FLEX · voices + sequencer

Mac keyboard

• Notes: A W S E D F T G Y H U J K O L P ; = C up to E, a piano laid across the home row.
• Z / X octave down / up. C / V velocity down / up (hold Shift while playing for an accent).
• Space PLAY / STOP · Return START · 1–4 select track · R REC · , MUTE · . SOLO · hold / STUTTER · Esc all notes off.
• Poly (by MIDI): 4 voices plays columns 5–8 with the selected track's sound while the sequencer keeps tracks 1–4; 8 voices uses all eight and stops the sequencer.
• Click the page once first, so it has keyboard focus. Notes play the selected track's voice.

Sequencer

• Row A, step keys: click = step on / off. Shift+click cycles normal → accent → slide → both (§9).
• Under each step: drag NOTE and GATE up / down (or scroll); click TYPE to step through tie, rest, skip, ratchet ×2–×6, probability 75 / 50 / 25 %, octave up, random.
• REC on + a note from any keyboard: stopped, it writes at the cursor and advances (JX-3P style); running, it writes into the playing step.
• Chord track (polyphonic, JX-3P style): under the tracks. WRITE starts a new sequence: hold a chord (up to 6 notes) and let go, and that is one step; TIE holds the last chord a step longer, REST is a silent step, ◀ BACK erases a step. OVERDUB adds notes to the existing steps from step 1. It plays with PLAY, on voice columns 5–8 (all 8 with Poly 8) using the selected track's sound. KEY TRANSPOSE, then a key, transposes it (C3 = none).
• Clearing: Clear steps (this track) and Clear all tracks in the track settings; CLEAR on the chord track.
• Mute / solo / stutter, quickest: the M, S, ST buttons on each track card (press PLAY first so you can hear it). Shift+ST = only that track stutters.
• As on the hardware: MUTE or SOLO (or , / .) latches; then tap tracks or step keys 1–4, or type 1–4. STUT on screen latches too: press it, tap tracks, press it again (pressing it on and off with no taps releases every stutter). On the Mac keyboard, hold / and type 1–4: the real hold, incl. the 0.6 s long hold = only that track. INTRO / STUTTER sets the loop window, 1–16 steps (§14).
• B MODE → KEYS: row A = sharps, row B = naturals, C to D (§13).

Page 4 · WAVE: VCO D, wavetables, PPG modes, joystick, speech

• VCO D runs on the voice's digital lane with VCO C; the firmware sums them, so a 4th oscillator costs no hardware. A, B, C and D are the 4 partials the Synclavier mode will use.
• Wavetables on C and D: TABLE 1–8 are FLEX originals, 9 is the E950 bank; 10–16 take your own .wav tables (.wav… under the joystick; Serum-style 2048-sample frames are detected, 128 = PPG wave size). WAVE is the position (PPG wave number 0–63); ENV, KEY and LFO 2 sweep it. D can be fixed, follow C with an offset (the PPG sub), or sweep on ENV 2.
• ENGINE: MODERN (smooth, band-limited) · WC360 (stepped waves, clock follows the pitch, RC tracking filter, no panel filter) · WAVE 2 (as WC360, into the CEM3320 stand-in) · WAVE 2.2 / 2.3 (8-bit words, a fixed-rate DDS at 183.8 kHz with truncated addressing, waves 60–63 = tri / pulse / square / saw, then the upper wavetable). The 2.3 plays wavetables through the top 8 bits of its 12-bit DAC, so it sounds like the 2.2 here.
• Joystick: drag it (VEC turns on): A left, B top, C right, D bottom, equal-power. On the main unit alone, XF1 = X and XF2 = Y. LFO→X and ENV→Y animate it, like a vector envelope.
• SPEECH (VCO D): TALK turns VCO D into a TMS5100-style LPC chip, the Speak & Spell family: chirp or noise through a 10-stage lattice, 25 ms frames in 8 interpolation steps, with the chip's own tables (from MAME, BSD-3). bent: the note sets the chip clock, as on a circuit-bent unit, so formants move with pitch. in tune: 8 kHz, and the note picks the nearest of the chip's 32 pitch periods (52–195 Hz). D's WAVE picks one of 16 phonemes (IY IH EH AE AH AW UH OO U ER M N Z S SH F), so ENV 2 or follow-C sweeps make it talk. SPEED stretches the frames; GLITCH bends the "ROM"; VOICE = whisper / auto / buzz. The phonemes are FLEX's own, from formant data quantized to the chip's tables: no TI vocabulary.
• Your own Speak & Spell ROMs: VSM… loads TMS6100 VSM dumps (16 KB each; pick both halves together). They are identified by CRC against MAME's list, which also picks the chip tables (TMC0281 / 0281D / CD2801). Words are found through the ROM's pointer table. Set TALK to ROM bent (C4 = the original speed and pitch; other notes bend the chip clock) or ROM sung (8 kHz, the note replaces the word's pitch). D's WAVE picks the word, a note plays it, LOOP repeats it, GLITCH corrupts it. Nothing from TI ships with the simulator.
• Table 9 · E950: the wave bank of Paul Schreiber's Synthesis Technology E950 Circuit-Bent VCO (32 × 256, from the E951 repo, MIT, © 2023 Eric Brombaugh). WaveEdit / E352 / E370 banks load with .wav… (64 × 256 is detected).
• The PPG ROM wavetables, TI speech ROMs and GPL WaveEdit banks are not included. Load tables you have the right to use.

Controllers

Every MIDI input is heard at once. The profile comes from the port name (or pick one in the Controller menu): a keyboard plays the selected track; the two below get their own maps. No editing on the controllers is needed: both maps use the factory settings.

How far to trust it

Each filter solves the same circuit equations as a SPICE model with real differential pairs and the board's part values, and matches it on every check in the table below. They are topology models, because the chips' internal schematics aren't published; bench recordings of real chips are the next step.

The sequencer core is the same C file the timing processor would run; its tests check timing to one sample, swing, polymeter, ratchets, ties, slides, skip, mute, solo and stutter.