Skip to main content
Topic: [TOS #5/8] Re: Should I dither? (Converting 24bit to 16bit FLAC) (Read 992 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

[TOS #5/8] Re: Should I dither? (Converting 24bit to 16bit FLAC)

Given my history on this forum I am probably not the best person to advise on dither.
But I can tell this.
In practice you will not hear a difference in vast majority of 24-16 bit conversion, so strictly speaking it is not neccessary to dither if we consider primarily the audibility.
In theory however it is highly recommended to dither on every bit reduction.
I use plain TPDF dither (mostly) or lightly shaped dither like modified-e-weighted, or low-shibata. This I can recommend to you since it has been my workflow this year when I need to go down to the 16 bit.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #1
I use plain TPDF dither (mostly) or lightly shaped dither
Sigh.

Given my history on this forum I am probably not the best person to advise on dither.
Indeed!
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #2
I use plain TPDF dither (mostly) or lightly shaped dither
Sigh.

Given my history on this forum I am probably not the best person to advise on dither.
Indeed!


I know that I do not know the details of TPDF, but still the recommendation of TPDF as basic dither algorithm is OK from me, since I experienced no issues with that type of dither this year. Noise-shaped that I mentioned he can also try.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #3
Please review the link you quoted instead of ignorantly repeating the same false dichotomy.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #4
There is an example of batch processing in Sox's package. It can be run by drag and dropping multiple files from a folder (e.g. an album) onto its shortcut on the desktop but must be updated by correct parameters.

Even if people here would disagree, when using command line, I would in your place try to use TPDF and then modified-e-weighted/low shibata dither on one or two "typical" file of converted musical genre and listen to the results on your own. If you cannot hear the difference, then choose whatever you like, I would go for TPDF in that case. If you prefer more one of the shaped dithers and/or hear a difference in favour of them, then choose one of the noise shaped ones.

But Foobar GUI is much more convenient.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #5
Even if people here would disagree, when using command line, I would in your place try to use TPDF and then modified-e-weighted/low shibata dither on one or two "typical" file of converted musical genre and listen to the results on your own.
Why would anyone here disagree to trying different algorithms?

If you cannot hear the difference, then choose whatever you like, I would go for TPDF in that case.
You like one over the other even though you can't hear a difference.  How reassuring!

BTW, why do use the term "TPDF" for dither without noise shaping?
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #6
The SoX docu also uses this kind of wording for default flat dither to make clear what kind it is. "SoX automatically adds TPDF dither when the output bit-depth is less than 24 and any of the following are true:..."
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #7
Another successful thread hijacking to promote jumpingjackflash5 approved TPDF™ dithering.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #8
The SoX docu also uses this kind of wording for default flat dither to make clear what kind it is. "SoX automatically adds TPDF dither when the output bit-depth is less than 24 and any of the following are true:..."
Is SoX specifying the density function for shaped dither?
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #9
Another successful thread hijacking to promote jumpingjackflash5 approved TPDF™ dithering.
...and further demonstrate his perennial ignorance in the face of repeated attempts to educate him.

Are you suggesting his posts be binned per TOS  #5?
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #10
Is SoX specifying the density function for shaped dither?
I for example understand TPDF as being flat until described otherwise. Nothing to waste to much hair splitting energy. Another sentense from the SoX docu:
"With no options, this effect will add triangular (TPDF) white noise. Noise-shaping (only for certain sample rates) can be selected with −s"
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #11
If that is all they have to say about noise-shaping then you could have simply said no.

This isn't hair-splitting.  It speaks directly to what TPDF means and how it relates to dither.

In this case not only do we have an individual with no objective basis for his recommendations, but one who doesn't understand the terminology or even how it's supposed to work.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #12
Even if people here would disagree, when using command line, I would in your place try to use TPDF and then modified-e-weighted/low shibata dither on one or two "typical" file of converted musical genre and listen to the results on your own.
Why would anyone here disagree to trying different algorithms?

If you cannot hear the difference, then choose whatever you like, I would go for TPDF in that case.
You like one over the other even though you can't hear a difference.  How reassuring!

BTW, why do use the term "TPDF" for dither without noise shaping?

As stated by others just because SoX's documentation does it too ... "... Dithering deliberately adds a small amount of noise to the signal in order to mask audible quantization effects that can occur if the output sample size is less than 24 bits. With no options, this effect will add triangular (TPDF) white noise ...."

Yes, if I cannot hear a difference, I would use TPDF (being the default), which is my subjective preference for 44100 sample rate that he asked for. I would probably use shaped dither for 48000, because noise shaping can work *theoretically* (practically I cannot prove it) more effectively at this rate (there is 2 kHz more space to shift noise to inaudible frequencies). I think I know exactly what irritates you on that statement, but in this thread another used asked about how to dither. So I tried to show him my workflow which works very well for me, I even posted my batch files. What is wrong ? Anybody can suggest his workflow, too ....

Since our conversation can be confusing for the user that asked for advice, let me be more clear (for him). If you do not hear a difference nor have a subjective preference, you can use the SoX default dither algorithm with peace in mind and forget about dither problem. If you would like to use noise shaped dither at 48000 or even at 44100 Hz (which is also the case if you use Foobar GUI where you cannot select dither type), it is also completely OK, it theoretically brings even higher SNR, but in practice you (probably) will not hear a difference in ABX tests.


Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #13
As stated by others just because SoX's documentation does it too ... "... Dithering deliberately adds a small amount of noise to the signal in order to mask audible quantization effects that can occur if the output sample size is less than 24 bits. With no options, this effect will add triangular (TPDF) white noise ...."
We get to add reading comprehension to your list of deficiencies.

Maybe I should rephrase my question...
why do use the term "TPDF" for dither without noise shaping?
...to:
What in the documentation specifically gives you reason to think noise-shaped dither isn't also triangular (TPDF)?

I would probably use shaped dither for 48000, because noise shaping works *theoretically* (practically I cannot prove it) better at this rate (there is 2 kHz more space to shift noise to inaudible frequencies).
Hmm, only "theoretically"?  Why do you think noise shaping must move noise into frequencies that are only inaudible?  If you think Fletcher-Munson is theoretical then you're sorely mistaken.  It would be one thing if you even understand the theory, but sadly you clearly don't.  You don't even appear to know how SoX performs noise shaping.

What is wrong ?
Funny how you pretend this has something to do with providing a batch file.  Bennetng called it correctly, so it isn't just me.

Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #14
As stated by others just because SoX's documentation does it too ... "... Dithering deliberately adds a small amount of noise to the signal in order to mask audible quantization effects that can occur if the output sample size is less than 24 bits. With no options, this effect will add triangular (TPDF) white noise ...."
We get to add reading comprehension to your list of deficiencies.

Maybe I should rephrase my question...
why do use the term "TPDF" for dither without noise shaping?
...to:
What in the documentation specifically gives you reason to think noise-shaped dither isn't also triangular (TPDF)?

I would probably use shaped dither for 48000, because noise shaping works *theoretically* (practically I cannot prove it) better at this rate (there is 2 kHz more space to shift noise to inaudible frequencies).
Hmm, only "theoretically"?  Why do you think noise shaping must move noise into frequencies that are only inaudible?  If you think Fletcher-Munson is theoretical then you're sorely mistaken.  It would be one thing if you even understand the theory, but sadly you clearly don't.  You don't even appear to know how SoX performs noise shaping.

What is wrong ?
Funny how you pretend this has something to do with providing a batch file.  Bennetng called it correctly, so it isn't just me.



Maybe then I am really confused about dither. For your questions: I think that by default SoX by default applies triangular (TPDF) noise. When using noise shaping it shapes (shifts) that added noise into higher frequencies, so yes, that noise is still triangular, but then shifted. Yes it may be better to speak about dither without noise shaping (being the SoX's default) and dither with noise shaping {and then talking about algorithm used for that noise shaping, not dither itself).

Yes noise shaping can move noise also to audible frequencies that are less perceived by human ATH curve and this way improve results.  For sure, especially at 44100 (or lower) SR we also shift to frequencies that we can hear (e.g.  13-19 (20) kHz). But if we have more space at higher (or even totally inaudible >20 kHz) frequencies then theoretically noise shaping is more effective, or no ?

If I am still wrong or mistaken you can bin this post and leave here just my other ones where I posted my workflow. I know I have weaknesses in theory of audio.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #15
Maybe then I am really confused about dither.
You are.

For your questions: I think that by default SoX by default applies triangular (TPDF) noise.
You forgot to include a very critical word: white.

Triangular probability density function has nothing to do spectral content.  White has everything to do with spectral content. Arny told you this already and I have sent you back to his link.  I would think that if you had bothered to study it rather than just assume you already knew it, things would be different now.

When using noise shaping it shapes (shifts) that added noise into higher frequencies, so yes, that noise is still triangular, but then shifted.
Shape does not mean shift.  The noise is not triangular, it's probability density is triangular.

Yes noise shaping can move noise also to audible frequencies that are less perceived by human ATH and this way improve results.
Less perceived by human ATH, you say?
https://en.wikipedia.org/wiki/Absolute_threshold_of_hearing

For sure, especially at 44100 (or lower) SR we also shift to frequencies that we can hear (e.g. above 13-20 kHz). But if we have more space at higher (or even totally inaudible >20 kHz) frequencies then theoretically noise shaping is more effective, or no ?
How does one determine what is effective?  Remember, we're talking about sound quality.

I know I have weaknesses in theory of audio.
It becomes a problem when you use this weakness as your sole means for justifying the recommendations you're giving others.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #16

I know I have weaknesses in theory of audio.
It becomes a problem when you use this weakness as your sole means for justifying the recommendations you're giving others.


Alright, then I'll try to hold back from those recommendations at this forum, because it leads to unwanted controversies where I am less theoretically equipped that the others, so it is not productive. I was just having good experience with some workflow(s), so I shared them. I might still occasionally contribute here, but shall not discuss e.g. dither type etc.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #17
Even if people here would disagree, when using command line, I would in your place try to use TPDF and then modified-e-weighted/low shibata dither...
jjf5 i have the feeling you mean some things correctly but mix things up when you write it down. The sentense i quoted above is a prime example. Think about it. I hope it is directly giving you a hint instead of confusing you ever more.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #18
Even if people here would disagree, when using command line, I would in your place try to use TPDF and then modified-e-weighted/low shibata dither...
jjf5 i have the feeling you mean some things correctly but mix things up when you write it down. The sentense i quoted above is a prime example. Think about it. I hope it is directly giving you a hint instead of confusing you ever more.


Thank you.

Yes I agree that sentence is confusing and suggests that e.g. modified-e-weighted/low shibata is "another/further dither" which it is not (it is the noise shaping algorithm). I tried to improve it in next posts.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #19
I was just having good experience
You may have had an equally good experience with some other "workflow" provided you didn't allow it to become tainted by your biases (some of which were likely derived from a less than adequate understanding of the subject at hand).
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #20
I was just having good experience
You may have had an equally good experience with some other "workflow" provided you didn't allow it to become tainted by your biases (some of which were likely derived from a less than adequate understanding of the subject at hand).

Alright. I have already stated that since my participation on some topics leads here to repeated confusions (this thread being another example) I'll try to avoid them, and if necessary to discuss them on another forum in my native language. It takes a lot of energy to debate them and sometimes we go in circles or I am being faced with public demonstration of what everything I still do not know. If possible, I still will be here for topics that are more general, audio software/hardware oriented etc.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #21
Another successful thread hijacking to promote jumpingjackflash5 approved TPDF™ dithering.
...and further demonstrate his perennial ignorance in the face of repeated attempts to educate him.

Are you suggesting his posts be binned per TOS  #5?
I posted this before I go to sleep and hoped stuffs can be settled today. I have a feeling that even if you binned his posts or locked this thread such things will still happen in other threads.

I think the screenshots explained what TPDF is not, if he is so unwilling to read Arnold's explanation.

I'll try to avoid them
https://hydrogenaud.io/index.php?topic=113149.msg931624#msg931624
Quote
I would not debate this way again
LOL

...in my native language
English is not my native language as well and I think your English is much better than mine.




Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #22
I recognize that you want to show that one thing is the type of probability density function used for dither - SoX default using triangular one (TPDF) and second thing is whether we utilize noise shaping or not and how (modified-e-weighted, low-shibata and other noise shaping algorithm). I (unfortunately) mixed this up in previous posts. But I agree we are now pretty far from original post.

Re: Should I dither? (Converting 24bit to 16bit FLAC)

Reply #23
Anyway through Christmas I will try to find time for some personal listening tests, so that I can keep or change my workflow also based on more objective (practical) experience. Theoretical level is one thing (and I am moreover showing weaknesses there compared to some other members), practical audibility and usability second thing. I apologize if I confused things in this thread, the moderators can clean up my posts in this thread so that only those which are relevant to the original topic and questions remain.

 
SimplePortal 1.0.0 RC1 © 2008-2019