If the unstoppable tide of virtual analogue synths is leaving you cold, you could consider diving into Poseidon 's refreshingly different 'Ocean of Sound'...
Over the last few years it's seemed as though everyone and their aunt has released a soft synth, and it's easy to become jaded by the prospect of so many virtual instruments that do pretty much the same thing. And so, to my shame, there are several diamonds in the rough that I've overlooked, including the whole product line from Virsyn, which started with the hybrid modular synth Tera (which was reviewed in Sound On Sound 's February 2003 issue). Since then, Virsyn have released another synth, called Cube (reviewed in SOS September 2003), the Tera-based sound bank Minitera, and the real-time vocal synthesizer Cantor (covered in SOS October 2004). Since then, most of Virsyn's output has been version updates providing significant functional additions to these products, but now they have a brand-new 'spectral modelling' synth called Poseidon, enigmatically subtitled 'The Ocean of Sound'.
Trying to describe what kind of a synth Poseidon is presents a refreshing challenge. For the first couple of days I found myself struggling to wedge it into familiar categories. Is it granular/graintable resynthesis? Or more like wavetable? Neither is correct, although there are operational similarities to synths that use these methods. So, let's look first at what Poseidon does before we get too hung up on how it does it.
Launching Poseidon (either as a stand-alone app or as a VST, AU or RTAS plug-in) brings up a subdued blue-grey interface, with a spartan layout divided between a panel of knobs, entry fields and pop-up selectors, and a preset browser. The browser is very similar to the attributes browser used in Native Instruments' synths and Kore. You can highlight various attributes listed by Category (instrument or sound type), Timbre (for example, 'bright' or 'dissonant'), Expression (such as 'plucked' or 'pad') and Style (musical genre). A list of presets that match the selected attributes is displayed to the left. Also in common with the latest versions of NI's synths is the ability to slot presets into a 128-patch bank that responds to MIDI Program Change messages.
Trying out a few presets shows the diverse range of sounds that can be produced with Poseidon. As well as more traditional synth leads and pads, Poseidon clearly excels at rich, evolving soundscapes and bright, percussive sounds. The big dreamy textures (and many other sounds) are reminiscent of Absynth, but there are also less 'synthy' sounding patches that give the impression of a virtual modelling synth. In particular there are rich plucked sounds and breathy brass and horn sounds. In between these sounds, there are a few less inspiring patches that don't show the synth off so successfully, suggesting a shortage of time when compiling and tagging the library. There is also a wide variation in level between patches. Having said this, two extra add-on packs that enhance the library were made available shortly after release, and more are sure to come. In any case, once you get your head around how this synth works it's possible to create sounds to your own taste remarkably quickly.
The best way to understand Poseidon 's synthesis method is to get to grips with what the spectral model graphs represent. The graphs are not colourful just to look pretty: the colours are used to represent a third 'axis'. We are quite familiar with audio waveform displays, which are graphs that represent a sound as amplitude changing over time. Most of us are probably comfortable with 2D spectrum graphs as well, which show how a static sound is made up from a series of individual sine-wave components. Poseidon 's models (and graphs) describe a sound by showing how the spectrum graph changes over time. Frequency components ('partials') are plotted on the Y-axis against time on the X-axis, with colour representing the relative levels of the partials. Natural sounds, such as the Sitar on the first page of this article, show a smooth evolution, and you can discern from the graph the instrument's characteristic frequency components and how they change and fade over time. Classic subtractive synth sources, such as the PWM saw in the screen on the next page, have strong components across many frequencies and so look like blocks of colour, or patterns. In another artificial model called Random Pings, illustrated here, you can clearly see individual components appearing and fading over time.
Poseidon recreates sounds additively using its 512 sine-wave oscillators mixed with residual noise, but this description doesn't really do it justice. The key is that it can change the sound over time, either mimicking a 'real' sound or creating something entirely new. This is where the similarity to wavetable or graintable synths comes in, except that instead of scanning through a table of waveshapes or audio snippets, you move (manually or via envelopes, loops and LFOs) through the various spectrum states. In this sense you could call Poseidon a spectrum-table synth.
Looking at the control panel, most synthesists will feel at home with the familiar tuning controls, LFOs, envelopes, filter controls and so on. However, you're clearly not dealing with a typical subtractive synth here; for a start, you'll see no familiar oscillator sections. To see the sound generation in action you need to click the Spectrum button, which replaces the attributes browser with a colourful graph display. The graph is a 'spectral model', a representation of the frequency components of a sound over time (see the 'Spectral Modelling Synthesis' box above for a more in-depth look at Poseidon 's synthesis method). Poseidon has over a hundred models in its factory library, with more being added occasionally. Some of these models are mathematically generated, and are stored together in a folder called 'Artificial'. The rest are created by analysing other sounds, which can be synthetic or real, percussive or sustained, and so forth. You can also analyse your own samples, vastly increasing Poseidon 's scope.
Let's look at a simple example with all controls at default settings, as in the screen at the start of this article. The spectrum model chosen is from a real source, a sitar. Playing a key, you see a vertical white line scan through the graph and hear the model 'played back'. When the model comes from analysing a sample, a patch with these settings will sound more or less like the sample being played back. However, unlike a sampler (playing back a digital audio recording) or a granular synth (playing back a series of small audio snippets), Poseidon is resynthesizing the sound additively, using a stack of 512 sine oscillators and residual noise generators. Besides, playing back the model in its original form is only one of many possibilities.
A set of parameters (as seen in the screen below) lets you control how Poseidon scans through the spectral model. Poseidon has several sampler-style loop modes, such as One Shot (no looping), Loop (forward looping), and Alternate Loop (forwards-backwards looping). The Loopstart and Length controls shown in the screen allow you to fine-tune which section of the model is used for looping. The Position control sets the start point when you hit a key, and can also be adjusted while notes are playing, allowing you to scan through the model manually or via LFOs, the Mod Wheel and so on. A standard ADSR envelope lets you set up movement through the model in four stages. Finally, the Time control adjusts the basic speed at which the sound engine scans through the model. The Time ratio can be up to 10 times real-time, and all the way down to zero, at which point the sound is frozen. If you're used to standard time-stretching, you will be amazed at how this sounds, because it is perfectly smooth all the way down, with no graininess. Again, this is because Poseidon is actually recreating the sound from scratch rather than trying to play back a sample.
Once you understand the basic concept of how Poseidon generates sounds, it's remarkably quick to create your own patches. There are a number of different ways of approaching sound design with Poseidon, but mostly I found myself working in a similar way to how I'd work with a wavetable synth. Once you've found a spectral model you're interested in, you can quickly set up a starting point, loop and speed, and add movement with the position envelope and/or an LFO. You have to think differently to how you would with a standard subtractive synth. A lot of the harmonic content and movement in a sound can be extracted from the model, instead of shaping and modulating traditional parameters such as pulse width or filter frequency. For example, if you want a pad with pulse-width modulation, you find a spectral model from a PWM'd sound source (such as in the screen on this page) and set up a loop that scans through it. The loop Length effectively determines the modulation depth and the Time factor sets the modulation speed. Similarly, a plucked or percussive sound can be created using a plucked or percussive spectral model, instead of trying to mimic it with envelopes.
Parameters that can be modulated have a drop-down menu listing all available sources (the LFOs, any of the three envelopes, and performance controllers such as velocity or mod wheel). Specific list items exist for combinations, such as EG1 scaled by velocity. There is a deliberate simplicity to this modulation system, which means that most of the control sources are either hardwired to certain parameters or only available on a select few. This makes Poseidon easier to learn and use, but is sometimes a bit limiting. For example, I would have liked to be able to modulate the Time control from an envelope and modulate some parameters with multiple sources. Virsyn say they are considering adding more modulation possibilities, and also possibly Macro controls, but only if they can find an interface solution that doesn't compromise ease of use.
As we've seen, many characteristics of a sound, including filter sweeps and envelopes, can be built into a patch by using a spectral model that has the desired characteristics inherent in it. However, Poseidon also has a filter, with various modes and controls. Although you can use this filter to emulate more familiar filter types to a certain extent, this is far from being a standard filter. Virsyn have called it an F-Domain filter, with the 'F' standing for Frequency, and although there are familiar mode names (low-pass, high-pass, comb) the results are often markedly different from a standard time-domain filter design. I won't pretend I fully understand the difference between the technologies, but apparently the filter is interacting with the spectrum at the synthesis stage, as opposed to operating on the audio signal post-oscillators. The filter has four parameters: Cutoff, Slope, Resonance and Width. Cutoff works pretty much as expected, but Slope gives you continuously variable control over the slope from zero to 72dB! The Resonance control is equally capable of extreme results, and also shapes the sound in quite a different way to what we're used to. Beyond subtle levels, it reduces all other frequencies until you effectively get a band-pass filter, almost as though it 'sucks' the whole sound into the resonant peak. The Width control then sets the bandwidth of the resonant peak. At the lowest levels, you get a peak that is so narrow that it can pick out single partials. When Resonance is high, you can sweep through the sound for tinkling or water-like effects. Comb Filter II has a series of notches spaced logarithmically, giving rather different sound-shaping possibilities to the regular Comb Filter. Promised for the first update (which should be out when you read this) is another filter mode which lets you shift or freeze the formant characteristic of the sound model. This will allow you to play models from real-world sources up and down the keyboard with no 'chipmunk effect'. Again, this is possible because the filter is part of the synthesis process, rather than a separate signal-processing module.
The sound-generation technique used by Poseidon is considered by many — especially those involved in academic synthesis research — to be the ultimate synthesis method. This is because theoretically it can synthesize any possible sound. It's certainly true that a very wide range of sounds is possible with Poseidon. However, a synth is more than just its oscillators. The biggest challenge, and maybe the biggest success of Virsyn's synths, is finding ways to make such an open-ended synthesis technology usable in a real instrument. With Cube, they chose morphing, using up to four sound-generator modules, with complex envelopes and modulator to crossfade between them. Poseidon takes quite a different approach, using just one of the sound engines from Cube, but with many possibilities for moving through a potentially complex sound space. This approach is easier to use, but the results can be almost as rich and evolving. In fact, to my ears, Poseidon actually sounds smoother and silkier than Cube. I suspect that this is partly because the effects in Poseidon are so nice. I also wonder if the fact that the synth has three less sound sources to deal with means that the single sound engine can afford to run at a higher quality, with higher sample rates.
The fact that you can create spectral models from your own samples gives you the potential for a limitless number of sonic starting points. You could use Poseidon like a cross between a sampler and a synth. It is certainly not a replacement for a sampler, because complex real-world sounds are not recreated perfectly. However, you can play the sound up and down the entire range of the keyboard in a way you can't with a sampler.
There are many possibilities that I'm only just beginning to get to grips with. You can emulate virtual modelling by sampling the characteristic of plucks and hits, and create classic analogue synth sounds, either using the mathematical models or analysed samples from the real things. This is aided greatly by the Unison/Ensemble mode, giving you up to eight unison voices, or various Sub or Fifth modes with up to six voices. You could even sample in a sweep through the wavetable from a wavetable synth and recreate classic wavetable sounds. Mainly, though, I found myself exploring the depths of the sound models with various Time factors and loops, looking for unique and inspiring textures.
I became beguiled by Poseidon very quickly, even before I understood how it worked. It has an ingenious and well-executed sound engine, slotted into an instrument that has a familiar enough design to make it usable. It's also capable of producing a wide range of real-world and synthetic sounds, from basic analogue leads to sumptuously rich textures and soundscapes. If your regular synths are feeling a bit stale, you'll find it a joy to be presented with a new way of thinking about sound design. To sum up, Poseidon is a beautiful instrument.