Skip to main content
Topic: Double dithering (Read 4910 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Double dithering

Hi, I guess this should go in here.

I was wondering what would happen if for example I dither a 24bit 48kHz audio file to 16bit 41.1kHz and then make some changes to it, like applying EQ or Reverb or maybe mix it with another file, and then dither again.

Would that diminish the quality of the sound? If so, how terrible would it be?

Would it be better to use an undithered file in the first place or just to skip the dither for the second time?

Thanks for your responses. B)

Double dithering

Reply #1
I'm not an expert on this. But I know some dithering algorithms (like Type I of Waves L2 Ultramaximizer) that is designed for post dither audioprocessing if necessary. However, as a rule of thumb dither should be applied at the last stage hence if you can avoid feeding a dithered wave into the second stage after which you dither again you should. Or use a dithering algorithm in the first stage that is best suited for that purpose.

"What if I need to do any further editing? Can I use IDR again? Can I use it with another dither algorithm?"

Waves IDR technology can be used over itself or in conjunction with any other dither algorithm. Therefore if it is necessary to use it multiple times you can without fear of harmful side effects. We do suggest that if any further editing is required to use normal noise shaping
The object of mankind lies in its highest individuals.
One must have chaos in oneself to be able to give birth to a dancing star.

Double dithering

Reply #2
It would be best to do all your editing at 24/32-bits and then dither once to 16-bit.

Double dithering

Reply #3
Actually, the best thing would be to do all your processing at 24/48 where dither is not needed, and then dither a single time when you convert to 16/44.

Once you have converted to 16/44 I wouldn't dither again, although the program you use to do the processing might do dithering at various steps. I've noticed that Cool Edit, for example, will dither when you simply amplify a wav (in 16-bit mode).

Double dithering

Reply #4
Yes, CEP does that if you have the option "Dither Transform Results (increases dynamic range)" in Settings --> Data.


Double dithering

Reply #5
Yes, CEP does that if you have the option "Dither Transform Results (increases dynamic range)" in Settings --> Data.


Thanks! I use Cool Edit all the time and never saw that setting, and the other day I was doing something where I really wanted that turned off.

Maybe if I look through all the settings I might find other treasures! 

Double dithering

Reply #6
So this means that CEP does dithering on every process you execute?

I've also noticed that foobar2000 dithers the output, and every single audio file has already been dithered during the mastering process, so... I'm getting a little confused about all this.

Also, a lot of audio programs claim to have internal 32bit processing, doesn't that mean that all signals are upsampled to 32 bits and then resampled to 16bits?

Any info on books, websites, or code related to dithering & resampling would be greatly appreciated.

Double dithering

Reply #7
I've noticed that Cool Edit, for example, will dither when you simply amplify a wav (in 16-bit mode).

That's a good thing.

btw, you can listen to a couple of dithering options here:

If you do a search for dither on HA (and on google, and on the syntrillium cool edit forums) you should turn up more information than you need! If you filter the results by my user name you might find some pearls of wisdom. (or not!)


Double dithering

Reply #8
Hi all,

I'll try and flesh out some detail on this one, wearing my CD mastering engineer's hat!

Dithering is a process ideally applied whenever digital audio from a higher bit rate is transformed to a lower bit rate. Note we are referring to bit rate , not sample rate.

The aim here is to reduce or minimise quantisation errors. The visual equivalent, as a metaphor, is like the process of copying a very high definition fine grain image onto a coarse grain lower definition film. The coarse grain film cannot handle all the high definition of the original image, but if we add a bit of soft focus or "fog" to the coarse grain film copy, then the granularity will not be so visible and the overall visual impression will be a "smoother" image with very slightly less detail and slightly less dynamic range.

In digital audio the "fog" or "soft focus" we add is a variety of noise - very low level hiss. (these can be "frequency shaped", but that's another story) The effect of this is to smooth out the granularity of coarser grained lower bit rates and give the impression of smoother treble and smoother very low level sounds around the zero crossing point.

So going (say) from 24 bit to 16 bit we dither to try and capture more perceptual detail than merely truncating the file from 24 to 16 bits. If we truncate - that is convert from 24 bit to 16 bit by simply throwing away 8 bits and no dithering -, we lose some fine detail. Ok its not a total disaster, but there is a small difference .

Now the  crucial point.  If we process at 32bit floating we should dither down to 24 bits after each process. If your software supports 32 bit files, you could do all your processing at 32bit taking advantage of the mammouth headroom available (over 1000dB!!) and dither to 24 or 16 bit for the final output - with suitable level adjustment. However 32 bit files are BIG files!!

Otherwise do all processing and files at 24 bit resolution and diher only at the final stage down to 16 bit. Each dither process as a minute amount of noise to your audio.

Crossfades -
If you have 16 bit files and crossfade, most processors will be using 32 or 24 bit processing, so without dithering the crossfaded section will be truncated. Also if you make a fadeout on a 16 bit file truncation will occur unless you specify dithering.

Multiple dithering adds more low level dither noise, note that in the case of 24 and 16 bit noise it is very low level and is generally much lower than tape hiss. 8 bit dither noise is a lot like severe tape hiss!!

The  rule therefore is to keep all audio at the highest resolution until the final output and then dither to the final lower bitrate, but in general terms whenever a level change or process is implemented at a higher bit rate, dithering should occur if the output is being saved at a lower bit rate.

I hope this helps the discussion.



Double dithering

Reply #9
I think it would be better to talk about "bith depth" instead of "bit rate".

Another more techical explanation would be that whenever you go from analog to digital or from a higher bit depth to a lower bit depth, quantization happens. This quantization always produces what is called quantization error. If no dither is used and plain truncation or rounding is used instead, the error produced is correlated with the signal, or in other words, the error varies with the signal. That kind of error is called quantization distortion. It is called distortion precisely because it is not constant and varies with the signal.

But, adding a little amount of a special kind of noise to the signal before the quantization happens, or in other words, dithering properly the signal, it is possible to totally decorrelate the quantization error from the signal, so that the error is now random or pseudo-random, translating into a constant noise. Then, there is no quantization distortion anymore, but quantization noise instead. As I said, this noise is not correlated with the signal but is a constant noise, same as analog noise.

Also, using dither it is possible to record and play signals that are below the LSB (least significant bit), or in other words, that are below the max. dynamic range possible with the bith depth used. For example, 16 bit allows a max. dynamic range or SNR of 98.08 dB (6.02 * 16 + 1.76). However, using plain dither it is perfectly possible to sample a signal of -110 dB and below.

As a summary, proper dither removes totally quantization distortion leaving just quantization noise, and "analogizes" digital, so that it behaves same way as analog in respect to noise floor behaviour.

All this accounts even for the simplest type of dither possible, which is spectrally flat dither. Other more complex types of dither such as noise shaped dither use a kind of noise that is spectrally shaped in order to be less audible for human ear.

SimplePortal 1.0.0 RC1 © 2008-2018