dithering question

[Westmalle]

Hi folks,

When you have a 44k/24 bit master, that needs to be converted to 44k/16bit, is it always necessary to add the appropriate dither?
The inherent noise of the recording I'm doing is on a constant level of -60dB.

Maybe it is good enough to add dithering to the fade ins and fade outs?

[zooloo]

Dithering 101 thread

[Westmalle]

Thanks for the link. I thought so.

I will only dither the fade ins and fade outs with 24bit dither.
I don't hear any zipper noise doing that.

As Mark mentioned above all you have to do is a few minutes of listening (at cranked levels in a quiet room) to 24bit files whose audio is disappearing slowly into the noise floor (i.e. fades or reverb tails) first at full resolution, then trucated to 16bit, and then with various dither algorithms added when requantizing to 16bit - to very easily hear that the fades and reverb tails are smooth at 24bit, fitzing and sputtering when truncated at 16bit, and at a good compromise of slightly added noise but much smoother than truncated when dithered to 16bits.

[Westmalle]

Quote:
Originally Posted by Nishmaster
I think what people fail to understand concerning dither is that at any given point in your program material, there is a certain probability that individual samples may fall at a very low amplitude LSB area regardless of how loud the material is. When truncating instead of dithering, you are effectively bashing those samples to digital zero, thus generating wideband distortion that gets worse as you approach the Nyquist limit and your probability of low amplitude signals increases and number of samples per a given frequency decreases (i.e. sampling the rising/falling edge of a waveform near the zero crossing).

This is that "edge" you hear, and it is completely audible and measurable. Intrinsically, louder program material has a lower probability of low amplitude samples, so it may not be as audible as quieter sources. However, as the RMS of the signal is decreased, the probability of low amplitude samples is increased until it is nearly 100%, resulting in zippering/sputtering/fuzzing/etc. Dithering combats this of course by randomizing slightly the amplitude. The point at which zippering occur and where the signal is buried in dithering noise may be close to the same, but the audible artifacts of white noise are not as objectionable as the distortion products caused by truncation.
Ah, now that's a very good explanation! Indeed, no matter how complex the wave, there have to be zero crossings, and any samples occurring at or near those zero crossings are susceptible to increased quantization error and thus the accompanying distortion. That actually makes a great deal of sense -- thanks!

But this I never thought about....

[David Rick]

Originally Posted by Westmalle

I will only dither the fade ins and fade outs with 24bit dither.
I don't hear any zipper noise doing that.

Truncating sounds different than dithering, even when you're not in a fade. In my experience, it mostly sounds worse, but if the original material is a bit dull-sounding, then those added harmonics can sometimes be a help. Standard TPDF dither at the 16-bit level tends to push things the other direction, towards "fuzzy". For that reason, I usually favor some kind of noise-shaping algorithm. My workstation has POW'R built in, and I mostly use Type 2, but sometimes Type 3.

David L. Rick
Seventh String Recording

[JEGG]

Originally Posted by Westmalle

Hi folks,

When you have a 44k/24 bit master, that needs to be converted to 44k/16bit, is it always necessary to add the appropriate dither?
The inherent noise of the recording I'm doing is on a constant level of -60dB.

Maybe it is good enough to add dithering to the fade ins and fade outs?

No, you don't always need dither-mostly you won't need dither if you're recording in a relatively noisy room-which includes most classical venues.

The issue will only be apparent with fade ins and outs. Here, simply check your recording. I have not yet had any problem with fade in and outs, INCLUDING raising dynamic level AFTER the fade ins and outs prior to truncating.

You know that with dithering you are still just truncating the file, right? You're just adding dither for the new bit length.

Is there an advantage to not dithering? Yes. Different forms of dither produce audibly different results. That's why there are so many types of dither. Therefore, dither is not neutral.

Are there disadvantages to not dithering? Yes, there can be. You already know what those are.

So what are you to do: Try it!

And if you have to crank the gain on your monitors up by 40 dB to hear the difference, then you really know you don't have to dither.