Kyra's FPGA-based synth engine is technologically innovative, but what's it like as an instrument?

Two years ago I reviewed the Waldorf Quantum, a superb synthesizer that I view as the company's finest achievement to date, so it was with some excitement that I approached the Kyra. This feeling was enhanced by its reassuring size and weight as well as its smooth, wobble-free knobs. I was a bit surprised by its small monochrome screen, but it was only when I looked at its rear panel that I was marginally disappointed; I have no idea why Waldorf used 1970s-style black plastic quarter-inch sockets and ring nuts and allowed them to protrude so far.

First Thoughts

The Kyra's basic unit of sound is a Patch, but you can program and play one of these only when it's inserted into a Part. This contains additional parameters — MIDI channel, level, pan, transposition, output pair and so on — that are edited and saved as part of an enclosing Multi. There are eight Parts within each Multi and these can be split by restricting them to constrained keyboard ranges, layered by making them respond to the same MIDI channel, or used multitimbrally by allocating different channels to each, or any combination of these. There's no voice stealing between Parts, so you can think of the Kyra as eight individual virtual analogue synthesizers drawing from a common pool of Patches. The strange bit is that you're always in Multi mode, so you only ever have access to Patches within the Parts of a Multi. Furthermore, the Part doesn't contain the Patch; it only points to the original so, if you modify and save a Patch within one Multi, you'll affect the same Patch in every other Multi that uses it.

The memory holds 128 Multis plus 3328 Patches spread across 26 banks, the majority of which are described as ROM. But, while you can't overwrite individual patches in the ROM banks, you can transfer banks of Patches to them, so it's not ROM at all but some form of flash RAM. However, it's the genuine RAM that worries me. The manual explains that this is maintained by a rechargeable battery that, in the absence of mains power, will sustain the memories for just a few months. If you're accustomed to synths that haven't had their CR2032 batteries changed in decades, you'll have to keep a close eye on this.

Rather than provide conventional VA oscillators, there are two Oscillator Groups (OGs) in a Kyra Patch, and these offer two modes of sound generation: Wave and Hypersaw. The first of these allows you to program the OGs independently, with each offering three simultaneous waveforms: sawtooth (actually a ramp), pulse and Wave, the last of which accesses a library of 4096 public domain single-cycle waveforms. Waldorf often refer to these waves as wavetables, but don't be misled; there's no wavetable synthesis on offer. There's hard sync of OG2 by OG1, and frequency modulation of OG1's Wave by OG2's. An Auxiliary oscillator control in OG1 offers two additional functions, acting as the volume control for a white noise generator in both OGs, or as the volume control for the output from a ring modulator that uses the Waves selected in OG1 and OG2 as sources. Consequently, you can't obtain ring modulation and noise simultaneously, and you can't generate different noise levels in each of the OGs. Each OG also boasts a detunable sub-oscillator offering four waveforms. This is a separate entity from the ramp, pulse and Wave generators because it doesn't contribute to, and isn't affected by, FM, sync or ring modulation. This makes it useful for enforcing the fundamentals of exotic timbres generated by sync and, in particular, FM.

Selecting Hypersaw mode defeats OG2 and causes OG1 to generate six detuned sawtooth waves as partials. You can control the spread of these partials as well as their loudness distribution and their stereo width to create anything from gentle chorusing to insane amounts of wobbling. In addition, a sub-oscillator function can steal two of the six partials and detune them by an octave. Hypersaw generates a limited but important range of sounds, including ensembles as well as many useful pads. However, I found that my Hypersaw Patches distorted unless I reduced their volumes to as little as five or six out of 127. One consequence of this was that, when I used MIDI CCs to fade Hypersaw sounds in and out of a performance, the coarse quantisation made this less smooth than I wanted.

The levels for the various waves generated by the OGs are mixed within the Groups before passing to the two filters in the filter section. These offer six filter types — low-pass, band-pass and high-pass, each with 12dB/oct and 24dB/oct options. Waldorf describe these as "accurate emulations of classic analogue ladder filters" and claim that "higher [resonance] values can cause large resonant peaks and, ultimately, self-oscillation" but nothing in the Kyra's filter section would cause it to self-oscillate in any conventional fashion, even if I used the modulation matrix to further boost the amount of resonance.

The default signal path is for the signals from OG1 and OG2 to be mixed equally and then passed through just Filter 1. However, there's also an OG Dual Mode that, in Wave mode, layers one OG1+OG2 pair with a second and allows you to detune them. The first pair is then directed to Filter 1 and the second is directed to Filter 2. However, you can't adjust the filters separately; apart from a couple of obscure modulation options, they act equally upon both pairs. If you invoke OG Dual Mode while in Hypersaw mode, a Hypersaw with 12 waveforms is generated, with alternate partials directed to Filter 1 and Filter 2. But, again, you can't adjust the filters individually, although you can adjust the stereo width of the resulting sounds.

There's also a Dual Filter Mode. This automatically invokes OG Dual mode, but it doesn't invoke the unison layer. With both Dual modes selected, the output from a single OG1 is passed through Filter 1, and the output from a single OG2 is passed through Filter 2, and you can now program the filters independently. However, you can't invoke Filter Dual Mode and Hypersaw Mode simultaneously, and selecting one always defeats the other. Since the relationships between these modes are a bit arcane, I've created the table below to help explain things.

Following the filters, the signal passes through a compressor that the manual calls a limiter. This offers light, medium and heavy responses, and I suspect that this is where some of the unwanted distortion I encountered had originated. Following this, the audio amplifiers shape your notes in the usual fashion, with EG1 hard-wired to their gains. There are also parameters for the static pan position, the LFO1 amount (to generate tremolo), and the LFO2 amount (to generate auto-panning).

The Effects

The signal now passes to the multiple effects units contained within each Patch. Although the control panel shows six of these, Waldorf claim that there are nine modules per Part, but even if you count the formant filter and compressor as separate modules, that's still only eight.

The chain starts with an EQ offering high and low shelves and a sweepable mid with variable Q. Treat this with respect; adding too much gain here will again cause distortion. Also, don't try to modulate it in the modulation matrix because you'll hear stepping. The EQ is followed by a two-band formant filter. Unfortunately, there's no Q control and you can't tune the peaks separately so, while it imparts a pleasing vocal quality to sounds, it's not as flexible as one might hope.

Next comes another compressor, followed by a distortion module. There are five distortion models: soft and hard rectifier, soft and hard bit-crusher, and Gnasher, and you can soften the distorted sound using a gentle low-pass filter. I'm not a fan of the first four models, but I quite like the Gnasher which, when used sympathetically, can add bite. But warm valve distortion it's not.

These are, in turn, followed by a delay that offers stereo and ping-pong modes and feedback of up to unity. I was a little surprised to find that changing the Delay time has no effect once the signal has been sampled, so you can't use it for modulated effects. Timing can be derived from a rather coarsely defined internal clock, MIDI and LFO3, the last of which means that you can create synchronised delay/modulation effects without using an external clock.

The delay is followed by a six-stage phaser and then a chorus, which I think is a strange choice of signal path. (I would almost always place the delay after these.) The Phaser offers bipolar feedback gain, a dedicated LFO that runs slowly enough for the gentle sweeps that we all love, good modulation depth, and control over the base frequency of the effect. Its menus provide a huge range of LFO waveshapes but, once you've experimented with these, I suspect that you'll return to the usual sine or triangle waves. You can also modulate the effect via the modulation matrix, and doing so creates a somewhat different sound. The chorus unit has four modes of operation: comb-filtering, flanging, chorusing and doubling. Its sound is pleasing but it lacks the deep swoosh of an analogue chorus. As elsewhere, it's very easy to create distortion in these units, so you must watch your signal levels to avoid this.

The final effect is a reverb unit with a claimed maximum time of 20 seconds, pre-delay, 'darkness' (a low-pass filter on the input to the reverb) and damping. I sometimes found this to be useful, but on other occasions it created a strange detuned effect that was rather odd. I would use this sparingly.

At the end of the signal path there's yet another compressor/limiter. This keeps the signal within bounds before passing it to an Output Mixer that allows you to determine the level of the Patch and its position in the stereo field on a chosen analogue L/R output pair. Again, distortion is your enemy here because, if you direct multiple Parts to a given pair, clipping may ensue. Finally, there's a master volume control that controls all of the analogue outputs including the headphones.

Modulation

A Kyra patch offers three contour generators. These are described as ADSRs, but in fact all three sport an additional Slope parameter that allows the Sustain phase to rise or fall while held. This isn't the same as a five-stage contour, but can be very useful when emulating percussive sounds or when programming sforzando for brass instruments. Contours 2 and 3 also have a parameter called Bass Delay that very slightly offsets their initiation to overcome what might otherwise be a problem caused by some of the underlying DSP. I never encountered the problem so I didn't need the solution, but it's nice to know that someone had thought about it. To say that the contours are fast is an understatement, and I always found myself slowing the attack and release phases a tad to avoid transient artifacts. The resolutions of the A, D and R times are a touch coarse but, to be fair, I didn't encounter any problems here either.

There are also three programmable LFOs per patch. Each offers 128 waveforms and each can be synchronised to MIDI Clock. You can program them monophonically (per patch, free-running) or polyphonically (per note, either key-sync'ed or with random phase). Furthermore, LFOs 1 and 2 offer delay, Dual antiphase (the free-running monophonic LFO is shifted by 180 degrees on the second layer in a 'Dual' patch) and Dual quadrature (the free-running monophonic LFO is shifted by 90 degrees on the second layer) options. Programmable start phases will also be added on the next firmware revision. You can modulate one LFO using any other (including itself) to extend the number of waveforms even further and, in addition to their hard-wired destinations, all three are available as sources in the modulation matrix. Unfortunately, their frequency ranges are a little narrow; a minimum frequency of 0.1Hz is too fast for very slow sweeps, while the recently updated maximum frequency of 51.2Hz is too slow to permit audio-frequency modulation of the filter and amplifier. Since the audio oscillators can't be used as modulators, there's no way to overcome this.

The modulation matrix itself offers just six slots, each with one source and three destinations. You have a choice of 34 sources (many of which are MIDI CCs and other controllers) plus 82 destinations. Unfortunately, there's no controller column in the matrix, nor can you cascade slots, so you can't ask this to do to something to that by an amount that's being determined by the other. However, you can direct multiple sources to a single destination as well as one source to multiple destinations. If you replicate one of the Kyra's hard-wired modulation paths, the two amounts are summed.

Second Thoughts

When you're ready to start programming the Kyra, you'll discover that there's much about it that's far from intuitive. This is also when you'll discover that the current version of the manual has errors. Some of these appear to be because it hasn't kept up with software updates, but others were wrong from the start. It would also be easier to understand the Kyra if a full block diagram was included. But my biggest complaint with it is that in places it's written like a brochure rather than a manual. Waldorf have undertaken to update this.

Consequently, it took me some time to work out the architecture of the Kyra and to learn how to navigate it. Some things — such as finding some of the amplifier parameters in the filter section — just took time to get used to. Some, such as the choices of which parameters had knobs and which were only accessible through the menus, were noticeably odd. Other things were more significant. For example, the Kyra uses potentiometers, so the knobs' positions are meaningless when you load a new Patch. Touching a knob will therefore cause the selected parameter value to jump instantaneously to the knob's physical position, not always with pleasing results.

The Shift Lock button is another oddity. This allows you to access all of the control panel's secondary functions without pressing Shift each time that you want one, but it's easy to forget that it's switched on and I found myself swearing when I thought that I had adjusted one thing only to find that I had modified another. But even when you've gotten to grips with all of this, programming can be frustrating, because you can only change knob-less parameter values using the + and buttons; there's no dial, nor even a data value fader. It takes nine seconds to scroll through the longest value list, so the Kyra clearly needs a software editor.

Two other aspects of the Kyra's programming system also caused me problems. Firstly, when you're editing a Multi, the big name in the centre of the screen is the name of the Patch in the selected Part, not that of the Multi itself, which I found confusing. Secondly, I found to my cost on several occasions that saving a Multi doesn't save any of the Patches that you've edited within it so, if you then jump to another Multi (whether intentionally or accidentally) all of your changes are lost. This is far from unique to the Kyra but, because there's no Patch mode, it's much easier to make a mistake.

Some Sounds

Eschewing all of its esoteric synthesis capabilities, the first patch that I created on the Kyra used just a single OG, square contours for both the filter and amplifier stages, and no effects. This should be the easiest sound for a synthesizer to create but, every time that I released a key, I obtained a series of impulses after the wanted sound ended. This had to be a bug, so I downloaded the latest firmware (at the time, v1.74) and this eliminated it. Having done so, I moved on to programming a selection of simple organ sounds. This was progressing well until I tried to defeat velocity sensitivity. I searched for the mechanism for a long time because it never occurred to me that this wasn't possible. I discussed it with the chaps at Waldorf and, by the time that you read this, a new per-patch velocity scaling parameter (which includes 'off') will have been added. However, there's no plan at this stage to add a choice of velocity curves.

I then switched tack and attempted to create the acoustic and electric pianos that the manual promised. I found that I could create some typical VA e-piano sounds, but acoustic pianos are (as I had expected) way out of reach. I then tried to program some orchestral sounds, but with only occasional success. For example, when I tried to imitate a solo trumpet, I heard a strange artifact as the filter opened, and I was unable to obtain conventional delayed vibrato because the Delay parameter determines the length of a ramp up to full amplitude rather than delaying the onset of the effect. I wasn't a fan of the Kyra's solo synth patches, either, especially since monophonic portamento is only implemented in legato and 'equal time' modes. On the other hand, I liked some of the Waves that allowed me to create FM-type sounds without invoking FM, and I enjoyed imitating LA synthesis by using one Part to create an initial attack and a second to create a sustain phase. I also found that applying FM and ring modulation to some of the more unusual Waves could create strange and interesting timbres, although I could never get the sync to exhibit that anguished tear that can be so wonderful on other synths.

Ultimately, I think that the real strength of the Kyra lies in its ability to layer Parts to generate rich ensembles, huge pads, and weird panoramic floaty things as well as synthesized percussion and rhythmic textures using its multiple arpeggiators (see 'Arpeggiators' box). Just be careful to keep track of the polyphony. OG Dual Mode uses two voices per note so it halves the Kyra's maximum polyphony to 64 notes. If you use this mode across eight layers, the maximum drops to just eight notes.

I really want the Kyra to be fabulous, so I'm going to hope that Waldorf will continue to develop it and turn it into the instrument that I hoped it would be.

While discovering all of this, I encountered two other oddities. The first concerned the contours' responses to repeated notes. As one would hope, the Attack phase of a repeated note picks up from the current Release level, which is good because it avoids the nasty sucking sound that afflicted low-cost synths in the late '70s and early '80s. However, instead of holes there were sometimes clicks when I played repeated notes. These were a consequence of playing the repeated note at a different velocity to the previous; although the algorithm thought that it was picking up the contour at the right point, there was a discontinuity if the second note was quieter or louder than the first. But perhaps my greatest frustration with the Kyra occurred when I tried to insert Patches into Parts in a Multi and transpose them to obtain the layers that I wanted. On other synths, the transpose parameter acts as a virtual CV, shifting everything that's connected to it. On the Kyra it's merely an OG pitch transposer, so your filter tracking parameters and other attributes are not shifted and the timbre of the sound changes. This means that, if you need to transpose a split or a layer, you need to reprogram the sound and then save the result as another Patch. I can't ask SOS to print what I think of this.

Final Thoughts

The earliest version of the Kyra was developed by a chap named Manuel Caballero, called the Exodus Valkyrie and shown at the 2018 Frankfurt Musikmesse. It was only when Waldorf took over development that it was renamed and reshaped into its present form, and I think that this explains many of my concerns about it; whereas Waldorf have released some polished products in recent years, the Kyra feels disjointed and unfinished.

Waldorf's marketing department might suggest that it's special because of its use of FPGAs, but I don't see that this has imparted any tangible benefits. They might also point to the 32x oversampling in the OGs, but I would point out that the signal is downsampled by a factor of 16 even before it reaches the filter. Nonetheless, if the firmware continues to be developed and if someone releases an editor to make programming more intuitive, it could become a solid instrument. But I can't promise that it will ever be exceptional; despite the undoubted synthesis power available, the waveforms are public domain, the modulation matrix is small by current standards, and the underlying 'patch plus all of its effects' architecture has been around since Novation released the Supernova more than 20 years ago.

I really want the Kyra to be fabulous, so I'm going to hope that Waldorf will continue to develop it and turn it into the instrument that I hoped it would be.