I regularly make stereo location recordings of orchestras, using Cool Edit Pro for editing and mastering. If I simply normalise my raw recording, when it is played back in a typical domestic environment with average background noise, it's necessary to turn up the volume for the quiet bits and then turn it down for the loudest sections.
So, for each movement of a piece I will look at the level of all the sections and boost the quieter passages by a fixed small amount, and reduce the level of any extreme peaks in the loudest parts. Then, by checking what headroom there is at the loudest part, I can amplify across the whole performance to achieve maximum volume whilst maintaining dynamic integrity between movements. I don't believe a stand-alone stereo compressor can achieve this effect.
Are there any rules of thumb that can be used in the adjustments in level that can be applied to passages with extreme dynamics? How do the producers for Deutsche Grammophon, Sony, and so on do it? I am not aware of having to adjust the volume in regular listening conditions when playing back commercial classical recordings.
Technical Editor Hugh Robjohns replies: In days of old you would ride the faders as the piece was being recorded, following the score to anticipate when you need to start pulling the fader down (to control the level of a loud section) or push it up (to lift a quiet section). These days, with 24-bit recording, this kind of adjustment tends to be made at the editing/mastering stage rather than during recording. It sounds like you're doing the same thing in Cool Edit Pro.
As far as compression is concerned, I would suggest some very gentle overall compression, typically with a ratio of 1.1:1 and a threshold around -30dBfs, but you'll need to adjust this to suit the amount of squash you require. This will reduce the overall dynamic range but maintain some dynamic variation. Be careful choosing your attack and release times to prevent any pumping effects. A multi-band compressor will make this easier.
Another popular technique exclusive to classical music is the parallel compressor. For this you need a compressor with a huge headroom margin and the ability to apply a lot of gain reduction (well over 20dB). The idea is to split the signal in two, with one path going to the mixing desk, and the other going to the compressor. The output of the compressor is then routed to the desk and mixed with the direct signal. Set the compressor to provide around 20dB of gain reduction when the music is at its loudest, and set the return level so that this compressed signal is about 20dB below the direct signal's peak level. Set attack and release times to prevent pumping. The idea is that loud passages won't be altered significantly because the compressed signal is so quiet in comparison. However, as the input signal falls in level, the compressor applies less gain reduction and so its contribution becomes more significant, raising the low level signal. With input signals of -20dB, the compressor won't be compressing at all, and so will combine at the same level as the direct signal, providing a 6dB overall lift. Obviously, you can play around with the compression ratio and threshold to vary the amount of lift you require. It is a very subtle and effective technique, but is a bit fiddly to set up.
Finally, if you're using a digital compressor, don't forget to offset the direct signal to correspond to the delay caused by the compressor's A-D/D-A conversion process, otherwise you will encounter phase problems.
I know that the white keys on a keyboard form a diatonic scale, but what does diatonic really mean?
SOS Contributor Len Sasso replies: To understand the meaning of diatonic, it helps to think of a scale not as a collection of notes, but rather as a series of intervals. The definition of a diatonic scale is that there are five whole-tone and two semitone intervals in the series and that the semitones must always be separated by at least two whole-tones. Using '2' to symbolize the whole-tone steps and '1' for the semitone steps, the major diatonic scale corresponds to the interval series 2212221. No matter what note you start on, following this prescription yields a major diatonic scale — the white keys starting on C is one example. It turns out that all possible diatonic scales are constructed by starting somewhere in the major diatonic scale and continuing until you reach the same note you started on. Those are generally referred to as the church modes: Dorian for 2122212, Phrygian for 1222122, Lydian for 2221221, and so on.
While the preceding definition is correct and functionally useful, it might leave you a little cold, as it does nothing to explain why those intervals are used or why the seven notes in a diatonic scale are chosen over the other notes in the 12-tone equal-tempered scale. For reasons deriving from the physics and maths of sound, the strongest harmonic relationship aside from the octave is the perfect fifth, which makes G the closest relative of C, for example. Since C stands in the same relationship to F as G does to C, it makes sense that a scale centered around C should contain both G (called the dominant) and F (called the subdominant). The next closest harmonic interval is the major third. Together, the root, major third, and perfect fifth constitute a major triad, and it's not too big a stretch to imagine that you might want to construct a major triad on the three notes C, F, and G. Do that and you have the seven notes in the C diatonic scale.
There's still the question of why there are five other notes in the 12-tone equal-tempered scale, and the answer contains a hidden but important compromise. You can make music, which is naturally called diatonic music, with just the seven notes of the diatonic scale. And if you did that, they would in fact be slightly different notes from the ones you find in the equal tempered scale. If you want to expand the system to accommodate diatonic scales in other keys, one natural way is to iterate the process of adding perfect fifths. This produces what is commonly called the cycle of fifths, but is actually a spiral of fifths that never really comes full circle. But if you make the perfect fifths just slightly flat, they do come full circle after 12 steps. Miraculously, you also wind up with notes that are close to the major thirds — they're a little sharp and a little more out of tune than the fifths, but still useable. This compromise gives us the 12-tone equal-tempered scale (equal tempered meaning all the intervals are the same). Relative to C, the extra five notes turn out to be where you find the black keys on the piano keyboard, and that's why the intervalic definition we started with works.
Lately, there seem to be numerous affordable hardware compressors on the market, and I've noticed that many of them (the Platinum Focusrites and the Joemeeks, for example) are described as optical compressors. What's the difference between optical compressors and other types of compressor, such as VCA, FET and valve compressors? Are there any relative merits to these different types of compressor and are they suited to any particular applications?
Editor In Chief Paul White replies: After microphones, nothing stirs up a group of music professionals so much as a discussion about compressors. Essentially, compressors are gain-riding devices that monitor the level of the incoming signal and then apply gain reduction in accordance with the user's control settings. Given this simplistic explanation, shouldn't all compressors sound exactly the same, in the same way that faders tend to?
Clearly compressors don't all sound the same, and there are a few good technical reasons why. Perhaps of less importance than some people might imagine is the gain control element itself, which can be a tube, a FET (field effect transistor), a VCA (voltage-controlled amplifier), an optical photocell arrangement (a light source and a light detector) or even a digital processor. Certainly all these devices add their own colorations and distortions to a greater or lesser extent, but what influences the sound most is the way the ratio and envelope characteristics deviate from theoretically perfect behaviour.
In an imaginary, perfect compressor, nothing happens to the signal until it reaches a threshold set by the user, after which a fixed compression ratio is applied. For example, if the compression ratio is set at 4:1, for every 4dB the signal rises above the threshold, the output rises by only 1dB. A modification to this is the soft-knee compressor where the ratio increases progressively as the signal approaches the threshold, the end result being a less assertive, less obtrusive form of compression.
Many classic designs don't in practice act like this perfect compressor however, as their compression ratio may vary with the input signal level. For example, some compressors work like a perfect soft-knee device until the signal has risen some way above the threshold, then the compression ratio reduces so that those higher level signals are compressed to a lesser degree than signals just above the threshold. The reason for this change in ratio is simply that many early gain-reduction circuits don't behave linearly, especially those using optical circuitry as the variable gain element. The components themselves are non-linear so when, for example, you combine a non-linear light source with a non-linear light detector, the composite behaviour can be quite complex and unpredictable — however, history has buried those optical circuits that didn't sound good, so we're now left with those that happened to sound musical.
The other very important factor governing the sound of a compressor is the shape of the attack and release curves. While a modern VCA compressor can be made to behave in an almost theoretically perfect way with a constant ratio and predictable attack/release curves, many of the older designs had very strange attack and release characteristics, and, in the case of optical compressors, this was originally due to the relatively slow response of a light and photocell compared with a VCA.
For example, the now legendary Universal Audio 1176 combined a fairly fast attack time with a multi-stage release envelope. Conversely, the Teletronix's LA2A's rather primitive optical components resulted in a slower and quite non-linear attack combined with a release characteristic that slowed as the release progressed. Indeed, perhaps the reason the traditional opto compressor has so much character is that there are so many places in the circuitry that non-linearities can creep in.
Having said that, some modern optical compressors use specialised integrated circuits that incorporate the necessary LED light source (which has largely taken over from the filament lamps and electroluminescent devices used in early designs) and detector element in a single package that incorporates feedback circuitry to speed up the response time and to linearise the gain control performance. Indeed, some of these are so well behaved that they can sound almost like VCAs, but using clever design, it should be possible to recreate the old sounds as well as the new using contemporary electronic devices, or imaginative software design come to that.
It's harder when it comes to saying what type of compressor is best for which job, but in very general terms, a well-designed VCA compressor will provide the most transparent gain reduction, which is ideal for controlling levels without changing the character too much. However, a compressor that allows high-level transients to sneak through with less compression can also sound kinder to material than one that controls transients too assertively, which is why some of the older, less linear designs sound good. That's not to say modern designs can't sound good too though — Drawmer pioneered the trick of leaking high frequencies past the compressor to maintain transient clarity while other manufacturers, such as Behringer, use built-in transient enhancers or resort to equally ingenious design tricks.
Optical compressors, especially those that don't use super-well-behaved integrated optical circuits (or those that use them imaginatively) usually impose more of their own character on the material being treated, making it sound larger than life. In this context, the compressor is as much an effect as a gain-control device, and such compressors are popular for treating vocals, drums and basses. The Joemeek and TFPro compressors fit this 'compression as an effect' category as they use discrete LEDs and photocells in a deliberately non-linear topography that's really a refinement of that used in some vintage designs.
Digital compressors and plug-ins can reproduce the characteristics of vintage classics, but only if the designers successfully identify those technical aspects of the original design that make it sound unique. If they don't, you end up with an approximation or caricature rather than a true emulation.
With so many brand-new budget-priced microphones out there these days, I'm wondering if it might be better to get a high-quality mic second-hand instead. I don't think I'd trust second-hand monitors, but does the same apply to mics? For the £150 or £200 I'd spend on a new mic, I could get a far better model second-hand.
SOS Forum Post
Technical Editor Hugh Robjohns replies: I would certainly support the idea of buying second-hand pro audio gear, but I would be very wary of buying gear in the price range you're talking about second-hand. My reasoning is as follows. Firstly, users of high-end professional equipment generally know what they are doing and so their equipment tends (with the inevitable exceptions) to have been reasonably well maintained. It's not always the case, but you can usually tell in an instant by looking at a piece of equipment whether or not it has been well looked-after, and if it looks OK it usually is.
Secondly, bona fide pro gear can be serviced and repaired. All the reputable speaker manufacturers will happily supply replacement drivers, and all the reputable mic manufacturers will be able to repair and recondition their microphones. So even if the gear has had a hard life, it will remain perfectly serviceable. You can still get spares for 30-year-old Studer tape machines, for example.
Conversely, budget audio equipment is made as cheaply as possibly, and while you can get remarkable quality for your money, most of it is not cost-effective to service — in other words, it is disposable. The current glut of Chinese-made mics offer exceptional value, but you certainly won't be able to get them repaired in the factory after 20 years like you can a Neumann, AKG or Sennheiser. Likewise, getting spare parts for a Fostex multitrack tape recorder is a lot harder than for an old Studer or Otari.
So, if the second-hand budget gear in question is in good condition and very cheap, then it may be worth the risk, but go into it with your eyes open — it may well prove impossible or prohibitively expensive to have this kind of gear repaired should it fail a week after you bought it. On the other hand, a second-hand truly professional product should remain serviceable for decades. I bought four Sennheiser MKH20 mics second-hand a few years ago, and one turned out to be faulty, but it was serviced by Sennheiser and came back like new, and, even adding the cost of the service, it was still a very good deal compared to the cost of the mics brand-new.
I am currently setting up a home studio, which I'm hoping to eventually turn into a professional facility, based around a Soundtracs Topaz desk, three Egosys Wamirack soundcards and a Pentium 4 PC, with numerous synths, samplers, effects and other outboard gear. I'm now looking to wire everything together using patchbays. Bearing in mind that my console does not accomodate balanced outputs and insert points (the only balanced connections on the console are at the input stages of all channels and the effects returns), can I use unbalanced patchbays, thereby simplifying the patch lead requirements? If you are going to suggest a balanced patchbay setup, could you describe where to connect and disconnect the ground/screen connections to avoid ground loops.
Reviews Editor Mike Senior replies: It sounds like you've already invested a good deal of money in the gear, and there's certainly enough there to produce high quality audio. However, if you're going to retain audio fidelity with so many pieces of equipment working together, I would try to balance as many of your analogue audio cables as possible. Even in my more modest home setup mains hum and induced noise are problems (which have taken upgrading to balanced connections to sort out), so if you're ever hoping to use your studio professionally you don't really have a choice. Even in commercial studios a lot of time can be spent dealing with hum, so it's worth planning for it now, in my opinion. Unbalanced connections are fine for a smaller setup than yours, but, at the stage you're at, I reckon it's a recipe for disaster.
The great thing about balanced connections is that lifting the earth connections between equipment to break earth loops is comparatively easy — just disconnect the earth wire at one end of the signal cable — but with unbalanced gear the same trick very rarely works in practice and will often make things worse. If you're wondering how to decide where to make this disconnection in your system, Mallory Nicholls suggested that his preferred method was "to connect cable shields at equipment outputs and not at equipment inputs" in his Studio Installation Workshops in SOS September and November 2002 (www.soundonsound.com/sos/sep02/articles/studioinstallation0902.asp and www.soundonsound.com/sos/nov02/articles/studioinstallation1102.asp). So, disconnect the shield just before it reaches the equipment inputs. If you're using any moulded cables, then you might have to perform some modification on the patchbay, but this is not usually too difficult to work out — it's what I did, and it's worked very well so far!
To incorporate any unbalanced devices within the balanced system, you have two main choices: unbalance at the input to the unbalanced device — connect one of the balanced signal wires to the jack sleeve, along with the earth wire, and don't disconnect the earth wire elsewhere — or use a balancing transformer to do the interfacing. The second solution is more costly, but may be the only way to solve any hum problems which the first solution may create. Maybe you'll be lucky and not get any appreciable hum using the first system, but if you do get hum then have a look at the Ebtech Hum Eliminators we reviewed in SOS March 2003 (www.soundonsound.com/sos/mar03/articles/studioessentials.asp) — there's an eight-channel one for £295 which would probably isolate enough connections to sort remaining hum problems out. I've only needed to use a two-channel one to sort out a persistent hum in my system, but yours is much more complex, and all of it will be connecting to the central desk, which multiplies the potential for hum.
I've been recording vocals using a Neumann TLM103 mic going through a Dbx 386 tube preamp, and using the Dbx's converters to send a digital signal into a Roland VS1680 multitracker. I understood the Dbx was virtually impossible to clip, but experience proves otherwise! Firstly, it's impossible to use the Dbx's 'Drive' tube emulation above its lowest setting without getting obvious red light peaking and distortion for any louder transients during a vocal take (I like to sing fairly close to the mic). Does this mean I'm not getting any tube warmth from the unit? Generally, due to this problem, I always use the 20dB pad which enables me to crank up the Drive dial a little, but not much. What is the purpose of its higher incremental notches if you can't really use them? Even with Drive set all the way down, and the digital metering on the output stage peaking between 12 and 16dBu but avoiding the red light district, there are still obvious frequencies in my voice which cut through the supposed soft limiting facilities of the Dbx type IV converters to produce distortion. Sometimes I have to do drop-ins of single vowels, vainly trying to grab a clean one at a comparable level to its neighbouring words. What am I doing wrong?
Reviews Editor Mike Senior replies: I own a Dbx 376 and use it for all my vocal recording, and I'd suggest that you definitely don't want to be lighting that input Peak LED — that lights when the input is clipping, and clipping is quite a different thing to valve warmth. Given that your TLM103 has a fairly high output level of 21mV/Pa, if you're giving your performance a bit of welly close up to the mic then you may well find that you have to have the input gain all the way down.
I also work very close to the mic — like you, I have the Drive control all the way down for most of my louder numbers. This isn't a problem, though — you're still driving the valve, simply by dint of the raw level coming from the mic, it's just that you don't have to add any gain on the Drive control to do it. The valve 'sound' for recording purposes is very understated in quality equipment, and you don't need to try too hard to get the benefits of the valve — you'll get all the warmth on offer just by running the valve comfortably within its normal working range. You don't need to overdrive the valve, as you would in a guitar amp.
You also asked what use the upper notches of the control were if you always sang too loud for them. The reason for having them is so that low-output mics, such as dynamics and ribbons, can also be boosted into the optimum operating range for the valve. Think of the Drive control more like an input gain control, and that should clarify things a bit. I'd also be tempted to leave the Pad out unless it's absolutely necessary — it'll just be adding extra components into the signal path, and that's not necessarily desirable.
So, if you're setting up your Drive control right, there remains the question of the gain management in the rest of the chain. The first thing to realise is that it is possible to get nasty distortion out of the Dbx Type IV compression if you push it too hard, even if you don't theoretically get digital clipping. The best tactic, in my opinion, is to treat the converter just as you would any other and leave plenty of headroom. In this case, without compression, the majority of the signal will probably be hitting the -16dBFS mark, although this depends on your own performance dynamics. The most important thing is that you try to avoid making the -4dBFS light come on at all. Set the channel up while rehearsing so that only the -8dBFS light ever comes on. Because of the way in which the Type IV conversion process works, the moment the -4dBFS light comes on, the converter is effectively limiting the signal, so if (once you've set things up) you cook things a little hot in the middle of a take and the -4dBFS light comes on, you'll only be limiting the spikiest peaks. Type IV is great at peak limiting, but that's all it should be used for — use a compressor to reduce the dynamic range if necessary. Your description of your metering levels ("the digital metering on the output stage peaking between 12 and 16dBu but avoiding the red light district") shows me that you're running the output too hot: the 12dBu and 16dBu lights correspond to the -8dBFS and -4dBFS lights when the meter is switched to read the digital level, so if these are coming on most of the time then you've strayed too far into the danger zone. Also, bear in mind that even the digital output metering in the Dbx 386 is analogue, so the real peaks in your audio signal will probably extend beyond the meter reading. And because of the Type IV process, the output meter will only hit the 0dBFS light if it's seriously abused, so just avoiding the red light does not necessarily guarantee clean audio.
If you're getting distortion through the Roland VS1680 even on unclipped material, double-check that Dbx's sample rate is set correctly and that you're clocking the VS1680 from it — if the Roland is set to run from its own internal master clock then you may encounter a variety of strange spits and pops.
When digital and analogue gear is used in the same system, setting up the gain sensibly throughout the recording chain can be a bit of a minefield. However, it's worth taking the time to get it right, because otherwise all your recordings will suffer. You certainly shouldn't have to be dropping in words to avoid clipping — that's something you should be doing for artistic reasons to get the best possible performance.
Please can you explain the difference between 'soloing' a channel and using the other buttons marked 'PFL' and 'AFL' to listen to it. They seem to do very similiar but different things. Enlighten me!
Technical Editor Hugh Robjohns replies: The PFL, AFL and Solo buttons found on the channel strips of professional mixing desks can be confusing if you're unfamiliar with their uses, not least because different manufacturers have different names for, and different ways of arranging these functions.
PFL stands for Pre-Fade Listen. It allows you to monitor the channel in question's signal level at a point immediately prior to the channel fader, and will therefore include any EQ or dynamics that might have been applied on that channel. Thus when setting up a channel's input gain using PFL, it's important to bypass any EQ and dynamics processing, otherwise you won't know what the actual headroom is at the front end. On mono channels, PFL is mono. On Stereo channels PFL should be stereo, but some cheap desks derive a mono PFL signal for both mono and stereo channels.
AFL, which stands for After-Fade Listen, is similar to PFL in function, but takes its signal from a point immediately after the channel fader, showing the level of the channel's contribution to the mix. AFL is also mono on mono channels.
Solo, more correctly known as Solo-in-Place (SIP), is an after-fade listen taken from after the pan control as well as the channel fader. It is therefore a stereo signal even on mono channels. The idea is to allow the monitoring of a channel signal when panned to its appropriate position in the stereo image. SIP is usually achieved by monitoring the main mix buss and muting all the channels other than the one you pressed the SIP button on. However, this means that you can't use SIP while mixing because it destroys the mix on the mix buss, muting aux channels as well as main channels. (PFL and AFL only affect the signal routed to the monitor outputs.) That's why SIP is often described as 'destructive solo monitoring'. Usually, you'll want to solo a channel and hear it with any associated effects returns, so selected channels can usually be made 'safe' from the SIP function, so that they continue to contribute to the mix when all the other channels are muted. A lot of desks have a single 'solo' button somewhere near the fader which can be configured to provide any or all of these functions.
Is there any real difference between morphing from one sound to another and crossfading? In many cases, the two sound very similar.
SOS Contributor Len Sasso replies: Morphing and crossfading are really two entirely different processes and apply to different situations. Crossfading takes place between two audio files, typically non-destructively in a sequencing environment or destructively in a sample editor. The effect, of course, is that one sound fades out as the other fades in. Morphing takes place between two groups of settings for an audio device, either hardware or software. In that case, one sound also dissolves into another, but the intermediate sounds are not simply a mix of the starting and ending sounds.
If you have sequencing software and a synth plug-in that can be automated, here's an experiment to quickly convince yourself that there really is a difference. Set up a basic oscillator and lowpass-filter patch without any envelope applied to the filter cutoff. Record rather long clips with the filter wide open, then with it relatively closed (but with the oscillator still audible). Now crossfade between the clips over a fairly long period. Next use animation to slowly sweep the filter cutoff across the same range. Compare the crossfade with the animated filter, which amounts to morphing between the open and closed states. Of course, morphing usually involves many more parameters, and the results are correspondingly more complex and interesting.
Cubase SX has a Detect Silence feature and a Hitpoints feature that seem to me like they do pretty much the same thing. Are they different, and when should I use them?
SOS Contributor Len Sasso replies: Yes, those features are really quite different. Silence-detection software, of which SX's Detect Silence is an example, works by searching for areas in an audio file where the level drops below a given threshold for a specified minimum amount of time. For more detail about its uses, see the Q&A on extracting loops in SOS June 2003 (www.soundonsound.com/sos/jun03/articles/qa0603.asp). Beat-slicing software, of which Hitpoints is an example, works by searching for abrupt changes in an audio file, presumably signalling individual events such as drum hits in a percussion file.
As the names imply, one of the main differences between Detect Silence and Hitpoints is that Detect Silence marks and optionally removes whole sections that it interprets as 'silence' (threshold, duration and other criteria are typically user-definable), while Hitpoints marks individual points in the audio file, the theory being that the region between two points represents a single event. In a simple audio file, say a kick-drum part, Detect Silence could be used to separate the individual kicks, but in a busy percussion part, it's unlikely that individual events are separated by silence. Detect Silence would be no help in that case. On the other hand, Hitpoints would be little help picking out regions separated by silence in an audio file (such as a track from an audio sampling CD) because each region would most likely contain a large number of hitpoint markers, but none at the boundaries of the silent portions. In short, use Hitpoints for slicing up audio files into individual events and use Detect Silence for identifying (and possibly removing) low-level regions in an audio file.
I've heard that with Windows XP, old versions of Norton Speed Disk do not work and that the latest version of Speed Disk included in Norton Utilities has a number of features removed and is no better, if not worse, than the defrag utility that Windows XP comes with. I'm about to move up to Windows XP to take advantage of Cubase SL, and have used Speed Disk in the past in favour of the defragger that came with Windows 98 because of the huge speed difference.
SOS Contributor Martin Walker replies: It's not just a case of older versions not working — when moving to another operating system it always pays to be careful with utilities such as Speed Disk that work at a low level, since they have the potential to do some serious damage to what is, after all, unknown territory for them. If you're intending to update an existing Windows installation to XP, you should first investigate when XP-compatible updates or upgrades have been released for all such utilities, and, if so, you should install these before XP; otherwise you should uninstall them altogether.
As for the speed difference, there are two issues — the time it takes to defragment a drive, and how much faster it operates afterwards. I discussed the various approaches to defragmentation most recently in SOS April 2003, and although using the Disk Defragmenter bundled with Windows 98 was like watching paint dry, personally I've found the bundled Win XP defragmenter to be very effective. It has a handy Analyse function that lets you view how fragmented the drive contents are before committing yourself, and once you do, it's far quicker to use.
It can also optimise disk layout for both boot and application launches, laying out data in contiguous chunks to eliminate excessive head seeks. If you have Task Scheduler enabled it will start optimising after the first boot, so your second boot should be faster. However, while Windows XP is constantly fine-tuning file positions on each boot, Microsoft estimate that 90 percent of the optimisation is done in the first couple. Even with Task Scheduler disabled you can run the new Defragmenter by hand to achieve the same improvements.
The only other advantage of Speed Disk in the past was its ability to place the swap file (or page file) outermost on a partition to provide the fastest performance, but with many musicians now routinely using 512MB or more of RAM, this becomes less important, especially as some people are routinely disabling the swap file altogether.
Although I was an enthusiastic user of Norton Utilities 2000 running on Windows 98SE, I've not personally upgraded to the latest 2002 version compatible with XP, largely because XP's own utilities all seem so much more effective than in previous versions of Windows, and because XP itself seems to need far less tweaking for audio purposes than its predecessors. Why not install Windows XP and see how you get on before deciding whether or not to invest in a suite of utilities?