audio by a sample rate converter (which could have upsampled and changed to a higher bit depth to process filtering prior to returning the lower bit depth output format) is what a good conversion process should do. That dithering is applied at the end of processing just before lowering bit depth. Applying dithering before going through SRC is not useful. The final dithering is the responsibility of the SRC designer since they are the only ones who can control that application.

But none of that has to do with outputting audio to hardware where no SRC is happening.
Old 9th February 2017
  #282
Quote:
Originally Posted by pentagon View Post
..The final dithering is the responsibility of the SRC designer since they are the only ones who can control that application.
I was curious about this.. I assumed if dither was necessary after a live SRC then the Apollo would naturally have it included in the process. But then again you would think dither is necessary when bouncing to disk in pro tools but from what I've heard is that is not automatically applied. Out of curiosity I emailed UAD to get some more details on the live SRC process.
Old 10th February 2017
  #283
Gear Guru
 
Karloff70's Avatar
 

Quote:
Originally Posted by pentagon View Post
You aren't doing sample rate conversion when going out your hardware.

Obe1 was asking about live src on incoming audio. Dithering being applied to audio by a sample rate converter (which could have upsampled and changed to a higher bit depth to process filtering prior to returning the lower bit depth output format) is what a good conversion process should do. That dithering is applied at the end of processing just before lowering bit depth. Applying dithering before going through SRC is not useful. The final dithering is the responsibility of the SRC designer since they are the only ones who can control that application.

But none of that has to do with outputting audio to hardware where no SRC is happening.
Thank you! Consider me unconfused again.
Old 10th February 2017
  #284
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Obe1 View Post
I'm not sure if the Apollo live SRC function works in floating point or 24bit. I guess I could email their tech dept.
Some time ago I discussed with one my friend (hardware developer) application of several my algorithms (done in 64-bit float) at FPGA for realtime work, but he say me that it is too complicated, even for modern chips.

Probably, new one cristall computers can process 64-bit float.

Anyway need serious adoptation of a PC algorithms to limited resources of hardware decissions.

If see to level noise of a some hardware devices, looks like, they perform integer math. But it is general my understanding of the issue. Of course, I don't have full information about DSP devices (including Apollo).

For float point math I expect own SRC level of noise about -170 ... -200 dB.
Old 10th February 2017
  #285
Here for the gear
 

2 Questions:

Should i ask a customer to apply dither when exporting 24bit pre-master files for mastering?
After the mastering processing is completed and i want to export a 24bit master file, will i apply dither again?
Old 10th February 2017
  #286
Motown legend
 
Bob Olhsson's Avatar
 

Quote:
Originally Posted by pentagon View Post
If you are going from 24 bit to 32 bit float in conversion, no dither is necessary since there is no bit truncation in that process ...
This is a misunderstanding. Bit truncation happens as part of all signal processing including resampling because the number of bits is expanded to infinity. Input bits is utterly meaningless. Truncation happens when moving to 32 bit float too. It's just minimized but not prevented. There's no simple way to dither the truncation.
Old 10th February 2017
  #287
Lives for gear
 

Quote:
Originally Posted by Bob Olhsson View Post
because the number of bits is expanded to infinity.
At no point in digital resampling are "bits expanded to infinity." That would create an overflow.
Quote:
Originally Posted by Bob Olhsson View Post
Truncation happens when moving to 32 bit float too.
There are no 24 bit fixed point numbers that cannot be expressed 100% as a 32 bit float point number.
Old 11th February 2017
  #288
This was the response from UAD when I asked them how their live SRC worked:

"The SR Convert mode will convert the incoming SPDIF digital audio to whatever Sample Rate the DAW session or Console is set to (if there is no DAW Session loaded).

The Bit Depth will of the resulting audio files recorded into your DAW will be determined by the setting you choose within the DAW preferences.

As I understand it, you won't need to load a dither plug-in in Pro Tools to get good results.

I hope I communicated all of that well. Let me know if you have any additional questions.
"

I don't know if that explained anything or not.
Old 11th February 2017
  #289
Airwindows
 
chrisj's Avatar
Quote:
Originally Posted by pentagon View Post
At no point in digital resampling are "bits expanded to infinity." That would create an overflow.
There are no 24 bit fixed point numbers that cannot be expressed 100% as a 32 bit float point number.
True as far as it goes, but it typically goes into a DAW that is either 32 bit or 64 bit floating point buss.

At that point, ANY MATH operation other than a very limited set (like 6.02 dB gain boosts or cuts?) will generate values that must truncate when rendered as a 32 bit floating point number.

This applies even when you're starting with a 32 bit float, and multiplying by say 1.001. You will truncate the mantissa, though the exponent will give you nice resolution down into the quietest details… but they are truncating too, just at their respective levels of detail. Almost anything you do is going to truncate the mantissa… again, other than certain very specific math activities.

And yes, it's an overflow. That's what working with floating point math is like \

Bob doesn't mean the concept of infinity as a number, he means there are simple multiplies that will give you a decimal expansion that MUST be cut off somewhere, no matter how big your output word. For instance, 1 * (1/3) has an infinite expansion, 1 * (1/2) doesn't.
Old 11th February 2017
  #290
Lives for gear
 
Hyder boy's Avatar
 

So should I dither my hardware inserts in pro tools or NOT! LOL


But seriously, should I?
Old 11th February 2017
  #291
Quote:
Originally Posted by Hyder boy View Post
So should I dither my hardware inserts in pro tools or NOT! LOL


But seriously, should I?
I think that's covered earlier in the thread. You'll find it somewhere in these 10 pages.
Old 11th February 2017
  #292
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by bojobojo View Post
2 Questions:

Should i ask a customer to apply dither when exporting 24bit pre-master files for mastering?
After the mastering processing is completed and i want to export a 24bit master file, will i apply dither again?
If mastering in 24 bit and above (including float point formats), applying of dithering have no practical sense. Because rounding distortions is too low or absent (depend on input and output bit-depth).

In general, applying dither have practical sense for conversion to 16 bit only. It is work as smoothing nonlinear distortions during rounding (truncation).

For some cases it may be applied for conversion from 16 bit. In this case dither's noise can mask distortions, if they still present in source.
Old 11th February 2017
  #293
Lives for gear
 
Hyder boy's Avatar
 

Quote:
Originally Posted by Obe1 View Post
I think that's covered earlier in the thread. You'll find it somewhere in these 10 pages.
Yea, I know it has. But with all the conflicting opinions, I wish one of these brainiacs could dumb it down for us mortals.
Old 11th February 2017
  #294
Motown legend
 
Bob Olhsson's Avatar
 

It's trivial to try it both ways and listen! I did and decided it was better to dither. Then I found a few developers who hadn't coded their dither properly and luckily they fixed their code.

Theoretically it should be dithered but that doesn't mean you can't get away with not dithering. Still why bank on "getting away" with not dithering when properly coded 24 bit dither should be inaudible while the distortion from not dithering may or may not be? That's just being lazy.
Old 12th February 2017
  #295
Gear Maniac
 
Yuri Korzunov's Avatar
 

Quote:
Originally Posted by Bob Olhsson View Post
Theoretically it should be dithered but that doesn't mean you can't get away with not dithering. Still why bank on "getting away" with not dithering when properly coded 24 bit dither should be inaudible while the distortion from not dithering may or may not be? That's just being lazy.
If use 24 bit as target format, dithering have no sense.

Because level noise of 24 bit about -140 .. -145 dB. Level noise of a good DAC about -120 dB.
Old 12th February 2017
  #296
Lives for gear
 
JulenJVM's Avatar
Quote:
Originally Posted by Yuri Korzunov View Post
If use 24 bit as target format, dithering have no sense.

Because level noise of 24 bit about -140 .. -145 dB. Level noise of a good DAC about -120 dB.
Of course this distortion is not audible, but when I apply dither, I hear new things on my mixes, related mainly to spatial placement of instruments. Truncation distortion seems to take away small details from the music, that dither helps bringing back. That is also subjective, but my own experience.

Two questions, please. Are you able to hear DC Offset? Do you take measures to remove it?
Old 12th February 2017
  #297
Lives for gear
 
edva's Avatar
Quote:
Originally Posted by JulenJVM View Post
Of course this distortion is not audible, but when I apply dither, I hear new things on my mixes, related mainly to spatial placement of instruments. Truncation distortion seems to take away small details from the music, that dither helps bringing back. That is also subjective, but my own experience.

Two questions, please. Are you able to hear DC Offset? Do you take measures to remove it?
I do not disagree with your statement about dithering. I dither too.
But, those questions are somewhat misleading as regards dither.
To state the obvious, DC Offset filtering, and what is being filtered, have a completely different effect on audio than does dither, or what it is "filtering". IMHO.
Old 12th February 2017
  #298
Lives for gear
 
JulenJVM's Avatar
Quote:
Originally Posted by edva View Post
I do not disagree with your statement about dithering. I dither too.
But, those questions are somewhat misleading as regards dither.
To state the obvious, DC Offset filtering, and what is being filtered, have a completely different effect on audio than does dither, or what it is "filtering". IMHO.
The point I wanted to make was not about the effect of the process, but rather about dismissing it because the problem is not audible. I wanted to point out that we deal with issues outside the hearing range, because we know the theory behind it, so why treat dither differently?
Old 12th February 2017
  #299
Lives for gear
 

Offset is audible (in, for instance, editing an audio file)
Old 12th February 2017
  #300
Lives for gear
 
JP__'s Avatar
 

But a DC filter is audible too. Its all a comporomise. Same with dither. After 10 pages the thread repeated again and again. Bob makes all clear already. Just make some test and use your ears.
You could just consider the topic solo and argue with the hearing threshold or you can see it as one single part of a chain of processes. Could make a significant difference in a real world scenario. Its like theory vs practise, like so often.
If you couldnt perceive the influence of 24Bit dither in a real practical scenario I just would question your monitoring (or maybe your converter/DAW/soundcard is doing it right already). Even the influence of different dithers is clearly audible in a real world scenario.
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump