Came across this today, thought it may be of interest to some folks here. Everything this guy writes is worth reading IMO. He's owned his own self built mastering business for 15 years, and he's written a bunch of practical guides to mixing, mastering, etc. Tremendous free resource.
*****************************************************************
https://forum.ableton.com/viewtopic.php?f=1&t=214124A few years ago I produced some audio examples of different types of dither, so that people could more easily hear what dither does and what a couple of different dithering options sound like. As there are even more options for dithering algorithms these days, I figured it was time to update my examples and talk a little bit about what seems to be one of the more confusing aspects of music production for people. You can download all of the audio examples and graphs I’ll be talking about here:
http://innerportalstudio.com/new-dither-examples/For these examples, I used a 24bit sample of a ride cymbal with some reverb applied. I then converted this to 16bit wav files in various DAWs using the dithering options they offer. Specifically:
- Rectangular, Triangular, POW-r1, POW-r2, POW-r3 from Ableton Live 9.
- The only dithering option in Presonus Studio One.
- UV22HR from Apple Logic Pro X, though it also offers the same POW-r options that Live does.
- Goodhertz dithering from Audiofile Engineering Triumph.
In addition, I also created a 16bit wav file version using no dithering at all, this is called truncating.
The next step was to cut off the all but the very end of the reverb tails of these files, and normalize the remaining portion to -0.5dBFS. This was done because dither noise is extremely quiet, with all but it’s very peaks around -96dBFS, well below the noise floor of most playback equipment. Boosting only the tails of the audio files allowed me to raise the overall level of the files to make the dither noise itself audible at normal listening volumes. These files are located in the folder called “Dithered Ride Tails”.
I recommend listening to the truncated version first, so you can hear what it is we’re trying to achieve with dithering in the first place. At the very end of the truncated sample, you can hear what sounds like digital noise as the least significant bit toggles on and off trying to replicate the very quiet end of the reverb fading out to silence. By adding dither noise, we make this last little bit of fade out much smoother and more natural sounding, at the expense of a very tiny bit of noise.
Remember, in these examples I’ve boosted this noise A LOT just to make it audible, in normal use, it’s so quiet as to be almost completely inaudible. Plus there’s some tricks with dithering to reduce how much of it we hear even more, which I’ll talk about shortly.
I included the full length ride samples without trimming or normalizing as well, in case anyone wants them to hear how dither sounds in more real world situation. You’ll find them in the folder called “Original Rides”. Though I highly doubt that many people will be able to hear the dither at all, even on what is arguably one of these best examples for demonstrating it’s purpose. It’s just extremely quiet, just imagine trying to hear it on a full mix!
In addition to the ride cymbal sample, I also created a 24bit sample of nothing but silence. This was also converted to 16bit using the above dithering options, but in this case it was so I could provide FFT analyzer images of just the dithering noise itself for visual comparison. I used DMG Audio’s Dualism plug-in for the FFT analysis. The scale was set from 20Hz to 20kHz, and from 0 to -144dBFS (effectively 24bits) to make the shape of the dithering algorithms easier to see. Keep in mind that a 16bit file has only a range to -96dBFS when you look at the graphs, so anything below that will be discarded. All the graphs are unsurprisingly located in the folder labeled “Graphs”, and you can see them on my site:
http://innerportalstudio.com/new-dither-examples/Why are they shaped differently? That’s one of the tricks I mentioned earlier, and it’s called noise-shaping. Rather boring and predictable name, I know. Since our ear is most sensitive around the 2kHz range, the dither noise in the various algorithms is created to be stronger in the frequencies away from this sensitive area. Most of the time it’s boosted way up by 20kHz, beyond the range of most human hearing, but the actual shape and slope of the boost varies depending the algorithm.
Each manufacturer has what they consider the ideal way of doing this, sometimes, in the case of POW-r, with different options for different kinds of music. You can hear this in the subtle tonality of the noise in some of the different dither examples, as well as seeing the exact shape in the graphs I provided. Some of the options like Ableton’s Triangular and Rectangular dithers are almost perfectly flat, however that doesn't mean they are less effective.
Ideally this gives the producer the flexibility to choose the dithering that best suits their material on a song by song basis. But again, this noise is so incredibly quiet that for most music, you’ll never hear it. Which is ideal anyway, as dither was created to be as inaudible as possible in the first place. I’ll admit that as a mastering engineer, even I rarely audition different dithers, since with most material there’s no audible difference anyway.
Once in awhile I’ll get a very dynamic song with lots of quiet passages, certain ambient or even orchestral songs fit this category. In those cases I might try out a few different dithering options, though even then the differences can be almost impossible to hear, even in my studio.
The point of all this is make you realize that while dithering does fulfill a useful role in the audio production process, it’s arguably the least important aspect and isn’t something people should worry too much about. Certainly add dither if you can when you’re rendering your mixdown or master to a 16bit file at the end of the writing process, but don’t lose sleep over which dithering option is the best. The differences are incredibly subtle, even to those people with well-trained hearing, and in almost all cases the dither is so far below the noise floor of any playback chain that no one will hear what dither you used, or even if you used it at all.
I hope this helped you not only understand why we use dither, but also highlight some of the differences in the various options available to us.
Peace and beats,
Tarekith