ATFS1

Get the highlights.

Formant Sequences. Eight formants and their breath, frame by frame, 126 times a second: vowels, words, whole phrases at the pitch you play.

Eighty-eight algorithms, drawn. The FS1R showed them in a few pixels. Here every operator is a box you click, with its level live.

Eight spectral forms. One partial, a group, only the odd ones, a resonance or a formant: each operator draws its spectrum as you turn it.

The front panel works. ATTACK, RELEASE, FORMANT and FM, the data knobs, PLAY and EDIT: the 1U front drives the editor.

Make your own Fseqs. Drop a recording and the ATFS1 tracks its pitch and formants, or type vowels and it builds one. The original could not.

The film.

ATFS1 rendered in 3D, in three shots. The sound is the engine playing Shoo Bee Doo over the Vowel Choir.

Formants and FM in one operator.

Every voiced operator is a carrier inside a window that restarts at each period of the note. Change the window and the carrier and the same operator gives one partial, a group of harmonics, only the odd ones, a resonance or a vocal formant; give it FM and it does what a DX7 does. Next to each one, an unvoiced operator turns white noise into a band around a formant: the breath and the consonants.

The laws come from Yamaha's 1995 patent, the FS1R manuals and the measurements published by the FSVR project on real units; the code is our own and was checked against those numbers.

Voiced operators: 8 per voice, 8 spectral forms: sine, all1, all2, odd1, odd2, res1, res2 and frmt, with skirt, bandwidth and resonance
Unvoiced operators: 8 noise formants per voice, with bandwidth, skirt and resonance; normal, or locked to the pitch or to the voiced formant
Algorithms: the 88 of the FS1R with their feedback loops; FM depth of 3.37 cycles at full level
Envelopes: a 4-stage amplitude EG with hold and a frequency EG on every operator; pitch EG; filter EG
Modulation: LFO1 to pitch, amplitude, frequency and filter, LFO2 to the filter; Formant and FM knob matrices
Filter: per voice: low-pass 24, 18 and 12 dB, high-pass, band-pass and band-reject, with negative resonance and input drive
Formant Sequences: 8 voiced and 8 unvoiced formants plus pitch, 126 frames per second, up to 512 frames

ATFS1 Voice page with the algorithm graph

16

operators per voice, 8 voiced and 8 noise

88

algorithms with feedback

32

voices, every one with its filter

21

Formant Sequences, plus any you make

Hear it.

Every clip is the engine on its own, playing a factory preset from a MIDI file written for it. Nothing added afterwards, only the level matched across clips. Download the MIDI and play it through the preset yourself.

Talking Formants
A Formant Sequence analyzed from a spoken phrase, played at six pitches.

0:12 MIDI
Vowel Choir
Eight formant operators per voice follow a looping o-a vowel, with symphonic chorus.

0:19 MIDI
Shoo Bee Doo
A scat phrase turned into formant tracks and sung back by the synth.

0:13 MIDI
Formant Strings
The FORMANT knob (CC 80) moves three formant centres while the chords play.

0:18 MIDI
Tine Piano
Two FM stacks with a 14:1 modulator for the tine, through the tremolo.

0:23 MIDI
Formant Lead
A mono lead whose mod wheel slides the formants from o to e.

0:13 MIDI
Yah Yah Riff
A rhythmic vowel sequence on a loop, one note per bar.

0:13 MIDI
Formant Scratch
The mod wheel scrubs through a spoken Fseq, back and forth.

0:12 MIDI

A phrase becomes a synthesizer part.

A Formant Sequence stores how the formants of a voice move: frequency and level of eight voiced and eight unvoiced formants, plus the pitch, 126 times a second. Play a note and the operators follow it, so the synth says the phrase at your pitch, at your tempo, looped or scratched with the mod wheel. The ATFS1 comes with 21: vowels, choirs, scat and spoken words.

ATFS1 Formant Sequence page — The Fseq page: the formant tracks of a spoken phrase, the vowel generator and the playback settings.

One operator, eight spectra.

The form decides what the window does to the carrier. Pick one and the spectrum above the envelopes redraws, so you hear and see what the operator gives before it reaches the algorithm.

all1 and all2. A group of harmonics around the operator frequency; the skirt opens it from one partial to a bright band.
odd1 and odd2. Only the odd harmonics, from a hollow tone to a square-like buzz.
res1 and res2. Three lines on the harmonic you choose with the resonance: a whistle that moves through the partials.
frmt. Grains at the note pitch with the carrier on the formant: the vowel colour of a voice, a horn or a string.

ATFS1 Operator page — The Operator page: a resonant operator, its spectrum at A3, its envelopes and every parameter it reads.

What the FS1R never had.

The original asked for an editor on a computer and a lot of patience. The ATFS1 brings the editor inside, and the features owners kept asking for.

Your own Fseqs. Drop a WAV or AIFF and the ATFS1 tracks its pitch and formants; or type vowels and it builds a sequence. The FS1R could not create them.
An Fseq per preset. No limit of six; every preset carries its own, and they export as FS1R sysex.
The whole voice on screen. 88 algorithms drawn with live levels, a spectrum per operator, envelopes you drag, and help that says what each parameter really does.
All 32 voices filtered. The FS1R halves its polyphony with the filter on; here every voice keeps it. An authentic switch brings the limit back.
Smoother sequences. Optional interpolation between frames and speeds from 10 to 1000 %.
FS1R sysex. Import voices and Fseqs from the hardware or other editors, and send your voices back.

ATFS1 Play page — The Play page: presets, the algorithm with live levels, the output spectrum and the Formant Sequence.

Talking, singing, ringing.

8

Voices

Phrases and scat sung back from their formants, a robot, choirs, whispers and a vowel lead.

5

Pads

Formant strings, a glass choir, vowels that morph, resonant sweeps and an odd-harmonic pad.

6

Keys and bells

A tine piano, formant bells, a glass marimba, a clav and an organ.

4

Leads

A formant lead on the mod wheel, a square-like odd lead, resonant brass and a breath flute.

4

Bass

FM, formant, odd-harmonic and wah basses.

5

Sequences and effects

Rhythmic vowel loops, a formant scratch on the mod wheel, wind and space talk.

Tech specs.

Sound engine

8 voiced operators with 8 spectral forms and 8 unvoiced noise-formant operators per voice
88 algorithms with feedback, FM depth of 3.37 cycles
Amplitude and frequency EG on every operator, pitch EG, LFO1 and LFO2, Formant and FM knob matrices
Per-voice filter: LPF 24/18/12 dB, HPF, BPF, BEF, resonance −16 to +100, input drive
Insertion, variation and reverb effects, two-band EQ

Formant Sequences

21 factory sequences: vowels, choirs, scat and spoken words
WAV and AIFF analysis into a new Fseq; vowel generator
126 frames per second, up to 512 frames, loops one-way or round, scratch on the mod wheel
Speed 10 to 1000 %, optional interpolation

Extensions

Full-screen editor: algorithm graph, spectrum per operator, draggable envelopes, inline help
32 voices with the filter on every voice (authentic 16-voice limit as an option)
FS1R sysex: import voices and Fseqs, export voices and Fseqs
Working 1U front panel with LCD, perform and data knobs
On-screen keyboard and output spectrum

Formats

VST3, Audio Unit (macOS) and standalone app

Minimum

macOS Mac with Apple silicon (M1 or later), or Intel Core i5 quad-core from 2017 or later, with macOS 11 Big Sur or later.
Windows PC with Intel Core i5-6500 (6th generation, 2015) or AMD Ryzen 5 1600, or better, with Windows 10 or 11, 64-bit.
4 GB of RAM, a 1280 × 800 screen and a 64-bit host for VST3 or Audio Unit, or the standalone app.
Disk space: 80 MB on macOS (all formats), 24 MB on Windows.

Measured CPU load

Setting	Apple M4	AMD Ryzen 9 5900XT
One voice speaking a Formant Sequence	0.4 %	0.7 %
16-voice vowel choir with chorus and reverb	6.8 %	13.9 %
32 voices of a morphing vowel pad, 16 operators each	8.7 %	21.9 %

Percentage of one CPU core at 48 kHz, sound engine only. With the editor open, the interface adds about 4 % on the M4. Hosts run each instrument on one core, so single-core speed is what counts.

Download ATFS1 1.0.2.

Free, no account, no activation, no telemetry.

macOS

macOS 11 or later. Apple silicon and Intel in one universal binary. VST3, Audio Unit and standalone app. 43 MB.

Download for macOS

How to install

Unzip the file.
Copy ATFS1.vst3 to ~/Library/Audio/Plug-Ins/VST3.
Copy ATFS1.component to ~/Library/Audio/Plug-Ins/Components.
Rescan plugins in your DAW. In Ableton Live: Settings, Plug-Ins, Rescan.

Windows

Windows 10 or 11, 64-bit. VST3 and standalone app. 14 MB.

Download for Windows