Dispensations

2023-07-06 MSK 015 music composition

The MSK 015 Quad VCA is an important milestone for North Coast Synthesis because it's the last missing piece for building a complete synthesizer: with this module and the rest of the North Coast line, it's possible to make music without needing any modules from other manufacturers. Partly to celebrate the release, and partly because I needed some better audio demos to post in the storefront, I spent most of June writing and recording music with my own modules.

I call the result Dispensations; it's posted as an album of twelve individual tracks on the audio server, and you can listen to the whole thing in a single 44:30 track with this player.

Writing the music

My goal for this project was quantity and speed more than quality. I wanted a dozen tracks that would use my modules, especially the Quad VCA, and they needed to all be good enough that I wouldn't be ashamed to have them posted in my storefront, but I wasn't planning to sell CDs of this music or try to get paid gigs performing it, and although the project did end up taking me about a month start to finish, I didn't want it to take two or three months.

With that consideration in mind, I aimed to write the tracks assembly-line fashion, with as much shared between them as possible. I used the same instrumentation for all the tracks: a sweet "lead" sound with at most two notes played at once; a "pad" sound playing chords with up to four simultaneous notes; a monophonic "bass" sound with a grittier timbre than the lead; and percussion.

Since I don't have enough synthesizer equipment to patch up a whole band's worth of sounds all at once, the plan was that I'd take the MIDI files for all the tracks, concatenate them, and then record a single album-length take of a given patch for all the tracks; multiple such takes for the polyphonic sounds, where I might not be able to patch up all the voices at once. Then, assuming the timing was good, I could mix down the single-patch recordings on the computer to create the final tracks.

I wrote the music in Lilypond, creating sheet music PDFs and MIDI files from a single source code file per track. The twelve tracks each have basically the same three-part musical structure, so I was able to create a template for that structure with the appropriate markup for the instrumentation, number of measures, and so on. Then I could just copy the template and put in the music as such for each track. The template did require some modification for tracks like "Three of every sort" (which is in 12/8 time instead of the usual 4/4) and for the multi-ending repeat structures I ended up using in many of the tracks, but I think it still saved time over starting completely from scratch on every track. Here are the sheet music PDFs.

1. Apostrophe
2. Age of reason
3. Yeah gimme that one
4. Danse de spongieuse
5. Coast Guardian angel
6. Three of every sort
7. Fanfare for the undeserving man
8. Boogie for John Nelson
9. I.B.A. unapproved
10. Above the law
11. Age of inflation
12. In the eschatology of the Latter-Day Kyonistas

Choosing titles for the tracks was almost the last thing done, followed only by choosing the visual art; for most of the project I was just calling the tracks "Track 1," "Track 2," and so on.

Since I couldn't build a modular-synth patch that would play all of the instrument sounds in their final forms at once, I ended up using a software MIDI synth with basic General MIDI sounds to preview the tracks while I was writing the music. I just had to imagine how they would sound with the final patches. Much like a "real" band recording in a studio, I didn't get to hear anything close to the final mix until I was actually making the final mix. The delayed gratification worked surprisingly well: most of the finished tracks did come out pretty close to what I was imagining when writing the sequences, though I had to record some extra takes for some of the sounds, where the first patches I set up sounded good tested alone but not good in the larger mix.

Percussion

I wanted to have a variety of different musical styles among the twelve tracks, while still keeping the same instrumentation, and I figured I could accomplish that by using many different drum grooves. I got one of those "999 ready-made drum machine patterns for people who failed the IQ test to become actual drummers" books (not its real title) and worked my way through it, basically writing the percussion section for each track first, before any of the other instrument sounds.

The good side of using the book was that it did give me a head start on different styles that I wouldn't have thought of using if left to my own devices. The bad side was that it was only a start. I had decided to stick to just four percussion sounds, so I wouldn't have to spend many hours patching different ones: a low kick-type bass drum; a higher, grittier snare-like sound; and a short and a long cymbal. Patterns from the book often called for a wider range of sounds, and I had to decide which ones to leave out and which ones to map onto which sounds in my limited set.

On the other hand, the book was written for a grid-type drum machine with very limited options for timing, and my MIDI setup could do a lot of timing stuff like variable swing and tuplets which the grid-type machine couldn't. If you pay money for a book that promises "999 drum patterns," you're going to be disappointed when many of the 999 are just along the lines of "hit the kick on beat 1 and the hi-hat on every beat" even though that actually is a pretty good drum pattern, so nearly all the patterns in the book were much more complicated. I ended up modifying them a lot for my purposes, sometimes to the point of unrecognizability.

The big challenge for patching a drum sound with North Coast modules was that I don't currently offer a "noise" module. I did some trial and error patching between two Middle Path VCOs until I got a noise-like spectrum that worked well for the snare sound. In order to save recording time I patched the bass and snare in a single patch and recorded them as left and right channels simultaneously, later splitting them on the computer. Here's the patch.

Both oscillators in the Middle Path VCO at lower left feed into its shaper section, producing a complex spectrum on the "cos" output which drives the FM input of one oscillator in the second Middle Path. Then its sawtooth output drives the FM input of the other oscillator, and the square output of that one is a sufficiently messy spectrum to sound like noise. I run it through the Coiler VCF and use the high-pass output. The resulting sound is sensitive to the settings of all the oscillators and the filter; by adjusting the knobs I can get something like white noise, the coloured noise I ended up using, or something with more of a pitch to it. Then the snare sound just uses a Quad VCA channel to apply an envelope to this spectrum.

The bass drum sound is basically just a filter ping. Triggers from the MIDI interface activate a Transistor ADSR, which applies an envelope to the snare spectrum, the same as with the snare sound but with different attack and release settings. That becomes input for the Leapfrog VCF, which is set to feedback mode and almost enough feedback to oscillate, at a barely-audible pitch in the low bass range. The output down there isn't strong, so it goes through a Transistor Mixer patched with feedback of its own to boost the signal; then into a Clouds module, whose input stage is driven into distortion by the very strong pulse of low frequencies. Much of the audible content of the kick drum sound in my tracks is actually the higher frequencies created by that distortion. This may seem an unnecessarily complicated bass drum patch, but given that I wanted to record it simultaneously with the snare patch, it made sense to share parts of the setup between the two. It's also what I ended up with after a couple of failed tries with simpler patching; once I got a sound I liked I didn't want to mess with it.

The patch for cymbal sounds is two voices of basic subtractive synthesis, just with an unpitched noise-like spectrum as input to the filters instead of setting up the oscillators to track. The hi-hat and the cymbal per se are basically the same sound, with different envelope settings. Here I'm using a similar setup to the last patch of the shaper output from two oscillators in one Middle Path feeding the FM input of a third oscillator and then that one's saw output feeding the FM input of a fourth, but here I'm using the pulse output of the final oscillator, with the pulse width adjusted to be narrow for a sound that emphasizes the treble. That gets multed into two Coilers, and then two Transistor ADSRs per channel control the VCA and filter envelopes separately. I'm using the bandpass outputs of the Coilers. The sound from this patch is perceptibly "electronic," not really what a real cymbal would sound like, but given that I'd chosen not to use a noise generator I think it works well in the finished tracks.

Lead

For the lead sound, I built a patch that could record two rhythmically independent voices at once, with very similar sound to each other but different stereo panning, making use of the multiple channels in the Quad VCA. That way I could record the entire lead part of the music (containing at most two simultaneous notes) for the whole album in a single pass.

In this patch, the gate control voltages each activate two Transistor ADSRs, for amplitude and filter cutoff. Pitch CVs are multed to one oscillator and one Coiler VCF per voice.

For each voice I'm using both the oscillators in a Middle Path, tuned close together with the pulse outputs patched into the shaper sections overriding the default triangle patching. One oscillator gets pulse width modulation from a channel of the Fixed Sine Bank, creating a bit of chorus sound. The Middle Path shaper section basically just works as a mixer on pulse waves in this configuration, rather than creating more complicated distortion. Then the output from the oscillators goes into one Coiler per voice, with an envelope on the filter cutoff.

The patching around the Quad VCA may be of interest: it is serving here as a dual stereo VCA with manual panning. The audio and envelope for voice 1 go into a multiple and are then patched into channels A and D of the VCA. The audio and envelope for voice 2 are patched into channel B and then the normalling applies them to channel C as well. With the A+B output treated as "left" and the C+D output as "right," the channels are respectively voice 1 left, voice 2 left, voice 2 right, voice 1 right. When making the recording I panned one voice a little to the left of centre and the other a little to the right by adjusting the VCA knobs, letting the module do the mixing.

Bass

The patch for the bass is another subtractive patch, and it uses fewer modules because it's a monophonic patch: I wrote the basslines to contain only one note at a time.

For this part I wanted a darker, grittier sound, so I'm using the Middle Path in "firm sync" mode. Although both oscillators in the module will track the pitch of the note (through the normalling of the V/oct input), the master oscillator on the left is the one that determines the main pitch in this mode. The slave, on the right, affects timbre rather than pitch, even with a significant amount of exponential FM from the Fixed Sine Bank. "Firm" sync on the Middle Path also uses the pulse output from the master oscillator, so pulse width modulation on the left also affects the sound. The result is a complicated and changing spectrum with a lot harmonics that still tracks with the selected pitch - well-optimized for the tiny PC speakers on which I mixed the album, which make purer bass tones harder to hear.

Only one channel of the Quad VCA is needed for this patch, so in a larger setup, the other channels would remain available for other voices or stereo shennanigans.

Pads

I'm a big fan of chord progressions and "pad" sounds, as I've written before in my articles on Fat sounds and thick pads, Ambient chord progressions, and so on. So I definitely wanted to include some sounds of this kind in the album tracks - but it also meant that I spent a lot of time tweaking the patches for the pad sounds, and I eventually had to call time on the project and go with the recordings I had even though they weren't perfect for all tracks. I recorded two different patches, with multiple recordings to cover all the notes in each chord, and then mixed them all down on the computer, using one patch for some tracks, the other for others, and both patches for a couple of tracks. That is not even counting the earlier patch experiments that I recorded but didn't use. If I had had other goals for the project, and unlimited time, I would likely have ended up with a unique pad or chord patch for every song.

The first of the two patches I ended up using looked like this. This one is intended to create a very "fat," diffuse, and evolving chordal background that comes from all directions at once without grabbing too much direct attention to itself. I used it in tracks 1, 2, 3, 7, 9, and for parts of tracks 6 and 12.

Although I played this as a monophonic patch, it's best understood as two subtractive synth voices mixed together. Each filter and oscillator has a separate envelope, all quite slow but with different settings, so that the sound will evolve over the course of each note as different parts of the sound come in at different times. I also use the "sin" and "cos" outputs of the Middle Path shapers, which have identical spectra but different phasing, as left and right channels to create a pseudo-stereo effect.

The gate and pitch CVs from the MIDI interface are multed to drive all four ADSRs from the gate, and all the VCOs and VCFs from the pitch. Note that the passive multiples I'm using are normalled, so patching into one input and not the other in the 2HP module means the single input will drive all six outputs.

In the first sub-voice, the pulse outputs from both sides of the Middle Path are taken out to Leapfrog filters before being patched back into the oscillator's sine shaper, replacing the default normalling from the oscillator triangle outputs. This patching creates the opportunity for the spectrum to evolve more over the course of the note (through the envelopes applied to the Leapfrogs) and also to pulse width modulate both oscillator signals with sine waves from the Fixed Sine Bank. The "sin" and "cos" outputs of the shaper then go to two channels of the Quad VCA.

The other sub-voice uses no filter, just the default normalled patching of the Middle Path's sine shaper. With the level knobs turned low, that creates a clean enough spectrum not to need filtering. The third (centre) input of the shaper is driven by another sine wave from the Fixed Sine Bank, creating a variable phase shift that further expands the pseudo-stereo effect. The outputs from this Middle Path go to the two remaining Quad VCA channels.

The Quad VCA is being used here in a stereo configuration. The first sub-voice left and right signals are on channels A and D respectively, and the second sub-voices's signals are on channels B and C. The first and second ADSR amplitude envelopes are applied (via multing and normalling) to those same pairs. Then the VCA's A+B and C+D outputs are left and right for the mixed signal. As a final step it goes through a Mutable Instruments Clouds module (stock firmware, delay mode, configured as a reverb and spatializer) to further enhance the motion and stereo effect.

After recording that and trying some experimental mixes I found that for some of the tracks it was just too diffuse and slow; in some tracks I really wanted more aggressiveness in the chords, especially for tracks that used the chords not so much as pads but to establish a rhythm. So I also made the following patch and used it for the remaining tracks: 4, 5, 8, 10, 11, and parts of 6 and 12. This is a duophonic patch, meaning that I could record two notes at once and get my full four-part harmonies with only two recordings.

This patch is almost identical to the "lead" patch above: a duophonic subtractive patch using the pulse outputs of the Middle Path sections routed back into their sine shapers. The differences are that the modulation from the Fixed Sine Bank is applied in different places, and (not visible in the diagram) the settings of the ADSRs are different, much slower than for the lead.

In fact, although this patch worked well overall for the tracks where I used it, the slow attacks I set as a compromise for most tracks didn't work well in every track. If I were redoing these mixes, I would probably start with the reggae-styled track 10 ("Above the law"), which would benefit from much more aggressive chord stabs.

Cover art

I used neural network generative models (several different ones) to create per-track cover images. I'm still learning this technique and it's a little hard to control. Most of the results couldn't be confused with human-created paintings yet, and I'm disappointed that it was harder than I had expected to generate higher resolutions, or to keep specific objects in a scene while changing other things. That was especially the case in the first one, at upper left in this montage: what's shown is my very first attempt at making an image for this track, out of about a hundred. I really liked the face on the statue in that image. I tried to keep the face while changing the background, and couldn't! Even with "inpainting" to designate parts of the image as frozen unchangeable and regnerate the rest, all my attempts at changing the background just screwed up the context of the face and made it look worse. Issues with small models and accuracy are also apparent, for instance in the track 3 image: I asked for watercolour style and the small model I used for the image did well with that, but it also oversimplified the way black and white keys work on a piano keyboard, in a way that a human artist or even a larger generative model would not do.

But it was fun playing with the settings, and took relatively little time to generate a stack of acceptable if not great cover images, compared to searching for appropriate stock images some other way. I think this kind of application may be one of the best ones for image generators, because here I don't need the output pictures to have any specific content; I'm just trying to capture a mood with each one.

◀ PREV What's your soldering success rate? || Inrush current and measuring it NEXT ▶