Discussion in 'Mastering' started by beyarecords, Mar 13, 2005.
Which is better to use when bouncing files down to lower bit/sample rates and why?
Whichever one sounds better on the recording that you're working on.
Well, that's how it works.
I've found there's such a huge misunderstanding about dither in general. I've seen posts where people seem to think dither choice significantly changes the overall tone of the music and this simply isn't the case (to my ear at least).
What dither does:
By adding a tiny bit of low level noise to the least significant bit it stops the audio from fitzing and burbling out as the audio fades to nothing - which happens if you just truncate when converting a file from 24bit to 16bit. A great explanation of this is at Nika Aldrich's site -
and at Izotopes' site -
There are some comparison sound files that might be of use to you to check out at
There are a few basic criteria as to how to judge what dither is best to use:
1) it should smooth out any re-quantization distortion so that the least significant bit areas sound as close to the original hi-res source as possible
2) the added noise should be as quiet as possible
3) what noise is still audible should sound "pleasant"
The way I've tested dither choices is to take various 24bit sound files that are gain staged down for their duration to the areas around the least significant bit, cranked my DAC and power amp full out, and then used my DAW's ability to toggle truncating output to 16bit on and off to hear how they sound when truncated, and then apply various dither choices to listen to what seems to work best based on the above 3 criteria. Since what constitutes "pleasant" (criteria 3) is a subjective thing - what dither choice is "best" will vary from listener to listener. You'll need great monitors & DAC, a quiet amp, and a quiet room, to do this testing accurately.
off subject, when you get a chance, can you drop me an email. thanks
"I've found there's such a huge misunderstanding about dither in general. I've seen posts where people seem to think dither choice significantly changes the overall tone of the music and this simply isn't the case (to my ear at least)."
yeah.... that's maybe because it does..... :lol:
maybe not the whole "tone" of the music.... but the quality of the higher frequencies......
it think that pow-r 3 makes the highs sound more like plastic...
This is not actually the best way to test dither. The problem with this method is that it undermines the way in which many of these algorithms are supposed to work.
POW-r, for example, plays with the concept of the threshold of hearing. You have a threshold of hearing for all frequencies (i.e. Fletcher-Munson) but the threshold is not the same for all frequencies. At 4kHz the threshold can be as low as -8dB SPL (you can hear a 4kHz sine wave at as low as -8dB SPL) and at 16kHz the threshold can be around 50db SPL and at 30Hz it's around 60dB SPL. This means that you can't hear signal present at 30Hz unless it is present at a very loud amplitude.
With this in mind, why make the noise floor produced by dither flat and the same at all frequencies? Why not contour it such that it falls below the threshold of hearing at all frequencies? We can put a lot more noise in in the low and high frequency ranges than we can at 4kHz, so if we take some of that noise that would have been at 1-4kHz and instead substitute it with noise in the low and high regions we can maintain a noisefloor that is lower than the threshold of hearing for all frequencies.
The problem is that the designers of these algorithms do this dependant upon a certain listening level. They assume that the levels will be set at traditional, audible listening levels, such as 0dB FS correlates with 95-100dB SPL. With the levels set this way (actually much higher than most listening) the noise-floor introduced by the very effective dithers will fall below the threshold of hearing for all frequencies and will thus be completely inaudible.
If, however, you crank your circuit path such that 0dB FS is equivalent to much greater than 100dB SPL then you are defeating the psycho-acoustic principles that allow these algorithms to work. Suddenly you will raise the dither noise at some frequencies above the threshold of hearing for those frequencies and they will become obvious. If you do this you will notice a coloration to the material for those frequencies where the noise level crosses the threshold of hearing first. Different dither/noise-shaping algorithms do this differently, so different ones will have different frequencies become noticeable first. If you do the test this way, that is what you will be comparing. This is not, however, useful because this is not representative of actual listening levels and the signal will therefore not actually present what you are hearing.
The best way to try dither algorithms is as follows:
Calibrate your system like you always do.
Use good monitors
Use a quiet room
Put different dither choices on the material.
Listen for differences WITHOUT unduly raising the level of the performance (by, for example, normalizing the results, null tests, cranking the amps, etc).
Choose the one you like most.
If you can't determine the difference then the dither is doing its job. Move on to the next project.
You bring up very valid points that when evaluating dithers it is crucial to listen to the material at normal listening levels using the various dither choices so that you can perceive what it is sounding like in the circumstances that the music will actually be listened to in. This I completely agree with you on.
However - I stand by my statement that putting these things under the microscope so to speak is an enormously educational experience - and in essence allows you to hear what each one of these is actually doing. ie. - If a highly noise shaped dither happens to produce an annoying whine similar to having a tv set on - even if this happens barely audibly - it helps to know that when using it we're introducing these things (even if enormously subtly) into the audio. Or being able to hear that some dithers actually seem to smooth the fitzing out as things disappear into the least significant bit more than others - not something that is always easily perceivable until you crank the fades up. This is not to say that we should ignore the forest for the trees by not listening to these things at normal playback volume - just that we can not discount examining closely what these soft little noise makers are actually doing down there.
Good points. Of course the problem is that what you are describing only happens when you artificially raise the level in most of these algorithms. What suddenly gets exposed as a high pitched whine may very likely ONLY surface if you push the level outside of normal boundaries (effectively "breaking" the system) and allow that region of noise to suddenly cross the threshold of hearing. If it is creating the whine such that it will cause a problem then we will hear it at reasonable levels. If we can't hear it at those levels in our controlled listening environments on good speakers... then our clients nor the audience will either. The problem is really CREATING problems that don't otherwise exist by using the equipment improperly and then claiming that these problems are a part of the normal operation of the equipment.
There have been many very interesting and informative posts in this thread. But I'd like to hear some comments from you ME's on they original subject of this thread. UV22HR compared to POW-r. I have both and don't notice a difference between the two, but then again I mix on BX8's in my bedroom which I'm sure is an acoustically horrible place for critical listening.
So how does UV22HR compare to POW-r?
Separate names with a comma.