You are here

Audio Architect 4

Software Synth
Published March 1999

The Audio Architect main window, with one of the sample networks loaded, and both the toolbar and module palette docked.The Audio Architect main window, with one of the sample networks loaded, and both the toolbar and module palette docked.

As computers continue to increase in power, so the number of rôles they can play in the studio increases — and the makers of Audio Architect would have us do our synthesis in software. Kevin Perry finds out if it can really replace those old Moogs and Prophets.<image0>

Audio Software Limited's Audio Architect (AA) is another member of the growing family of 'virtual synthesizers' for your PC (such as Reality and Generator), and has been around longer than most: version 1 was covered in Sound On Sound way back in April 1997, when the Pentium 200 MMX was the king of the CPU hill. Since then, the power of PC hardware has increased dramatically, as have our expectations of music software. Does this British package have what it takes to compete in such a competitive market? I settled down to a Christmas with AA to find out (no, not that AA).

The Package

The parameter window for the LFO module.The parameter window for the LFO module.

AA comes on three floppy disks, together with an old manual, a v4 addendum, a version user guide and a set of v4 release notes. Whilst the existing manual is easy to read and understand, an up‑to‑date version is urgently required (as is a grammar and spelling checker) as many changes have been made since the version to which the manual refers; it is also inconvenient to have to flick through the manual, the addendum and the release notes to find the latest details about a particular synthesis module (and that's without having to then go hunting in the online help file). A glossary of synthesis terms, with two useful appendices about sound and electronic music in general, rounds out the manual. I'd like to see the software coming on CD rather than floppy disk, which would be more convenient in this day and age, especially since I have a nasty habit of losing or corrupting floppy disks.

Installation, however, is painless: double‑clicking on the setup icon on Disk 1 launches the installation routine, which prompts for the next disk at the appropriate time. Along with the application, over 60 'networks' (AA‑speak for a complete virtual synthesizer) are installed, along with half a dozen wave files used in the networks. On running AA for the first time, you are prompted for a password: no dongles, disk‑based installs or other offensive copy‑protection schemes here.

In addition to the new modules and enhanced existing modules, AA v4 has been re‑written as a 32‑bit application, which apparently improves performance by more than 500 percent, and enables networks to be edited during audio playback.

The Interface

The parameter window for the wave oscillator module. Note the Analyse button which launches the Wave Analysis tool.The parameter window for the wave oscillator module. Note the Analyse button which launches the Wave Analysis tool.

Upon launching AA, the user is presented with a blank network worksheet, a palette of synthesizer modules and a toolbar with familiar icons for loading, saving and the like. Both the palette and the toolbar can be dragged anywhere on screen and resized for maximum flexibility. AA also uses standard Windows key shortcuts, such as Ctrl‑S for save and Ctrl‑X for cut, wherever applicable.

A module is added to the network by the simple method of selecting it from the palette and clicking on the worksheet to place it. As you would expect, the modules may be rearranged by dragging them around the worksheet, and may be cut, copied and pasted between different networks: AA permits multiple networks to be open simultaneously, which is very useful, especially for doing 'before and after' work.

Connections are made between modules by dragging the source module over the destination with the right mouse button depressed: once the mouse button is released, you are presented with a list of sources and destinations to connect. The list of connections for a module can also be displayed by double right‑clicking on a module once it has been placed in the network. Connections are classified as either control or audio connections: whilst this allows the sample rates for audio and control functions to be set differently (so as not to overwhelm the CPU with control calculations which do not have to be performed as frequently as the audio calculations), the audio and control connections are the same colour on screen, which is a missed opportunity.

Finally, the parameters for an individual module can be accessed by double left‑clicking on the module in question. A pop‑up box allows the parameters to be edited: for a change, the user is sensibly presented with sliders and text‑entry boxes rather than the virtual knobs which so many audio applications use (see screen above). Most of the modules display a numerical value alongisde their sliders, which is particularly useful for the ADSR module.

The output of a completed network can be sent to a soundcard for auditioning in real time, or to a WAV file, either to be used elsewhere or if the PC is too slow to do the calculations in real time. AA can be configured to generate a sound based upon either MIDI input or 'virtual' MIDI input from the sequencer module, on a network‑by‑network basis. In both cases, the network must be activated: Calculate mode must be selected via the toolbar or by pressing the spacebar. In the present version, the MIDI input has to come from an external MIDI port or from a virtual MIDI device like Hubi's MIDI Loopback. AA's own virtual MIDI port is planned for the next version.

AA can be put into a High Priority mode, which improves its performance at the expense of the responsiveness of the user interface: this is useful for slower machines or for squeezing the last ounce of performance out of a faster machine. The Report Time option, however (which, according to the manual, causes a dialogue box to be displayed at the end of a calculation, containing details of the time the calculation took and the length of the generated audio file) does not seem to work: the results appear instead to be written to a file (mae.log) in a moderately inscrutable format!

The Modules

As well as the expected modules for a virtual analogue synthesizer (oscillators, LFOs, ADSRs, low‑pass filters, mixers and amplifiers), AA has some more esoteric ones, in particular the phase displacement (PD) oscillator (as used on the Casio CZ synths) and the user‑configurable Finite Response Filter (FIR).

AA v4 adds more modules rarely (if ever) seen in commercial software synths — an interpolating waveshaper for both audio and control signals, and a wave oscillator (combined waveshaper and sawtooth oscillator) — and enhances the program's existing modules. The FIR can now be turned into an Infinite Impulse Response filter, the LFO module now has a retrigger option, the ADSR module can be configured to use either exponential or linear curves, and the MIDI input module can make use of user‑definable controllers.

DX‑series synthesizers are notoriously hard to program, but if they'd had an interface this simple, it might have been a different story.

The waveshaper passes an audio signal through a user‑defined wavetable: the graphs can either be drawn by hand (which is fine for simple ones) or can be generated automatically by applying Fourier Transforms to existing wave files using the waveshaper's (or wave oscillator's) wave analysis tool. Sadly, AA crashes when this feature is used on any of the wave files that come with the program, but when you find some audio that can be analysed, the results are intriguing. Interestingly, there is also an equivalent module for control signals: one use for this is to create new envelopes with more than four stages.

In Use

Many years ago, because I couldn't (and still can't) afford an EMS VCS3, I tried to recreate one in software on the Commodore 64. Too many attempts to reproduce convincing Space Ritual sounds later, I decided against a career in software synthesis and returned to playing Ultima IV. So I was hoping that in Audio Architect I would finally have something which would let me create these sounds, for importing into Cakewalk sequences or further abuse within Sound Forge.

I started by opening one of the networks that comes with AA, which range from emulators of analogue synths through analogue‑style sequences to way‑out sound effects. Five minutes later, the PC spontaneously rebooted. In fact, every time I used AA, I got no more than about 15 minutes or so of crash‑free use: I got spontaneous reboots, invalid operations, division by zero errors, invalid page faults... Just leaving AA running a network triggered by a sequence caused it to crash in about five minutes. To put things in perspective, the same machine runs Cakewalk Pro Audio (8.01) and Sound Forge (4.0d) continuously for hours with no problems — in fact, I can't actually remember the last time these programs crashed!

To prove that it wasn't just this machine, I tried it on a second machine (not designed for music use) running Windows 98. I never heard anything other than a repetitive noise‑based sound, and crashes were even easier to achieve, although less spectacular (AA hangs but Windows itself remains running): open a network and put AA into Calculate mode; while still in Calculate mode, close the network; open another network and watch AA lock up. Or, put a MIDI‑triggered network into Calculate mode without a MIDI driver loaded (this is not as crazy a situation as it might sound: I have an external MIDI interface which I often don't turn on if I'm just doing sound editing).

In 'normal' editing mode, this doesn't seem to happen — although I never left AA open for more than about 15 minutes, which, as you may imagine, made it difficult to test in anger — but that's hardly the point: release version software should not be so unstable. With this in mind (and most of Christmas now gone), I got down to actually creating some new sounds.

I first tried to recreate some of the classic WWB (wibble‑whoosh‑bleep) sounds that I can (on a good day) get from my ailing Casio CZ1000. In just a few minutes, I had a network making bubbling and burbling sweeps that wouldn't have sounded out of place on an Ozric Tentacles track. The ADSR allows times of up to 10 seconds for attack, decay and release, which is a godsend, as too many real synths don't offer anything like this amount. Being able to modulate the phase of the PD oscillator by any control signal is a great improvement over the CZ series, and allows for some sounds that probably couldn't be achieved any other way. Mimicking the CZ's 8‑stage envelope using the interpolating control shaper (wave shaper for control signals), however, wasn't that intuitive — would it be too much to ask that software synth manufacturers simply provide an envelope generator with more than four stages?

Back on Earth, it's also easy to create networks that simulate more traditional analogue or even FM‑based synths (some DX‑type networks are available for download from the AA web site). Having never had the 'pleasure' of using an FM synth (apart from the limited SY35), I was impressed by how quickly some nasty FM sounds could be generated: the DX‑series synthesizers are notoriously hard to program, but if they'd had an interface this simple, it might have been a different story.

Comparing the analogue sounds generated by AA with a Korg MS10 showed me the great benefits of a synth of AA's nature: no dirty pots, no background noise, and controls marked in seconds rather than 1‑10. AA's filter also does both 12 and 24dB/octave slopes, although I would have liked to see a high‑pass or band‑pass filter offered as well. The AA sounds may not have been identical to the MS10's, but in a track I think you'd be hard pushed to tell the difference. Real‑time control may appear to be lacking in AA compared to a real analogue synth, but parameters can be modified over MIDI, and, with a bit of time and effort, real‑time edits could be programmed using the wide variety of controls available. You may have to change your way of working, but that's as likely to help as to hinder you.

I have to admit that I had less success with the wave oscillator and wave shaper modules. Although I did get some usable textures from these modules (especially with vocal samples), I felt that using them was a bit 'hit and miss'. The FIR, on the other hand, whilst also potentially complex, is explained in the manual in enough detail that you can make educated gueses: it's a pity that you can't apply the FIR to control signals.

Hooking up an external MIDI keyboard to AA is easy: drag the MIDI module onto the network, double‑click its icon, select the MIDI port and channel from the drop‑down lists, put AA into Calculate mode and tinkle those ivories. A global setting for the MIDI port and channel, however, would be extremely useful, especially for those of us with keyboards not hooked up to the first MIDI port. For a typical network, latency from pressing a key to hearing a sound was under a second, although it may well be more if triggering AA from a sequencer running on the same PC. It's not really usable for live work, but is perfectly fast enough for auditioning sounds during editing. It would also be hard to trigger accurately from a sequencer, since the latency does vary from sound to sound, and almost certainly with CPU load as well (so it might vary if you are running audio tracks at the same time).

Conclusion

To be fair, AA has a number of features that other commercial software synths simply do not have; it also has an uncomplicated and Windows‑friendly user interface that even a novice could get to grips with easily. And it doesn't use an infuriating and outdated copy‑protection scheme like certain other music applications I could mention.

On the other hand, it doesn't make use of current Windows technologies like DirectSound or DirectX, interfacing with another application such as a sequencer is clumsy, and it doesn't provide some of the modules (such as basic chorus and reverb) that might be considered essential in this day and age. By comparison, Generator (which works as a DirectX plug‑in, is optimised for use with DirectSound drivers and comes with basic effects modules) costs £169, and something like Stomper can be downloaded from the Internet for free.

But the bottom line is that Audio Architect is simply too bug‑ridden to be considered for serious use: its lack of stability beggars belief. I find it hard to believe that this is the fourth version of Audio Architect rather than an early beta version — in fact, it appears as if some parts have not been beta‑tested at all. Beta versions or free software you can forgive for being unreliable, but not something which costs a few pennies short of a hundred pounds.

If the bugs in Audio Architect were fixed and its price halved, it would be a serious competitor in the virtual synthesizer market; in its current state, though, it simply can't be recommended.

Wot, No Intel?

You may notice a distinct lack of Intel components in the specifications of the two PCs I used during this review (see 'Test PC Specifications' box on next page), one of which is my dedicated audio PC. How come, when everyone recommends using genuine Intel components for audio work? Am I mad? Was it some bizarre accident? Or something more sinister yet?

In fact, it was a conscious decision on my part to buy and use the parts I chose. Until a month or so ago, I was using a Cyrix 6x86MX (PR233) CPU for audio work. That's right: a Cyrix. One of those CPUs whose floating‑point performance is lacklustre, to put it politely. Did it cause problems? No: until recently, mixing of audio tracks was done using integer (whole number) calculations, at which the Cyrix CPUs excel (Martin Walker's figures in a previous Sound On Sound demonstrate this). Moreover, since I was mainly using MIDI, money spent on floating‑point performance I wouldn't need would have been better spent on memory, software or another one of those nice little 19‑inch MIDI modules that I seem to collect.

It was only with a recent upgrade to Cakewalk 8, which uses an enhanced floating‑point mix engine, that I decided that it was worth replacing the CPU. So I retired the Cyrix and went and bought... an AMD K6‑2. Why not a Pentium II or Celeron?

Firstly, I'm still mainly using MIDI with a dozen at most audio tracks, for which the K6 is perfectly adequate — and it is as fast at general Windows operation than an equivalently clocked (and more expensive) Pentium II. In fact, the IBM/Cyrix CPUs easily represent the best price/performance ratio for a MIDI musician on the PC.

Secondly, although the Cyrix's floating‑point performance is weak (at best 33 percent slower than an equivalently clocked Pentium II), the floating‑point performance of the K6 and K6‑2 is not as bad as many make it out to be: at best it is 50 percent faster at floating‑point calculations than a Pentium II at the same clock speed (if audio applications made use of the 3DNow! extensions present in the K6‑2, it could be up to 300 percent faster!).

Below is a table of CPU usage by processor and effect (the Pentium 166 MMX figures are taken from Martin Walker's March 1998 article; I chose a selection of the effects he used to which I also have access). All tests were conducted on mono data at 44.1KHz — where possible, I have used both 16‑ and 24‑bit calculations (24‑bit calculations provide significantly better dynamic range than their 16‑bit equivalents).

Intel Pentium 166 MMXAMD K6‑2‑400
Cakewalk CFXChorus (16)5.5%2%
Chorus (24) ‑‑3%

Sonic Foundry


Chorus (16)


29%


10%

Chorus (24) ‑‑14%

Waves


C1 Compressor (16)


15%


6%

C1 Compressor (24) ‑‑5%
L1 Ultramaximiser (16)20%3%
L1 Ultramaximiser (24) ‑‑6%
Trueverb (16)72%35%
Trueverb (24) ‑‑31%

The Cakewalk CFX use integer calculations (the more recent Cakewalk effects use floating‑point), and all the others, as far as I am aware, use floating‑point calculations. Given the relative prices of the K6‑2, the Celeron and the Pentium II (and their motherboards), I think the figures speak for themselves.

Minimum Specifications

The Audio Architect manual states that it is compatible with Windows 3.x, Windows 95 and Windows NT. However, since the program has been re‑written as a 32‑bit application, it will no longer work with Windows 3.x. Audio Software's web site does list Windows 95/98 and Windows NT compatibility.

A Pentium 133 or better is recommended for real‑time playback, although as with any audio application of this type, the faster the CPU the better. No minimum memory specification are given, although 32Mb is probably the minimum that anyone should consider using for audio work under Windows 95 (a minimum of 64MB for Windows NT is likewise reasonable).

A 16‑bit soundcard with Windows MME drivers is required: there is no need for DirectSound drivers or the like. However, DirectSound drivers are likely to provide lower playback latency than standard Windows MME drivers.

Test PC Specifications

AUDIO PC

  • AMD K6‑2‑400 (100MHz buss).
  • Aladdin V Super Socket 7 motherboard (512K L2 cache).
  • 64Mb PC100 SDRAM.
  • Matrox Millenium G200 AGP.
  • Korg 1212 I/O soundcard (v1.1 Windows MME drivers).
  • MOTU MIDI Express XT MIDI interface.
  • Windows 95 OSR2.5 (USB supplement).

'BUSINESS' PC

  • Cyrix(IBM) M2‑PR333.
  • FIC VIA VP2 motherboard (1Mb L2 cache).
  • 128Mb SDRAM.
  • VideoLogic Apocalypse 5D Sonic soundcard (DirectSound drivers).
  • Windows 98 SP1.

Pros

  • British!
  • No copy protection.
  • Logical and simple user interface.

Cons

  • Very unstable and bug‑ridden.
  • Doesn't integrate neatly with other audio applications.
  • Limited effects modules compared to similar programs.
  • Doesn't make use of recent enhancements to Windows.

Summary

In the increasingly crowded software synth marketplace, unless you're determined to support a British product, it's hard to recommend Audio Architect at this price and with this level of quality.