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: Dithering necessary for upsamplig (Read 8120 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Dithering necessary for upsamplig

I'm wondering whether it is useful to dither when you upsample from 16 to 24?
Thank you
Cheers,

Dithering necessary for upsamplig

Reply #1 –
I don't think so.

Dithering necessary for upsamplig

Reply #2 –
Quote
I don't think so.
[a href="index.php?act=findpost&pid=258994"][{POST_SNAPBACK}][/a]

I don't remember where but I red somewhere that it is also recommended when you upsample. So I'd like to know if it's true and if yes why (cos I can't see why).

Dithering necessary for upsamplig

Reply #3 –
yes you should do light dithering to make it sound a bit better and then upsample with ssrc because you dont dither at 24 and above.

Dithering necessary for upsamplig

Reply #4 –
Quote
yes you should do light dithering to make it sound a bit better and then upsample with ssrc because you dont dither at 24 and above.
[a href="index.php?act=findpost&pid=259034"][{POST_SNAPBACK}][/a]


Incorrect.  The only time you need to dither is when reducing bit depth.  Going from 32 or 24 bits to 16 bits is when you might need to dither.  Remember that dither is nothing more than added noise which helps cover quantization noise.  Up and down sampling is altogether unrelated.

Dithering necessary for upsamplig

Reply #5 –
Quote
Quote
I don't think so.
[a href="index.php?act=findpost&pid=258994"][{POST_SNAPBACK}][/a]

I don't remember where but I red somewhere that it is also recommended when you upsample. So I'd like to know if it's true and if yes why (cos I can't see why).
[a href="index.php?act=findpost&pid=258998"][{POST_SNAPBACK}][/a]

Upsampling is not the same thing as increasing bit depth. Upsampling means increasing the sampling frequency, like 44.1 kHz -> 48 kHz. When changing sampling frequency i guess dithering is always useful.

Dithering necessary for upsamplig

Reply #6 –
Quote
when reducing bit depth
Not only then, when changing the loudness also dither should be applied. I did an extreme experiment: make two sine waves, ie 250 Hz and 8020 Hz and look at the IMD+N (distortion) component after reducing the full-scale signal with 70 dB. Properly dithered this gives 0.3 % distortion whereas Sound Forge hits the 2 %, where it is clear that no dither has been applied.

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

Dithering necessary for upsamplig

Reply #7 –
Quote
Quote
when reducing bit depth
Not only then, when changing the loudness also dither should be applied. I did an extreme experiment: make two sine waves, ie 250 Hz and 8020 Hz and look at the IMD+N (distortion) component after reducing the full-scale signal with 70 dB. Properly dithered this gives 0.3 % distortion whereas Sound Forge hits the 2 %, where it is clear that no dither has been applied.

Regards,
Jacco
[a href="index.php?act=findpost&pid=259394"][{POST_SNAPBACK}][/a]


It's best to perform all loudness changes in very high bitdepth and to dither one times at the end, when converting down to 16bit, in order to add as few noise as possible.
There is absolutely no sense in dithering when increasing only the bitdepth. It makes the audio even worse cause it adds noise.

Maybe many people don't know how dither works. In 16bit at very low volume a sine wave can't be drawn properly using the few available sample values - it looks rather rectangular. This is then audible as "quantization distortion". But in 32bit the same sine wave looks pretty smooth because of a lot more sample values.
So by changing from 16bit to 24bit the curve form won't change and will sound the same, because the added last 8bits of each sample aren't used - they are just 0s.
But when converting from 24bit to 16bit, adding a small amount of white noise makes it possible to have the correct sine wave mixed into the noise.
So less distortion - but more noise. Some people prefer to have distortion.
I know that I know nothing. But how can I then know that ?

Dithering necessary for upsamplig

Reply #8 –
Quote
There is absolutely no sense in dithering when increasing only the bitdepth.
Also if the 16 bits signal is not properly dithered in the first place? Can it be that if you apply proper dither to a non-dithered 16 bit source which is to be upsampled (only in terms of bitdepth) to 24 bits, the resulting output spectrum is more independant than the one you started with. I did some experiments (with ssrc, this is really an outstanding piece of software to do dither) which points into that direction.

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

Dithering necessary for upsamplig

Reply #9 –
Quote
Quote
There is absolutely no sense in dithering when increasing only the bitdepth.
Also if the 16 bits signal is not properly dithered in the first place? Can it be that if you apply proper dither to a non-dithered 16 bit source which is to be upsampled (only in terms of bitdepth) to 24 bits, the resulting output spectrum is more independant than the one you started with. I did some experiments (with ssrc, this is really an outstanding piece of software to do dither) which points into that direction.

Regards,
Jacco
[a href="index.php?act=findpost&pid=259487"][{POST_SNAPBACK}][/a]


You can never get more bits than you already had. "more independant" ? I don't know what exactly you mean here. 24bit has more capabilities, but nothing you could reach by dithering when upconverting.
What is ssrc ? (I hope I don't appear unreliable because of this question.)
I know that I know nothing. But how can I then know that ?

Dithering necessary for upsamplig

Reply #10 –
Quote
What is ssrc ?
Abreviation for "Shibatch sample rate converter" In my humble opinion the best sample rate converter out there, at least the best one I have seen.

Quote
You can never get more bits than you already had.
I do not know if this is relevant and I will also not comment on it. Although my feeling tells me that you can't. What I mean is this: suppose that your 16 bit recording was not properly dithered which results in an effective "bitdepth" of lets say 10 bits. In other words, the distortion + noise of the resulting 16 bits signal is so worse that the dynamic range is 10 bits large. By upconverting to 24 bits with proper dither, there might be chance that one can go to the 16 bit level of performance. Better than that seems to me not possible.

I have never seen an article on this in literature but simple experiments point into that direction.

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

Dithering necessary for upsamplig

Reply #11 –
Jacco,

No offense, but you're way off the mark here.  Take a break and read this.

It will give you a good baseline on how digital audio works.  Enjoy.

Dithering necessary for upsamplig

Reply #12 –
It is very hard for non-native English speakers (like me and dekkersj, I think) to read all those articles that don't bother to use 'primitive' English.

Anyway, what you said, dekkersj, sounds pretty much like a placebo effect. Be careful to not violate rule no. 8 !
The properly dithered upconverted 24bit audio will have the same distortion + noise that the original has and, additionally, extra noise caused by the extra dithering. Possibly the extra noise masks some of the original distortion and therefore it might sound better to you.
In any case, is dither available for upconverting at all ? Not in CEP at least.
I know that I know nothing. But how can I then know that ?

Dithering necessary for upsamplig

Reply #13 –
Hi,

I have read the article and it brought me little extra knowledge, apart from the suggestion that it is very hard to make good 16 bit stuph. But I guess I am wrong wrt "making more independant" of something that is not properly dithered. The distortion and noise introduced by the not proper dithering action cannot be removed afterwards, unless it is a deterministic signal. This is of course not the case for audio.

In SSRC there is the possibility to apply dither in upconverting.

Regards,
Jacco
Logical reasoning brings you from a to b, imagination brings you everywhere.

Dithering necessary for upsamplig

Reply #14 –
The definitive answer is that you should not dither when just:

increasing the bitdepth
increasing the amplitude by a factor of 2
using cut/paste/edit/move type operations without cross fading


Almost any other operation should, in theory, be dithered before truncation to the target bitdepth.

e.g. if you resample 44.1k 16bit to 96k 16bit, the resampling will generate as many bits as the calculation allows (e.g. floats, Int32 etc etc - just as 3 divided by 2 generates as many decimal places as you care to count, even though both numbers are integers) - the output should then be dithered to 16-bits.

If your calculation is working internally at the target bitdepth, there's nothing to dither, and the result is not even accurate to that bitdepth. At 24-bits, that doesn't matter much - at 16-bits, it's a very bad way of working - very few programs do this though. Most 16-bit outputs from complex calculations are generated using a much higher internal precision. Unfortunately dither is often neglected at the output though!

Cheers,
David.

Dithering necessary for upsamplig

Reply #15 –
However, there is a counter argument that, when going from 16-bit to 24-bit, it may be beneficial to fill the last 8 bits with noise. This may help to smooth out non-linearities in a bad DAC.

With a good DAC there is no advantage, and with a carefully noise-shaped 16-bit signal, noise added to the 17th bit will remove the benefit of noise shaping because it'll be louder than the digital noise floor in the original signal!

So, in conclusion, the answer is still no!

Cheers,
David.

Dithering necessary for upsamplig

Reply #16 –
I think I got it!
Thanks everybody.