Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Question - advice on dither  (Read 5976 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Question - advice on dither

Since experienced members here I'll try to ask practical question for my friend that is having a small collection of recording that he digitized from his home tapes before they were worn out (not performing well now). So now he has them in 24/96 FLAC and did some slight processing on them. If he wants to make CD quality FLAC (16/44.1) from some of the recordings, he uses SoX for this and results are OK, but he wonders if he must use dither (TPDF or noise shaped) as the tape recording already have the tape hiss present. Or he would be better off not to use dither at all in this case, since noise is already heard on recording?

Re: Question - advice on dither

Reply #1 –
Dither is still required. The conventional wisdom is that noise shaped is preferable if no further processing will be done, otherwise use TPDF. Personally, I'd use TPDF since there's already audible noise.
Regards,
   Don Hills
"People hear what they see." - Doris Day

Re: Question - advice on dither

Reply #2 –
Quote from: splice on
Dither is still required.
Are you sure? I'd say there is very likely enough noise in the recording already to make dither redundant.

Re: Question - advice on dither

Reply #3 –
Dither is the addition of random noise before the bit depth conversion. Adequate dither eliminates the distortion due to the quantization error that is unavoidable when reducing bit depth. With appropriate test tones the quantization distortion (when done without dither) is readily recognized but with recording of most real audio the distortion is inaudible unless the bit depth is less than 16, possibly less than 12.

Tape hiss is not the best random noise for the purpose but it does work well enough. It can be easily conceptualized as a signal (the desired audio) mixed with some random noise: dither. Adding more dither of a desirable nature will not make things worse but will not make any audible improvement either. This can be easily enough tested with a spat of ABX trials.

Re: Question - advice on dither

Reply #4 –
Yes in this  case theres is audible noise because originals were on tape which was recorded about 20 yrs ago.

Re: Question - advice on dither

Reply #5 –
If as you wrote in this particular case adding dither will not make audible inprovement then it is a useful i formation for me (my friend).

Re: Question - advice on dither

Reply #6 –
Your line of reasoning is probably:
Dither is adding noise,
tape hiss is noise,
hence why bother adding even more noise (the dither)?

Sounds logical but this is not the way it works.
When you reduce the bit depth from 24 to 16, you get a systematic error (quantization error).
This of course applies to the entire audio signal (including the hiss).
Dither de-correlates this error.
A good read: https://en.wikipedia.org/wiki/Dither
TheWellTemperedComputer.com

Re: Question - advice on dither

Reply #7 –
Quote from: Roseval on
Your line of reasoning is probably:
Dither is adding noise,
tape hiss is noise,
hence why bother adding even more noise (the dither)?
Yes, exactly, provided that the noise which is already there is sufficient in level. In this case adding more noise doesn't do anything useful. If the noise already present isn't sufficient in level, you need to add dither. In the OP's case, it would be very unlikely that there's insufficient noise. More likely, the noise is way above the quantisation level.

Is there anything in the Wikipedia article that contradicts this?

Re: Question - advice on dither

Reply #8 –
I also read about the principles described above - one of the reasons I was not able to give my friend straight advice.

In this case there is an audible hiss above the levels which usually tapes have, and the Dolby B or C was not used when recording he original.

Re: Question - advice on dither

Reply #9 –
The important point is that the noise needs to have a minimum level in order to randomise the quantization sufficiently well. "Proper" dither tries to add the minimum amount of noise necessary to achieve this. The dither can even be spectrally shaped so that its audible effect is reduced, even though its level isn't reduced. Other than that, there's nothing special or magical with dither.

In your case, the level of tape hiss is going to be far above the level of quantization noise at 16-bit. I don't know of any analog storage medium that was low-noise enough to require dither for conversion to 16-bit PCM.

Re: Question - advice on dither

Reply #10 –
Thanks for explaining.
 

Re: Question - advice on dither

Reply #11 –
Quote from: pelmazo on
Yes, exactly, provided that the noise which is already there is sufficient in level. In this case adding more noise doesn't do anything useful.

If tape hiss would be a random noise it probably won't make sense to dither.
But is it random noise or does it have a distinct pattern?
TheWellTemperedComputer.com

Re: Question - advice on dither

Reply #12 –
Seems to be normal hiss, but on different tracks it slightly differ in volume.

Re: Question - advice on dither

Reply #13 –
Since you seem to ask things about dither every once in a while you may try to convince yourself once and never wurry again!
I guess you decided to go with SoX already. How about creating a file with silence and apply all dither types found in SoX?
Amplify the files by 40dB and listen. Pick what noise sounds best and done.
There are 10 shapes including sloped and different shibatas SoX noise shaping
Use -a with dither to avoid pointless dither on silence. I just had Placebos Black Market splitted only to find several minutes of silence dithered.
If you really create CDs with the downsampled music make the frame boundaries fit CD specs to avoid hiccups at tjhe gaps. https://hydrogenaud.io/index.php/topic,101691.msg838986.html#msg838986
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Question - advice on dither

Reply #14 –
Thanks for the tips. Yes I use SoX command line works very well. Before I asked because of my own concerns, this thread is because of my friend as I wrote at first.

Re: Question - advice on dither

Reply #15 –
In an effort to minimize theoretical hand-waving...

Feel free to provide samples showing how dither creates audible changes to a recording that contains tape hiss.  Feel free to do the same with vinyl as the source.

I only request that the recording be peak-normalized to within no less than -4dBFS prior to the addition of dither.   Ideally I would like 0dBFS, but think it's fair to allow headroom for samplerate conversion in order to prevent clipping.  I'd prefer actual musical content rather than signals synthesized solely as an attempt to demonstrate a difference, though these would be interesting as well.

Use whatever shaping you would normally use for conversion. I'm not interested in tailoring shaping for the sole purpose of revealing audible differences.

Objections to the specifics of my request are welcome, at which point I'd be more than happy to split the discussion if they would otherwise detract from this topic.

Re: Question - advice on dither

Reply #16 –
While it won't satisfy graynol's request, I can state that I've done many tests. I'm sure other people have too. I doubt their results were significantly different from mine, I'm just not interested in repeating such things.

With CoolEdit, which I used to record and process many hundreds of LPs and cassette tapes, it is possible to generate signals of virtually any level without the complication of analogue noise contamination. CoolEdit has a very good spectral view and a some other useful, if not ultimate. investigatory tools such as its statistical analysis and frequency analysis graph.

With generated signals it is very easy to see the quantization distortion vs undistorted but noisy results of dithering from bit reduction, adjusting the spectral plot parameters as necessary. One can also easily mix other noise, such as recordings of plain tape hiss and unmodulated LP track with the signal, to compare the results with the many types and levels of dither CoolEdit can supply. One can repeat these test as many ways as desired with music recordings from LP and cassette.

Very low level signals show the results most clearly of course. With a good, quiet, headphone amplifier, which I fortunately have, one can turn up the volume enough to listen for differences.  I thoroughly convinced myself of what I wrote in my first post. Anyone with similar tools can do the same kind of investigation. It is pretty hard to dispute the data, as revealed by the tools mentioned. It would be surprising if anyone could produce listening tests that did not agree with the reality of the data (except, perhaps by the methods of consumer audio journalists).

In spite of the lack of audible difference, I dither my analogue conversions as a matter of course. I record in 32 bit and convert to 16 bit only as the very last step. I use dither parameters I chose many years ago that met my aesthetic tastes. I don't recall everything, but basically they are the options which produced the least amount of added noise that made quantization distortion undetectable in spectral view. The dither itself, with no signal, is quite below my hearing detection at any reasonable listening level so its addition carries no penalty.

Re: Question - advice on dither

Reply #17 –
From what you have discussed it seems to me that in this particular case dither is not absolutely neccessary but if done it won't do harm and will diffuse possible quantization error. So I will recommend to my friend to try without dither first especially on tracks that have higher volume of hiss, if he is not satisfied with the results for some reason, then use (preferably light] dithering.

Thank you.

Re: Question - advice on dither

Reply #19 –
Quote from: greynol on
In all seriousness, how is your friend going to determine whether he's satisfied with the results?

In this simple case, it would be enough for him if he does not hear audible artifacts/further degradation on the created 16/44.1 FLACs. As I wrote some of the recordings are already of lower quality since part of them were recorded at home about 20 yrs ago.

Re: Question - advice on dither

Reply #20 –
I guess what most do is simply to apply dither and never look back, use some shape you like for a fuzzy feeling and your done.
In case someone always wanted to listen to dither but never had some at hand.
SoX dithernoise in loud in Uploads
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Question - advice on dither

Reply #21 –
He'll be conducting a properly-controlled ABX test?

I guess I'm a little concerned whether the conclusions reached will be based solely on how the results sound and controlled in such a way that the use of dither is the only variable.

Re: Question - advice on dither

Reply #22 –
Quote from: Wombat on
I guess what most do is simply to apply dither and never look back, use some shape you like for a fuzzy feeling and your done.
In case someone always wanted to listen to dither but never had some at hand.
SoX dithernoise in loud in Uploads

Thank you very much for those examples, they provide an "audible" insight to dithering even for me.

Re: Question - advice on dither

Reply #23 –
Quote from: greynol on
He'll be conducting a properly-controlled ABX test?
Good one :)
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Question - advice on dither

Reply #24 –
Quote from: greynol on
He'll be conducting a properly-controlled ABX test?

I guess I'm a little concerned whether the conclusions reached will be based solely on how the results sound and controlled in such a way that the use of dither is the only variable.

I will speak to him about it, he has foobar so he can do it  , maybe then he be more sure about the results.