The digital oscillator

Every MicroFreak sound starts in the digital oscillator. The analogue filter, the envelopes and the LFO are there to shape what it produces. Rather than one fixed design, the oscillator can switch between 22 synthesis models, and four knobs run them all: TYPE picks the model, and WAVE, TIMBRE and SHAPE change its sound in ways that depend on the model.

The oscillator plays from C-2 to G8. The keyboard covers only two octaves of that, so use the Octave buttons to move it up or down.

Wave, Timbre and Shape have two tracking speeds. Slow, the default, moves in small increments for precise sound design. Fast suits big sweeps while you perform. Set it at Utility > Browsing > Osc Knob Speed (Utility: global settings), or on the Device tab in MIDI Control Center. At either speed the knobs also accelerate when you turn them quickly.

The 22 types cover a good deal of synthesis history: classic analogue shapes, wavetables, additive synthesis, physical modelling, FM, granular synthesis, speech and, since firmware 5.0, sample playback. The table lists them in the order the manual describes them, by the name the display shows.

Two outside designers contributed models. Mutable Instruments allowed Arturia to use some of the open-source oscillator designs from Plaits, its popular 2018 Eurorack module. The manual introduces that group straight after KarplusStr, starting with V.Analog. BASS, SAWX and HARM come from Noise Engineering: they are special versions of the models in its Virt Iter Oscillator Platform, a module that ships with the same three models and can be flashed with others.

BasicWaves covers the two workhorse shapes of subtractive synthesis. A sawtooth contains every harmonic, odd and even, which is why it suits bowed strings. A square contains only the odd harmonics, which some ears find more musical. One knob morphs between the two, one adjusts their symmetry and one adds a sine sub-oscillator underneath.

SuperWave makes copies of a waveform and detunes them against each other, the classic recipe for big, lush sounds. Most waveform animators of this kind only multiply a sawtooth. This one lets you choose between four waveforms.

A wavetable stores waveforms in memory and reads them back at whatever speed gives the pitch you play. Each table holds 32 cycles, and each cycle is 256 samples long. Moving through the cycles gives timbral changes an analogue oscillator can’t produce.

A harmonic, or additive, oscillator builds a sound by adding harmonics together and changing the level of each one. Harmo sums up to eight harmonics like a traditional additive design, but it also sums complete waveforms, so it reaches more complex tones.

Kevin Karplus and Alex Strong found at Stanford University that looping a short burst of noise through a filtered delay gives realistic plucked strings and drums. Today we call that kind of technique physical modelling. An exciter, here a strike or a bow, sets a resonator vibrating, and the model imitates the physical properties of a real instrument.

V.Analog emulates the classic triangle, sawtooth and square waveforms, using two waves you can detune against each other. It is the first of the Mutable Instruments models from the Plaits module (The 22 oscillator types).

This Plaits model chains two processes. A waveshaper changes the rising and falling slopes of a wave. It can steepen a triangle’s rise until it becomes a falling saw, or curve the slopes, and every change alters the harmonics. A wavefolder then folds the wave back on itself when its level is pushed, instead of letting it clip. The original idea came from a feature of the Serge waveshaper.

FM synthesis grew out of John Chowning’s work at Stanford in the late 1960s. Modulate a waveform with others tuned to the harmonic series and you can imitate acoustic instruments. Move away from the harmonic series and you get bells and other complex, inharmonic tones. This Plaits model is a simple form of FM, with two sine oscillators each modulating the other’s phase, but it still covers a lot of ground.

Granular synthesis chops a wave into tiny particles, then rearranges, multiplies and layers them. In this Plaits model the particles are arranged into formants and filtered waveforms, which gives vocal, resonant tones.

Chords turns the oscillator into a four-voice chord generator. One key plays a whole chord, and because chord type and voicing sit on knobs, you can modulate the chord like any other parameter. It’s another Plaits model.

To hear inversions, play a C major chord and turn TIMBRE clockwise. The first inversion arrives at about 10, and more follow as you keep turning.

Speech goes back to Texas Instruments’ research in the late 1970s, which produced the Speak & Spell, the first talking toy. Vowels are the open sounds, like “aah” and “ooh”; consonants are the sounds that start, stop and shape them. This Plaits model can sing vowel formants or say words from its libraries, all played from the keyboard.

Press PARAPHONIC and the word is sung by four voices at once. The full list of word categories, and how to pick them over MIDI, is in Speech oscillator words and CCs.

Hit anything and it answers with a cluster of resonances. Usually the material soaks up the energy and you barely hear it, but instruments are shaped so that some resonances ring and others are damped. A modal resonator imitates this: an exciter sets a modelled body ringing, and you can change the body’s harmonic structure and its damping on the fly, by hand or from the matrix. It covers anything from woodwinds to strings and drums.

Noise mixes a noise source with a pitched oscillator. Its speciality is particle noise: tiny fragments of noise that appear when noise is sampled down to a lower rate.

BASS is one of three models from Noise Engineering, adapted from its Virt Iter Oscillator Platform. It is a quadrature oscillator built from a sine and a cosine oscillator. The sine goes through a balanced modulator, and the result is mixed with the modulated cosine. The three knobs push that design in different directions.

SAWX, the second Noise Engineering model, starts with a sawtooth and stretches its harmonic content, for example by modulating its phase with subsampled white noise. A chorus then adds time-shifted copies of the saw.

HARM, the third Noise Engineering model, builds a waveform by adding harmonics, or partials, to a fundamental frequency.

WaveUser works like Wavetable, with two differences: it plays wavetables you load yourself with MIDI Control Center, and SHAPE reduces the bit depth instead of adding chorus.

Firmware 5.0 added four oscillator types that play samples instead of synthesising sound. Sample is the most direct of them: load your own audio with MIDI Control Center, then set where playback starts, how much plays, forwards or backwards, and how it loops. The MicroFreak stores up to 128 samples, 210 seconds in total, and FW 5.0 comes with 52 factory samples.

Scan Grains plays the loaded sample as a stream of short slices, called grains. Each grain gets its own volume envelope so they blend smoothly. The engine travels through the sample from beginning to end at a speed you choose.

Cloud Grains puts the grain position in your hands. Instead of scanning through the sample, it plays grains from the start point you set, with detailed control over how they are generated.

Hit Grains is the third granular engine. It puts a sharp volume envelope on each grain and is designed for percussive granular patches.

Vocoder is the carrier for the MicroFreak’s built-in vocoder. Its waveforms are rich in overtones, which gives the vocoder plenty to work with. Connecting a microphone and playing the vocoder are covered in How the vocoder works.