The Acoustic Guitar Forum

Go Back   The Acoustic Guitar Forum > General Acoustic Guitar and Amplification Discussion > RECORD

Reply
 
Thread Tools
  #1  
Old 04-01-2010, 07:25 AM
rick-slo's Avatar
rick-slo rick-slo is offline
Charter Member
 
Join Date: Nov 2004
Location: San Luis Obispo, CA
Posts: 17,171
Default Dithering about dither

With my recording setup with its given noise floor and for my purposes of recording solo guitar music I am wondering if dither is helping in any audible way.

Below is a test clip of a single note stroke to decay (no fader) both without and with dither.

Recorded mono in 24 bit: Gefell M295 - NPNG preamp - Mytek Stereo 96 AD

It was then reduced to 16 bit with one part dithered (Sonoris TPDF/curve 2/16bit) and the other part not dithered.

I am having a hard time hearing any effect of dithering and I wonder if I should bother.

http://dcoombsguitar.com/Guitar%20Music/Other/DitherTest.wav
__________________
Derek Coombs
Youtube -> Website -> Music -> Tabs
Guitars by Mark Blanchard, Albert&Mueller, Paul Woolson, Collings, Composite Acoustics, and Derek Coombs

"Reality is that which when you stop believing in it, doesn't go away."

Woods hands pick by eye and ear
Made to one with pride and love
To be that we hold so dear
A voice from heavens above
Reply With Quote
  #2  
Old 04-01-2010, 09:37 AM
Pokiehat Pokiehat is offline
Registered User
 
Join Date: Jun 2009
Posts: 181
Default

Dither is meant to be very very low in level because quantization noise in a 16 or 24 bit recording manifests very very low in level and the purpose of dither is to decorelate/randomise quantization error.

It is possible to hear dither noise if you:

1) amplify it until its loud enough to hear distinctly at ideal listening levels which defeats the purpose of the dither.

2) listen to the dithered audio at very very high amplitude (dangerous) and observe fades to (perceived) silence where the signal passes under the noisefloor threshold. In a modern 16 to 24 bit recording this is going to be 80 to 100dB down from full scale. Most spectrum analysers don't even go down that far just in case you are interested in trying to 'see' it with SPAN or PAZ analyser via generating and dithering sine test tones.

Furthermore it is possible to hear quantisation noise if you keep on throwing away more bits (i.e. with a bitcrusher). At normal listening levels it really doesn't become a problem until you are have 12 bits or lower at which point you can hear it very distinctly. Keep on throwing away more bits and the recording turns into a harsh, highly regular type of distortion. The function of dither works best when applied to situations where you are throwing away alot of bits. Like 24 bit down to 8 bit but since that hardly ever happens you are unlikely to see dither used in audio like it has been used in images where it is more common to see low bitdepth images in use up until comparitively recently. Dither actually makes image files bigger but I think the trade off was more significant in the pre broadband days where you had to mark forum threads 'no 56k' because loads of jpgs back then would just kill a modem user's net connection.

I wouldn't concern yourself too much with dither for the following reasons:

a) Bad dither and bad noise shaping is worse than no dither at all.

b) Working with analogue sources converted to digital, you are not going to get signal to noise ratio in excess of 80dB. The magnitude of quantisation noise by truncating 8 bits from a 24 bit recording is going to be dwarfed by noisy electrical components, spill and ambient noise picked up by the mic etc.

c) The effect is so small that if you are worried about noisy recordings you are far better off trying to fix the source of the noise. This ties in with (b) in the sense that there is nearly always more significant sources of undesirable noise than quantization error.

(d) It is best left to a mastering engineer anyway who won't get it wrong and probably has access to more and more appropriate noise shaping algorythms than you do.

Last edited by Pokiehat; 04-01-2010 at 09:51 AM.
Reply With Quote
  #3  
Old 04-01-2010, 11:57 AM
rick-slo's Avatar
rick-slo rick-slo is offline
Charter Member
 
Join Date: Nov 2004
Location: San Luis Obispo, CA
Posts: 17,171
Default

Thanks Pokiehat. That is generally how I see it and more importantly hear it but of course most people are recording analog (mic'ing stuff) and pointing out bad things if you do not dither going from 24 to 16. I just have not been able to hear the difference at least with the noise floor I have. Also I too have felt that certain dither software made the sound worse it anything.
__________________
Derek Coombs
Youtube -> Website -> Music -> Tabs
Guitars by Mark Blanchard, Albert&Mueller, Paul Woolson, Collings, Composite Acoustics, and Derek Coombs

"Reality is that which when you stop believing in it, doesn't go away."

Woods hands pick by eye and ear
Made to one with pride and love
To be that we hold so dear
A voice from heavens above
Reply With Quote
  #4  
Old 04-01-2010, 02:17 PM
Fran Guidry Fran Guidry is offline
Registered User
 
Join Date: May 2002
Location: Walnut Creek, CA
Posts: 3,712
Default

Quote:
Originally Posted by rick-slo View Post
With my recording setup with its given noise floor and for my purposes of recording solo guitar music I am wondering if dither is helping in any audible way.

Below is a test clip of a single note stroke to decay (no fader) both without and with dither.

Recorded mono in 24 bit: Gefell M295 - NPNG preamp - Mytek Stereo 96 AD

It was then reduced to 16 bit with one part dithered (Sonoris TPDF/curve 2/16bit) and the other part not dithered.

I am having a hard time hearing any effect of dithering and I wonder if I should bother.

http://dcoombsguitar.com/Guitar%20Music/Other/DitherTest.wav
Rick, there's a much more powerful tool for this, it's called double blind ABX, and it's implemented for free in foobar2000. You can very easily compare the two files note by note and with about 15 tests you'll have a very solid statistical result.

Since the effects of dither between 24 bit and 16 bit is somewhere around -90dBFS I'm reasonably certain that you won't be able to detect a difference under anything near normal listening conditions.

Even a well constructed professional studio will have trouble keeping noise levels down to the point that truncation vs dither is audible.

This hasn't stopped many folks from making pretty wild assertions about the dramatic difference they perceive in sighted evaluations, of course.

Fran
__________________
E ho`okani pila kakou ma Kaleponi
Slack Key in California - www.kaleponi.com
My YouTube clips
The Homebrewed Music Blog
Reply With Quote
  #5  
Old 04-01-2010, 02:41 PM
rick-slo's Avatar
rick-slo rick-slo is offline
Charter Member
 
Join Date: Nov 2004
Location: San Luis Obispo, CA
Posts: 17,171
Default

Quote:
Originally Posted by Fran Guidry View Post
Rick, there's a much more powerful tool for this, it's called double blind ABX, and it's implemented for free in foobar2000. You can very easily compare the two files note by note and with about 15 tests you'll have a very solid statistical result.

Since the effects of dither between 24 bit and 16 bit is somewhere around -90dBFS I'm reasonably certain that you won't be able to detect a difference under anything near normal listening conditions.

Even a well constructed professional studio will have trouble keeping noise levels down to the point that truncation vs dither is audible.

This hasn't stopped many folks from making pretty wild assertions about the dramatic difference they perceive in sighted evaluations, of course.

Fran
Agreed Fran. Going from 24 to 16 you lose 48db of dynamic range so those bottom 48db plus a margin of 1 bit (another 6db) brings you to that -90db. My noise floor is not that low.
I do not record with peaking at the full scale of course. The peaks usually reach up to the area of -8 to -10 so I guess you could say the noise floor should be above say -80db but
that is still well below the noise floor I get.
__________________
Derek Coombs
Youtube -> Website -> Music -> Tabs
Guitars by Mark Blanchard, Albert&Mueller, Paul Woolson, Collings, Composite Acoustics, and Derek Coombs

"Reality is that which when you stop believing in it, doesn't go away."

Woods hands pick by eye and ear
Made to one with pride and love
To be that we hold so dear
A voice from heavens above

Last edited by rick-slo; 04-01-2010 at 04:28 PM.
Reply With Quote
  #6  
Old 04-01-2010, 04:48 PM
Pokiehat Pokiehat is offline
Registered User
 
Join Date: Jun 2009
Posts: 181
Default

You don't lose 48dB of dynamic range. You are never going to get 24 noise free bits even if you have a 24 bit ad converter. The least significant of those bits are just going to be noise. Your mic and mic pre are going to make sure of that.

Also it is possible for the human ear to resolve sound beneath the quietest part of the signal but I guess the point is that 100dB down, the human ear is unlikely to detect any sound whatsoever under normal listening conditions let alone be able to resolve signal from noise at this kind of level. So it all becomes very theoretical and there is a tendency for people to rely on spectrograms where quantisation error can be highly visible provided the y axis extends down far enough and where the program material has very narrow bandwidth. It is most visible using sine wave test tones since the sine wave has no harmonic content. Dither in this case has a highly visible effect in decorrelating the error inherent to the quantisation process but the problem is that under normal listening conditions, the audible artifacts of that error are basically on the threshold of inaudibility anyway. So sometimes it is easy to confuse the magnitude of what we see in spectrograms versus what we can actually hear in the programme material.

I don't want to say that dither is redundant or pointless because its not. Its just something that home recordists shouldn't concern themselves with because in all likelihood they have much bigger problems to be concerned with that have a much more significant affect on the quality of their recordings.

I'm talking things like proper micing technique i.e. elimination of ambient noise and spill using directional mics at the appropriate distance and orientation, elimination of common electrical problems which can produce undesirable and audible noise at normal listening levels i.e. ground loops etc. For the majority of people these are things which are going to result in a bigger improvement in your recordings. Leave the dither to the mastering engineer who is putting the final touches for your cd release. Don't got a mastering engineer or a cd release pending? No big deal. You probably wont be able to tell the difference anyway unless you listen in highly unusual circumstances designed specifically so you can hear the quantization noise/dither.

Last edited by Pokiehat; 04-01-2010 at 05:28 PM.
Reply With Quote
  #7  
Old 04-01-2010, 05:48 PM
rick-slo's Avatar
rick-slo rick-slo is offline
Charter Member
 
Join Date: Nov 2004
Location: San Luis Obispo, CA
Posts: 17,171
Default

Pokiehat, we know this stuff already and clearly we are talking about the dynamic range purely in terms theoretical bit ranges so when metioning 48 range reduction it is in those terms. In reality 24 bit AD converters do not have that range anyway. The one I use specs at Dynamic Range: 120dB A-weighted, 117dB total. My mics have a self noise of 13dB and my preamp specs a EIN Better Than -127dB 20Hz-20kHz @ 40db Gain. The real noise floor in my environment is the computer fan but it is right down there where the noise introduced by dithering would be so I am not that motivated to have an even quieter room - a better acoustic space in general of course would be great.
__________________
Derek Coombs
Youtube -> Website -> Music -> Tabs
Guitars by Mark Blanchard, Albert&Mueller, Paul Woolson, Collings, Composite Acoustics, and Derek Coombs

"Reality is that which when you stop believing in it, doesn't go away."

Woods hands pick by eye and ear
Made to one with pride and love
To be that we hold so dear
A voice from heavens above

Last edited by rick-slo; 04-01-2010 at 09:23 PM.
Reply With Quote
  #8  
Old 04-02-2010, 07:24 AM
Pokiehat Pokiehat is offline
Registered User
 
Join Date: Jun 2009
Posts: 181
Default

I'm not sure what you are trying to say with the numbers but really? Your real noisefloor for after taking into account ambient noise from the mic and thermal noise from mic, mic pre and ad converter is down 100dB from full scale?
Reply With Quote
Reply

  The Acoustic Guitar Forum > General Acoustic Guitar and Amplification Discussion > RECORD

Thread Tools





All times are GMT -6. The time now is 12:03 PM.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.
Copyright ©2000 - 2022, The Acoustic Guitar Forum
vB Ad Management by =RedTyger=