You are here

Sonic Foundry Sound Forge 4

Software & DSP Plug-ins By Martin Walker
Published March 1997

The arrival of audio processing software plug‑ins for the PC means that the platform can now compete as a viable digital audio workstation, and the latest update to PC audio editor Sound Forge supports the use of several software‑based processors. Martin Walker checks them out.

During the last 12 months, the PC has seemingly gone from being a poor relation to joint partner (along with the Mac) in the hard disk recording hierarchy. With the advent of programs like Cubase Audio and Logic Audio for the PC, many people are now regularly recording up to eight tracks of audio alongside MIDI, with both types of data controlled from a single package. The quality of audio has always been somewhat dependent on the individual PC, since most budget soundcards tend to pick up a small amount of interference from other expansion cards in the PC, as well as from the power supply and hard disk.

As the quality stakes have gone up, so various manufacturers have responded with more specialist, higher‑quality soundcards. These employ much more thorough methods of shielding and interference suppression to give audio qualities that are extremely close to those attainable by stand‑alone units, where the recording technology is entirely removed from the PC box and placed in its own shielded enclosure. As people's expectations have continued to rise, a backlash has become inevitable, and despite many cautionary comments from industry professionals, every person who confidently starts out using an integrated multitrack hard disk recording system on the PC finds out soon enough just what the limitations are. Once you have more than two tracks (or a stereo pair) with only a single stereo soundcard to output data, it is impossible to add EQ or effects on more than a global basis when plugged into an external mixer.

It was probably at this point that many PC musicians looked longingly towards advertisements and reviews for Mac products, where software 'plug‑ins' offering effects processing and EQ have been available for some time, albeit at professional price points. The concept of plug‑ins is one that will be new to most PC owners, but as software applications grow ever more comprehensive, so the idea of an expandable core program (which provides basic features) with separate plug‑in software modules available (both from the manufacturer of the original program and from third parties) becomes extremely attractive. In the rarified areas of sound processing, it is impossible for a single company to be at the leading edge of all aspects of sound manipulation. In areas such as 3D positioning and psycho‑acoustic tweaking, the previous development of Macintosh plug‑ins means that the majority of the hard work may be done already for any manufacturer who decides to enter the PC arena.

Although Wave editors have been available for the PC for some years, and many have included special effects such as reverb, filtering, and echo, as well as the more standard items like normalisation and fading in and out, these have in many cases been little more than gimmicks, and whilst giving a great deal of pleasure to people dabbling with their soundcards, the audio quality was generally far worse than the average budget effects unit. Stand‑alone budget effects units have come on in leaps and bounds over the last year — you can now buy a rackmount device with an 18‑ or 20‑bit 'engine' providing excellent signal‑to‑noise ratios and overall sound quality for under £200, and in order to move your processing onto the PC, this same level of performance must be matched or exceeded. Although 16‑bit sounds have a theoretical 96dB dynamic range, once you start manipulating audio data, you have to be very careful not to lose information along the way which can accumulate and give rise to audible degradation. For this reason, professional sound editing always uses higher resolution internally (see the 'Bit of a Problem' box, below), only returning the sounds to 16‑bit after processing has been carried out.


Running 24‑ or even 32‑bit calculations is obviously more processor‑intensive (high‑end workstations sometimes work with internal resolutions of 56 or even 72 bits), and it is for this reason that DSPs (Digital Signal Processor) chips are often used. These are used to offload the bulk of the arithmetic slogging from the processor to a separate chip. Every time you select a plug‑in, the appropriate algorithm code is sent to the DSP, so that for each point on the waveform, a number is sent, and the result returns much faster than the equivalent calculations would have taken when sharing the CPU. By having a reserve of DSP chips at your disposal (the so‑called 'DSP farm' beloved of Mac‑based Digidesign owners) you can decide how best to use your total DSP power for each project. This sort of power was needed until recently, simply because computers had insufficient power to carry out audio processing as well as all of the other housekeeping, such as disk read/writes, updating the screen display, and running the other programs that are trying to cope with MIDI data at the same time.

When I first started researching this article in late 1996, the concept of PC plug‑ins was exciting enough — being able to add further functions to an existing program is certainly most attractive, and it is nearly always easier to work in an integrated environment than have to move to another application to perform a specific operation not available in the main program. However, without added DSP power, it seemed that none of the more exotic functions found in Mac plug‑ins could be implemented on the PC, and despite excellent 'level manipulation' programs such as the L1 Ultramaximiser from Waves, it seemed most unlikely that many real‑time processing engines would be available until the power of PCs took yet another quantum leap forward.

All this has now changed, firstly with the release of Steinberg's WaveLab v1.5 (using 24‑bit calculations but no DSPs), which includes a selection of eight plug‑ins, even including EQ and reverb (see Paul White's review in last month's SOS). For the first time, a live signal at the input of your PC soundcard can be treated in real‑time by professional‑quality computer algorithms, and it looks as if, within the space of a few months, the limitations of the integrated 8‑track audio workstation (namely lack of real‑time treatment for individual tracks when multiple outputs are not available) has finally been addressed.

The next revelation occurred when Sonic Foundry (the designers of PC audio editing software Sound Forge) announced an update to Sound Forge v4.0a, which is available to existing Sound Forge users to download free from the Sonic Foundry web site. This update includes Microsoft's Active X technology (itself a form of plug‑in), which was originally designed to add features to the Internet browser Explorer, and acts as a sort of 'glue' to allow extra bits to be bolted onto your PC system (for example, the subset ActiveMovie caters for video and audio streaming). Sonic Foundry's support of this technology has enabled their own plug‑ins to achieve real‑time previewing (as long as the host PC's processor power is sufficient), but more fundamentally, any plug‑in designed to work with Sound Forge will also work with any other application that supports ActiveMovie technology. This is exciting news indeed, as more developers will be encouraged to support Sound Forge, but also, plug‑ins will now be useable by sequencers, editors or even Internet browsers in real time! Here, then, are some mini‑reviews of the current range of plug‑ins available from Sonic Foundry, along with one from QSound.

Sonic Foundry Noise Reduction Plug‑In

This plug‑in offers two types of noise reduction. The first can perform wonders on material that has any form of fairly constant background noise, such as hiss, rumble (from nearby machinery), electrical or mechanical hum, or indeed any other constant sound that was not intended to be part of the recorded signal. I have heard CDs featuring quiet acoustic recording where the whirr of an ADAT can be clearly heard in the background — this is an ideal candidate for the Noise Reduction plug‑in. The process is as follows: first, choose a short section of the recording where only the offending noise is present (less than a second will be enough), and press the Get button. This generates a noiseprint — an analysis of what noise frequencies are contained within your selection. Then, an envelope is created by means of the Fit button, which creates a linked number of points above the noiseprint. This envelope is used by the noise reduction system — any signal below the points is treated as noise. A setting of 6dB above the noise level is recommended as a start point. Finally, by pressing Preview, you can hear the noise reduction in action, and compare it with the untreated version using the Bypass button. If all goes well, the offending noise will be typically 20dB down on the original (a default setting, although any setting from 0 to 100dB of reduction can be selected), and will be significantly easier on the ear. All aspects of the process can be manually tweaked, and I fully expect that in the hands of an experienced operator, dramatic improvements will be possible. This process is very similar to the DeNoise function used in Dart Pro (reviewed in SOS January '97).

The other function of this plug‑in is click removal, and this is designed for short discontinuities rather than continuous background noise. Options include Find (which searches through any marked area, stopping if anything appears that looks suspiciously like a click), Replace (which removes the offending portion, replacing it with similar data from nearby), and Interpolate (which effectively draws in a line of new data between the values at the start and end points, and which is only effective for very short 'ticks'). Even with the default settings, appreciable improvements can be made using both types of Noise Reduction. This sort of process is invaluable not only for rescuing mistakes, but also for restoring older material from vinyl records which have not withstood the test of time as well as they might. Sound Forge v4.0a runs 50% faster than v4.0, and Pentium Pros will now work in real‑time with 44.1kHz stereo files. At around £200, this plug‑in is not a cheap proposition, but is worth its weight in gold in the right circumstances, and is still significantly cheaper and faster than Dart Pro.

Sonic Foundry Spectrum Analyser Plug‑In

This plug‑in allows you to translate the normal level vs. time image into the frequency domain, so that you are looking at spectral content. It is extremely comprehensive — so much so that at first it seems a bit impenetrable, even to someone like me who has used both hardware and software spectrum analysers extensively in the past. The signal can be viewed in both the format of a more usual spectrum graph, or as colour and monochrome sonograms, in which the horizontal axis represents time and the vertical axis represents frequency. The amplitude of each frequency component in this latter case is represented by the colour intensity of each point in the graph. This method of displaying spectral information is useful for identifying distinctive spectral patterns created from sounds such as speech, musical instruments, and bird calls (apparently it is possible for experts to identify birds simply by studying sonogram characteristics). When I first started looking at this plug‑in, two main applications sprang to mind: firstly, the obvious one of examining mixes for their frequency content (to highlight problem areas where two instruments clash or for studying other people's work), and secondly, on specialist audio courses, as well as at college and university level, for instructive purposes.

However, during the time I have had this plug‑in for review, it has proved more and more useful. It has been used to examine the EQ characteristics of mixers, spot subsonic rubbish that flapped my cones but performed no useful function in a mix, and examine the acoustics of my control room. But with the release of the Sound Forge v4.0a update, a whole new world opens up — real‑time monitoring. Now you can carry out spectrum analysis on your mix in as much detail as you want, whilst the track is playing. This turns the plug‑in into a far more useful tool, and makes the price of around £100 seem far more reasonable. It is less easy to make pretty displays with Sound Forge's Spectrum Analyser than it is on the one provided as part of Steinberg's WaveLab, but once you get to grips with it, the Sound Forge plug‑in is a far more powerful and analytical tool, and is the only one I know of that currently works in real‑time.

Sonic Foundry Batch Converter Plug‑In

This plug‑in is one of those unexciting options that doesn't strike you as being worth the expense. But I've had people on the phone in the past who have been given a few hours to convert 800 Mac AIFF sample files to their WAV PC equivalents for a cross‑platform game conversion, and if this sort of situation ever applies to you, this plug‑in may prove invaluable. Although batch converters are nothing new, and indeed you can probably find several in the public domain if you search hard enough, the exciting aspect of this one is that you can subject each sound of your batch to many additional automated processes before it is saved in the new format. Besides the obvious ones, such as mono‑to‑stereo, bit depth, and sample rate conversion, you can change the volume, add compression, normalisation or noise gating, apply a 3‑band EQ with fully parametric mid range and subsonic filter, remove DC offsets, trim out leading or trailing end silences, and fade a selected number of milliseconds at each end to avoid clicks. There is also a useful Extract Regions option, so that you can work with many effects in one large file, give them global treatments, and then run the Batch Converter to automatically separate this file into a series of smaller ones whilst converting other parameters. The downside of this versatility is the price — at around £150, only the well‑heeled multimedia person will be able to justify it. If you only need to convert formats, then it may pay to look for something simpler and cheaper. However, if the alternative is hours of drudgery converting hundreds of sounds by hand, this will pay for itself the first time it is used.

Qsound Qtools Plug‑In

This is a Sound Forge plug‑in from a third‑party developer — the ubiquitous QSound Labs. QSound seems to pop up everywhere nowadays, so it's nice to see a version of this technology available to PC Sound Forge users. For the background to this 3D spatial treatment, look no further than Paul White's interview with QSound Labs in the November '95 issue of SOS. Here, as in many other versions available for different platforms, there are two main treatments on offer. For mono sources, QSYS allows you to 'place' your sound anywhere between or beyond your speakers, as a sort of 'super panning' control. As long as your speakers are symmetrically placed, it works extremely well (although it must be admitted that some people hear these sort of effects more than others). Once processed, you can download your file to a sampler to create a permanent library of QSound‑treated instruments (extremely useful with percussion). The second treatment is QXpander, and this treats a stereo signal with similar fairy dust, resulting in a 'wider‑than‑wide' stereo image, whilst leaving the central mono information alone. Both work extremely well with the right sort of signals, and have a wide number of applications outside the area of ping‑pong gimmickry. For instance, when treating a complete stereo mix (or sub‑mix) the effect can be to move the reverb out into the room, leaving more space around individual instruments. At around £200, this plug‑in isn't cheap, and with the majority of plug‑ins going real‑time it is now lagging a bit by comparison, but it still offers one of the cheaper ways to get this revolutionary technology into your music. The main downside for me has to be the copy protection method (see 'The Protection Racket' box, below). QSound have to protect their investment, but when even the title bar of the QXpander window has to accommodate 'QSound's Patented Stereo Expansion Process', leaving only enough room for the first four letters of the sound filename, perhaps things are going a bit too far.

Looking Into The Crystal Ball

The plug‑ins reviewed here are the start of a new wave — others are just becoming available, such as the bargain‑priced Native Power Pack from Waves. Together, they provide a tantalising glimpse of where the PC is heading in its new role as a general‑purpose digital audio workstation. Although the Mac's DSP farm has yet to appear on the PC, and the DSPs on soundcards are dedicated to specific functions rather than being available for general use, this is sure to change as the lure of more profits begins to attract more Mac developers sideways. Start saving up now, or if you want to buy a plug‑in straight away, don't forget those from Glade — they provide fragrance enhancement in your studio for up to 45 days, at only £2.99!

The Protection Racket

In the transition of plug‑ins from the Mac arena, one element has survived which is somewhat unwelcome. The QSound plug‑in incorporates my first experience of hard disk install protection. This involves a copy‑protected floppy disk with (in this case) two installs on board. During each install, concealed data on the floppy is transferred to your hard disk as a hidden file. If this file ever gets moved or corrupted, you lose the install for ever. The floppy will then let you install for a second time, but if this installer corrupts, your program is useless. I know how much Mac owners grumble about this form of software protection, but now, having experienced it first hand, I can only add my voice to the rising clamour, and plead with manufacturers to adopt some other form of copy protection such as that used by Steinberg, whose Wavelab simply asks occasionally for the original CD‑ROM to be inserted. Hard disk installs are a time bomb waiting to go off for legitimate owners! During the course of this review, I upgraded my PC, and therefore had to uninstall the QSound plug‑in from my old machine. The special Uninstall program provided with QTools (don't use the standard Windows 95 version!) stubbornly refused to look at my floppy drive with the original disk, and despite all my efforts, refused to be uninstalled. As this happened over a weekend, there was nothing more I could do, and so I lost one of the two legitimate installs. On my new machine, the second install did exactly the same thing, but by then I had visited the QSound web site, which provides instructions on how to uninstall manually using a DOS command line if problems occur. I know manufacturers want to protect their investment (I've been a software developer myself), but I can't help feeling that if this is typical then many professionals will think long and hard before purchasing any program that features this form of software protection.

Upgrades & Downgrades

Call me cynical, but I think that upgrades mostly come in two varieties — those which look the same on the surface, but have an improved 'engine' resulting in faster or more reliable operation, and those which have new bells and whistles, which sometimes make the basic program feel more sluggish than the previous version unless you upgrade your computer. First impressions of Sound Forge v4.0a were therefore particularly impressive. Not only are there a host of new must‑have features, but a lot of things from version 3 have been speeded up significantly, such as screen redraws. These are now so fast that even on humble machines real‑time scrolling is possible — ie. when you play a long sample, the display updates the current portion of the waveform continuously in real‑time, at any zoom level!

Version 4 of Sound Forge has now been out for some months, but has many significant improvements over version 3 (which was reviewed in SOS May '96), including greatly accelerated processing, new professional reverb algorithms, redesigned pitch‑shift, and graphic fade with dithered or noise‑shaped fades. Other useful basic additions are a new vertical zoom (useful for examining low‑level areas of a waveform, and an excellent playback meter system with many options, such as holding peak and trough level (highest and lowest waveform values). New features include Sync support for the Record dialogue (which makes it possible to send and receive SMPTE or MIDI timecode while recording), and AVI support now allows audio to be edited in sync with video files — the list goes on and on. Suffice it to say that if you already have version 3, the upgrade to version 4 will give you enough new features to keep you occupied for months to come. For more on Sound Forge, see the overview of three PC audio editors starting on page 130 this month.

A Bit Of A Problem

Many people initially think that if 16‑bit audio can approach a theoretical 96dB signal‑to‑noise ratio, this resolution is quite sufficient for digital audio processing. This is not the case, and the easiest way to explain why is by example. Let's say that at a particular sample point in your waveform, its value is 25, and that you simply want to reduce the level of the waveform to half its original value. This would result in a sample value at this point of 12.5, but unfortunately fractions are not allowed, so we decide to call it 12 (by rounding it down to the nearest integer). After some more digital manipulation, we finally need to normalise our levels, which results in the gain being increased by 6dB (doubling the level). This final value will be 24 (a change of double the original error of 0.5), so we can see that the errors are beginning to accumulate, simply because we had to throw away some information earlier. Of course, in the world of 16‑bit audio, there are 65,536 possible values for the waveform at any point, so you might think that missing a half off will make absolutely no audible difference. The problem is that the inaccuracies occur every single time you carry out an edit that changes the waveform in any way — not only changing gain, but EQing, fading in or out, and normalising (this is just a special gain change). It is not just the extremely low‑level values that suffer either — at any point on the waveform, the digital result of an edit may not be exactly an integer value. Since these inaccuracies are cumulative (every edit results in more approximations), they give rise to audible deterioration, which gets worse after each process. Typical end results are grainy reverb tails and fade‑outs that sound unnatural, but you may also notice EQ that sounds 'edgy', and that you seem to be losing the 'ambience' from your sound, giving a result that is 'veiled', 'flat' or 'lifeless'. To minimise these problems, it is often recommended that the only operations you carry out with 16‑bit editors are track re‑ordering (no level changes) and click removal (operating over 1‑2ms, so that rounding errors are inaudible).

The solution is to take the 16‑bit input signal and convert it to a higher resolution such as 18‑, 20‑, 24‑ or 32‑bit. Then, during the entire time that your signal is being processed, calculations will be made with this higher resolution, and with correspondingly smaller errors. Finally, the waveform is truncated down to 16 bits, and the net result is a cleaner sound. This is one reason why it is not possible to compare digital audio editors solely on how fast they operate — it is the accuracy of the end result that counts, not just the speed of the computation!


  • Huge number of extra features.
  • Excellent sound quality and many improved algorithms compared with version 3.0.
  • Extremely fast operation, even running on non‑cutting‑edge PCs.


  • Sometimes too comprehensive for typical users.


A very good audio editor which has evolved into an exceptional one, with open‑ended plug‑in design which allows you to get more from your initial investment. With the latest ActiveMovie technology allowing other compatible programs to access the plug‑ins as well, this has to share the crown for the top PC editor with Steinberg's WaveLab v1.5.