If you're serious about digital audio you need to know about dither - both what it is and when to use it. So here it is: the SOS Digital Performer dither primer.
Dither is one of the trickiest concepts that we digital audio users ever have to get our heads round. It's inextricably tied up with the maths that lie behind digital systems, so that puts plenty of people off from the start! It also has the whiff of high-end hi-fi hogwash about it — like those £500 kettle leads. That said, it's important to know about it and how Digital Performer (or Audiodesk, for that matter) applies it.
The need for dither arises when you reduce the resolution of a digital recording, the most common situation being when you're working on a 24-bit DP project and need to make a 16-bit bounce of it, ready for burning to CD. If the bounce is done without dither, the last eight digits of every 24-bit sample is simply discarded, along with all the precious and subtle audio information they contained. You'll still end up with a playable, recognisable 16-bit file, but one which doesn't sound as good as it could or should, and quite possibly has acquired some grainy distortion, particularly on low-level signals.
Dither is a process, then, that aims to preserve the sonic attributes of a high-resolution recording when it's converted to a lower resolution, like the 24-bit to 16-bit bounce just mentioned. It does this by adding an element of randomness into the resulting digital signal which we perceive as very low-level noise, and which not only 'masks' the distortions caused by the drop in resolution but actually manages the seemingly impossible trick of extending dynamic range in the lower-resolution format beyond theoretical limits. For this reason (amongst others), it's always worthwhile recording at 24-bit resolution in DP, even if your final goal is distribution of your work on 16-bit CD.
As a digital audio workstation capable of recording and playing audio at 24- and 16-bit resolution (and with 32-bit processing resolution) DP needs to be able to dither: and it can, in several different ways, all of which we'll look at in a moment. First, though, a word of warning. The otherwise commendable DP5 manual (and other DP manuals before it) is a little short on dither detail, and contains a mistake which can lead to much confusion. Again, we'll look at this in a moment, but until MOTU clear this up, or actually re-configure DP to behave as the manual suggests, it's best to ignore it and stick with this column instead!
In recent versions of DP, two dither plug-ins, Quan Jr and Master Works Limiter, are included as part of the standard plug-in bundle. Both utilise the same unspecified dither algorithm, can reduce bit depth right down to 1-bit (which is loud, so be careful!) and have switchable dither and noise-shaping. This makes them useful not only for making a typical 16-bit bounce of a 24-bit recording, but also for preparing audio for lower-spec devices such as 12-bit samplers or those grungy 8-bit telephone 'hold music' systems. Master Works Limiter is clearly designed as a final mix-bus limiter and dither solution, whereas Quan Jr is more suitable for 'utility' tasks, such as dithering incoming or outgoing ADAT transfers, or doing the kind of 'creative' bit-depth reduction heard in some glitch music styles, in which case dither is best left off.
More flexibility and possibly better audio quality is delivered by a few third-party plug-ins. Perhaps the best known, and something of an industry standard, are Waves' L1, L2 and L3 limiter plug-ins (www.waves.com), which incorporate their IDR (Increased Digital Resolution) dither algorithm. In fact, two alternative algorithms are on offer, along with variable noise-shaping curves and a range of preset destination bit-depths. Wave Arts offer something similar in the form of Final Plug, which has the snappily entitled Triangular Probability Density Function, dithering down to 4-bit resolution, and a range of different noise shapes (www.wavearts.com). More configurable still is the dither module in iZotope's Ozone 3 (www.izotope.com), which boasts, amongst its three dither algorithms, the latest version of the very highly regarded Mega Bit Max, which is arguably better sounding than other dither heavyweights such as UV22 or POWR. As well as the usual noise-shaping options there are also more sophisticated facilities such as auto-blanking, which turns off dither noise in periods of digital silence, and harmonic suppression, which aims to maintain audio quality when bit-depth reduction takes place without (or with very low levels of) dither.
These days the most common scenario where dither needs to be applied in DP is when you've mixed your 24-bit project and are about to make a 16-bit bounce or copy of it. This can be achieved in several ways, depending on how you like to work:
Firstly, there's DP's built-in 'faster than real-time' Bounce to Disk function. This handy function, residing in the Audio menu, presupposes that you've already rendered to disk any Virtual Instrument or external device tracks, and can then make a mono, stereo or multi-channel bounce of the entire mix in a variety of useful audio formats. Just make sure you've selected 16-bit as the resolution (or bit depth, as it's referred to for some of the audio formats).
Secondly, you could make a manual, 'real-time' bounce to another application. Here you'd utilise some sort of inter-application audio link (perhaps Cycling 74's Soundflower, or the open-source Jack) to route the output of DP to an input in another audio program, such as DSP Quattro or BIAS Peak, which is running at 16-bit resolution. You could also use a self-contained solution like Audio Hijack Pro.
Thirdly, your audio interface's digital outputs could feed a 16-bit hardware recorder, such as a CD or DAT recorder.
In all these cases, the steps needed for successful dithering in DP are the same:
- Create a Master Fader track (if you don't already have one) which is routed to your usual main hardware outputs, and set it to the 0dB (unity gain) position.
- Place any 'final' processing plug-ins you want to use on this, and then place your preferred dither plug-in in an insert slot beneath them. Set its bit depth to 16, and dial in your desired dither algorithm, noise-shaping and other options.
- Now play or bounce your sequence.
This method should produce a perfectly dithered 16-bit copy of your 24-bit sequence, with better low-level detail and arguably a better overall sound than could have been achieved if the actual project was recorded at 16-bit. However, it's important that the 16-bit file or copy that is produced doesn't end up being processed further — even for something as simple as a change of gain — as that would in a sense undo the usefulness of the dither process. Use it directly for producing a CD.
Now we're getting a touch nerdy, but hang in there, this is important! By default, all of DP 's insert slots are in a pre-fader position in the signal chain, and this works great for most mixing tasks. It can really screw up using a dither plug-in on a Master Fader Track, though, if the fader position is set to anything other than 0dB (unity) or if you've automated the fader, to do a fade-out at the end of a song, say.
The reason? Your 24-bit mix passes through the dither plug-in, gets properly dithered down to the lower resolution, but then gets processed further at a higher resolution again by the simple gain change taking place at the fader. This is a digital no-no: the dither must be the very last processing step in the chain.
There's a simple fix, though. Look to the left of the lowest insert slot on the Master Fader Track and you'll see a tiny 'handle'. Drag this upwards to just above your dither plug-in insert slot and you'll see a more noticeable divider appear. By doing this you've configured the dither plug-in (and other insert slots below the divider) to be post-fader in the signal chain. So now your dither plug-in is placed after the gain change occurring at the fader, just where it needs to be, and digital decorum is restored once more.
Lurking at the very top of DP 's Audio menu is a rather mysterious Dither command. By default it's selected — it appears in the menu with a tick next to it — and if you choose it once more you simply deselect it. What does it do, and what's being dithered?
The situation is confused. The DP manual says 'it makes Digital Performer use dither whenever it must reduce the bit depth of audio' and goes on to describe a 24-bit to 16-bit Bounce to Disk operation. However, a few simple tests reveal that DP never applies dither to a Bounce to Disk operation, unless it's been manually applied with a plug-in as described above. So when does this Dither command apply dither?
Very simply, in three situations. The first is when you have 'Automatic Conversion' enabled, by selecting the Consolidated Window's little 'lightning flash' button, and you import audio that is at greater than 16-bit resolution into a 16-bit project. Here, the imported audio will be dithered. The second is when you manually select a high-resolution audio file in the Soundbites window, and convert it to 16-bit using the mini-menu's Convert Sample Format command. Again, the converted audio is dithered. The last situation is a little odd. If you select a 16-bit soundbite in your sequence and then apply an audio plug-in via the Audio menu (ie. as an offline process), the resulting processed soundbite will have also acquired dithering. The most likely reason for this is because the soundbite is being processed offline at higher resolution, so dither is required to mask unwanted harmonic artifacts when the audio is truncated and returned to the 16-bit environment.
Should you leave Audio menu dither on or off? My advice would be to leave it on; it can't harm any real-time processing or a Bounce to Disk, and it's altogether useful for file conversions. There is one scenario where you'd want to turn it off, though. Imagine you'd done a bounce of some audio with a 16-bit dither plug-in applied, but this resulted in a 24-bit file; it's an easy enough thing to do. The dither plug-in would have processed the dither, added dither noise, and then made the last eight digits of every sample into zeros. Now you need DP to chop off those last eight redundant digits and write a 16-bit file suitable for feeding to a CD burning application. The Soundbite window's Convert Sample Format command is perfect for this task, but you need DP to only truncate the meaningless digits, not apply a second round of dither. In this case, deselecting the Audio menu Dither is the perfect option, and the soundbite can be converted without fear of 'double dither'.
One or two other situations will cause you to think about dithering. One is the digital transfer of a 24-bit DP project to a multitrack hardware recorder (like an ADAT tape recorder or hard disk-based workstation) which uses a lower resolution. Here, you'd hook up your audio interface to the external device using some sort of multi-channel digital connection, such as a 'lightpipe' optical cable. Then you'd place a simple dither plug-in like Quan Jr, set to 20- or 16-bit resolution as appropriate, on every audio track in DP. After making the correct routings between tracks in DP and tracks on the external device, you could start this recording, begin playback in DP, and make your digital transfer.
Finally, consider this. If you're preparing a mix to be mastered by someone else, you mustn't touch dither with a barge-pole, as this will be done further down the line. Instead, bounce your mix (or your stems) at 24-bit resolution, with no level-boosting compression or limiting, and leaving plenty of headroom. The mastering engineer will thank you for it.
In a stroke of bad timing that borders on the absurd, www.plasq.com has withdrawn from sale Rax and Wormhole, which I recommended as network MIDI and audio problem-solvers in this column in the March and April 2007 issues of Sound On Sound.
Plasq's line is that they're concentrating from now on on their visual-oriented software (which includes the very popular Comic Life application) and that they'll continue to offer support for licensed users of Rax and Wormhole for a short while to come. They also claim that rather than see the products wither away they'll consider selling them to another company, offering them as unsupported freeware, or releasing them to the open source community. As yet, though, there's no indication of which way things will go, and I personally hope that the future is not freeware, as that way neither might make it unscathed into Mac OS 10.5, which is just around the corner. Plasq currently have a discussion going on in their forum (www.plasq.com/forums), to which current and prospective users are invited to contribute, and if you feel strongly enough I'd also recommend emailing Plasq at www.plasq.com/contact.
Installers for the software are still available, but the demo version of Wormhole has been withdrawn. If and when there's movement from the current position, I'll let you know.