Choir Machine is a “robot voice” choir synthesizer inspired by the sound of the choir in Wendy Carlos’s 1972 soundtrack to Kubrick’s “A Clockwork Orange”. In Beethoven’s “Ode to Joy”, she used a Spectrum Follower, a kind of early vocoder which turned a singing voice into a robot choir.

——————————————————-

CHOIR MACHINE
A formant choir for the Critter & Guitari Organelle
User Manual – Version 0.28

===============================================================
1 WHAT IT IS
===============================================================

Choir Machine is a formant synthesiser. It does not play back
recordings of singers; it builds each voice the way a voice is
actually built, by running a pitched glottal source through a set of
resonators tuned to the frequencies that define a vowel. Move those
resonances and the same note becomes a different vowel on the same
breath.

Eight voices sing at once. Each has its own vibrato rate, its own
small detune, and its own scattered attack, and that scatter is the
whole trick: eight voices that start at exactly the same instant
sound like one organ, and eight that start within thirty
milliseconds of each other sound like a choir.

Three modes share those voices. In syllable mode every key press
picks a random syllable from a weighted table and sings it with a
real consonant onset. In drift mode there are no consonants at all
and each voice wanders slowly through vowel space on its own
schedule. In vocoder mode the choir becomes the carrier for whatever
is plugged into the input.

AT A GLANCE

* Voices – 8, allocated as an ensemble
* Modes – Syllables (green), Drift (blue), Vocoder (magenta)
* Controls – 4 knobs, single page
* MIDI in – Omni; responds on all channels, velocity sensitive
* MIDI out – Keyboard echoed on channel 1
* Range – Four octave settings, -36 to 0 semitones

COMPATIBILITY

Developed and tested on an Organelle M running Organelle OS 4.1. The
patch is plain Pure Data vanilla – no externals, no extra libraries,
and nothing newer than the Pd 0.49 that the Organelle OS ships – so
it needs no OS update to run.

Original Organelle (Organelle 1): untested, and I have no OG here to
try it on. It should load, since nothing in it is version-specific,
but it was written with the M’s processor in mind and eight voices
plus a twelve-band vocoder is a lot of DSP for the older single-core
machine. If you run it on an OG, please report back – I would like
to know, and a reduced-voice build is easy to make if it turns out
to be needed.

ORAC: there is no ORAC version. An ORAC module is structured
differently from a standalone patch – different host conventions,
parameter pages instead of four fixed knobs – so this would be a
real port rather than a rename, and it would then share the CPU with
whatever else is in the chain. Not planned at the moment, but I am
not against it.

===============================================================
2 GETTING SOUND OUT OF IT
===============================================================

PLAYING FROM A COMPUTER OR A MIDI KEYBOARD

Connect a MIDI output to the Organelle MIDI In. Choir Machine
listens omni – on every channel at once – so nothing needs
configuring on the patch side.

That omni behaviour matters when you play a multi-track MIDI file at
it. A General MIDI arrangement will send every part, including drums
on channel 10, into the same eight voices. Mute everything except
the parts you actually want sung. Four-part vocal writing is the
natural fit: each part gets two voices, with the allocator
rebalancing as parts enter and leave.

VELOCITY

The Organelle’s wooden keys are not velocity sensitive – they send
the same value for every key press – so playing the patch standalone
gives you an even, unaccented choir, exactly as it always has.

An external keyboard, or a MIDI file with dynamics written into it,
gets the full range. Velocity controls two things. It sets how loud
the note is, on a curve with a floor under it so that a very soft
note is quiet rather than missing. It also shortens the attack, so a
hard note arrives sooner as well as louder – which is most of what
dynamics actually feel like under the fingers, and what a real
singer does when asked for a loud note.

The calibration is anchored so that the wooden keys land at
precisely the level the patch had before velocity existed. Nothing
about standalone playing changed. A full-force note from an external
keyboard is about a third louder than anything the Organelle keys
can produce.

===============================================================
3 THE CONTROLS
===============================================================

One page, four knobs. Knob 4 changes its job with the mode, and the
fourth screen line always tells you which job it currently has.

KNOB 1 – SUSTAIN

Attack and release together, weighted so that the release does
nearly all the audible work. The scale is exponential, so the knob
covers a very wide range of times without a useful setting hiding in
a corner of the travel: roughly a tenth of a second at the bottom,
about three seconds at the middle, half a minute around 90, and a
full minute at the last numbered position. At the very top the
screen reads Sustain inf and notes hold indefinitely – useful for
building a drone by hand, but remember that a held note still
occupies a voice.

KNOB 2 – VIBRATO

A probability, not a depth. At zero no voice vibratos; at full every
voice does. In between, the knob sets how many of the eight are
vibratoing at any moment, and which ones changes from note to note.
This is deliberate – a real section has some singers holding
straight tone while others do not, and a uniform depth control
applied to all eight sounds synthetic.

KNOB 3 – OCTAVE

Four fixed settings rather than a sweep, so you cannot land between
them. The screen shows the offset.

* Octave -2 (-36 semitones): bass register; formants are far above
the fundamental, so vowels are very clear
* Octave -1 (-24 semitones): tenor and baritone
* Octave +0 (-12 semitones): alto and mezzo; the default
* Octave +1 (0 semitones): soprano; high fundamentals start to skip
past the first formant

KNOB 4 – THREE JOBS

The label on screen line 4 changes with the mode.

* Syllables mode, screen reads “Consonants”: consonant loudness and
length together. Low is a smooth legato choir; high is percussive,
breathy and much more consonantal.
* Drift mode, screen reads “Drift n s”: average seconds per vowel
step, roughly 1 to 30. Nothing else – breath is fixed in this
mode.
* Vocoder mode, screen reads “Input”: input gain into the analysis
chain. Set it by ear until the vocoded result is present without
crackling.

===============================================================
4 THE THREE MODES
===============================================================

Press Aux to cycle. The LED tells you where you are: green for
syllables, blue for drift, magenta for vocoder.

GREEN – SYLLABLES (default mode)

Every key press draws a syllable from a table of fifty-six, weighted
by how common the sounds actually are in speech, so you hear
plausible ones often and odd ones occasionally. Each syllable
carries a consonant onset and one, two or three vowel targets – a
plain vowel, a diphthong, or a long winding one.

Because every voice draws separately, an eight-note chord is eight
different syllables arriving at eight slightly different instants.
That is what makes it sound like a choir singing text rather than a
synthesiser playing a chord.

Best with: moving parts and repeated attacks. Consonants need
note-ons to exist, so a mode that lives on articulation rewards
music that articulates.

BLUE – DRIFT

No consonants, no syllables. Each note enters vowel space at a
random point and then walks through it, one target at a time, on its
own schedule. Knob 4 sets the average seconds per step; every voice
picks its own interval around that average, so the eight never move
together.

The stepper deliberately alternates between the front half of the
vowel space and the back half, so consecutive targets are always far
apart. On each step, a voice reaches its target in the first quarter
of the interval and then holds it for the rest. That hold is what
makes the vowel audible as a vowel – a voice permanently in transit
between two targets only ever sounds like the mush between them.

Drift uses its own vowel set, deliberately more extreme than the one
syllable mode uses, with German qualities: roughly the vowels of
Liebe, ueber, Leben and Baer on the front side, and Vater, Sonne,
Sohn and Buch on the back. It also sharpens the first two formants,
which are the two that carry vowel identity. Both changes exist for
the same reason: syllable mode has consonants to mark where one
sound ends and the next begins, and drift mode has nothing but the
vowels themselves.

Best with: sustained part-writing. Slow four-part music – Dowland,
Tallis, anything with long suspensions – gives each vowel time to
register before it moves on.

MAGENTA – VOCODER

A twelve-band vocoder from 180 Hz to 6.4 kHz, with the choir as
carrier. Plug a source into the Organelle input, hold keys to give
the vocoder something to sing with, and speak or play into it. Knob
4 is input gain.

The output is gated by the carrier, so nothing leaks through when no
key is down. Voiced and unvoiced sounds are detected separately and
a noise carrier handles the consonants, which is what makes speech
intelligible rather than just rhythmically modulated. Odd and even
bands are pushed slightly apart in the stereo field.

Best with: a close microphone and clear diction. Vocoders reward
over-articulation; speak more crisply than feels natural.

===============================================================
5 HOW IT BEHAVES UNDER THE FINGERS
===============================================================

ENSEMBLE ALLOCATION

Notes arriving within thirty milliseconds of each other are treated
as one chord and the eight voices are shared out between them,
rather than the first note grabbing everything and the rest going
hungry. Play a four-note chord and each note gets exactly two
voices; play one note and it gets all eight.

As you add notes the allocator rebalances, taking voices from
pitches that have more than their share. Release notes and the freed
voices are handed back to whatever is still held. This happens
continuously, which is why a sustained chord thickens audibly as you
lift fingers off it.