leonardoMirabal/frodhes v0.9.4

Frodhes 0.9.4

on GitHub

Frodhes 0.9.4: a physically modelled Fender Rhodes, vibraphone and marimba for the M-VAVE FM-1. Every note is computed live from the mechanics of the instrument; there are no samples.

Download: frodhes.fwsc below (the same file as frodhes-0.9.4.fwsc; the newest is always at https://github.com/leonardoMirabal/frodhes/releases/latest/download/frodhes.fwsc). Try it first in your browser: https://leonardomirabal.github.io/frodhes/

Install

pip install mido python-rtmidi
python3 tools/fm1_install.py frodhes.fwsc

If anything goes wrong, switch the FM-1 off, hold OCT− and switch it on to get the USB rescue, then install again.

New in 0.9.4

• MARIMBA, a concert marimba. A sixth TYPE (ALGORITHM, or K1 on the main page): rosewood bars tuned 1 : 4 : 10,
  each over its tube, struck by yarn or rubber mallets, computed like the vibraphone (no samples). The low notes ring
  longer and swell for a moment just after the strike (the long tubes), the top ones are short and dry. Five
  octaves, C2 to C7, are the real instrument's; every other MIDI note plays too.
  • Its own VOICE pages: VOICE 1/2 STRIKE, MALLET (SOFT, MED, HARD, RUBBER), TUBE, WEIGHT;
    VOICE 2/2 RING, MUTE, NOISE, TUNE. The main page's middle cards are MALLET and TUBE.
  • MUTE instead of a damper (a marimba has none): your hand at the key-up. 0 lets every note ring out (the
    MARIMBA presets start there), up stops it sooner. The pedal still holds; All Notes Off stops everything. It is
    the same setting as VIBES' DAMPER (CC 103), so a sound saved on one keeps its value on the other.
  • No motor: no motor page (SELECT goes from VOICE 2/2 to COMPRESSOR), and the motor's CCs 47–51 change nothing on
    MARIMBA (they are kept for VIBES).
  • Two new presets: F15 MARIMBA (rings out, a little reverb) and F16 MRMB SOFT (soft yarn, a heavier mallet,
    more reverb and a little delay). Program change 14 / 15.
  • RING above its middle (64) makes the low notes ring longer, not louder: as RING rises the notes are turned
    down, the low ones the most (at 127 about 6–9 dB from C2 to G3, 3 dB at C4), so the hardest strike with the
    loudest mallet and tube still stays just under full scale instead of being squashed by the output limiter. At
    64 and under nothing changes.
  • Its level is set 2.5 dB above the Rhodes types, as VIBES, but every strike is held under −0.5 dBFS (the top
    bars' roll-off and RING's ceiling), so a chord comes out about 1.5 LU under the Rhodes F01 (measured on the
    FM-1: −1.50 LU on USB, −1.48 on the jack).
• DISPLAY page (GLO four times): COLOR, the screen's colour, AMBER as before and then 25 pastel hues around
  the colour wheel, from PEACH through the pinks, violets, blues, cyans and greens to LEMON, GOLD and MANGO;
  BRIGHTNESS, for a dark room: the screen and every light dimmed together, 10 to 1; LEDS, which keys light
  (none, the played keys, the played keys and the scale's root keys glowing faintly, or every key lit for playing in
  the dark, a played key going dark). Kept with the settings. After an update the FM-1 looks as before: AMBER,
  BRIGHTNESS 10 and ROOTS are 0.9.3's screen and key lights.
• Notes over MIDI light their keys, from USB and the TRS jack, as the FM-1's own keys do.
• DISPLAY LEDS: a fifth mode, SCALE, after ALL. As ALL, but only the keys of the scale are lit: the keys, white and black, whose own note is in the scale (SNAP and DEGREE alike). A played scale key goes dark; the keys outside the scale stay dark, also when pressed. With no scale on it looks as ALL. Saved LEDS settings keep their meaning.
• DISPLAY COLOR: RED, a full red, appended after MANGO (27 steps). Saved COLOR settings keep their colour.

Changes

• Settings and user presets saved by 0.9.3 carry over, and the FM-1 comes back on the sound it was left on.
• MIDI CC 79 (the bar TYPEs): 0–95 selects VIBES as before (0 and 64 included), 96–127 now selects MARIMBA
  (in 0.9.3 every value selected VIBES). CC 80 still picks the four Rhodes types.
• 16 factory presets (F01–F16); Program Change 0–15 loads them, 16–31 the user presets U01–U16 as before. The
  preset number saved by 0.9.3 (14 factory presets) is moved once, so you come back on the same sound.
• The start screen reads ELECTRIC PIANO + VIBES + MARIMBA.
• Less flash for the same sound: the bar instruments' pulse-rate table is stored in 16 bits (every value fits,
  nothing changes in the sound), about 17.6 kB less flash; the image (with the DISPLAY page) is 487,848 B of the
  581,564 B available.
• The developer CCs 9 (the mode cap) and 116's bar replies (112 / 113) work on MARIMBA as on VIBES.
• MARIMBA's MALLET card: its steps re-spaced between the detents (SOFT, MED, HARD, RUBBER stay where they
  were), so no key's level jumps at the changes of mallet family; VIBES' MALLET is unchanged.

Fixed

• A note held through a TYPE change to another engine (a Rhodes type to VIBES or MARIMBA and back, or VIBES to
  MARIMBA), whose sound the change fades out, stayed silent when struck again before being let go (a second MIDI
  note-on, or an FM-1 key on a note MIDI still held). It now plays on the new type. 0.9.3 had this too.
• Key lights under the sustain pedal: a key let go before the pedal went down no longer lights again when the pedal goes down (its note was only in its silent damped tail, or a MARIMBA bar ringing on). Under the pedal a key stays lit only for a note the pedal holds: one held when the pedal went down, or played while it was down, as long as it still sounds. Rhodes, VIBES and MARIMBA.

Tools

• tools/marimba_va.py (Gate VA-M: the marimba's model against its numbers before any C), tools/marimba_scan.py
  (the host scan, the rolls and stress cases, the musicality checks M1–M7), tools/hil_scan.py --marimba (the FM-1
  suites over USB and the jack), tools/hil_cpu.py --engine marimba, tools/mrb_ref_measure.py (numbers only from
  reference recordings, never copied, played or embedded); tools/gen_bars.py builds both bar types (--vibes-only
  still writes 0.9.3's header); tools/hil_jack.py: a tail that ends by its own decay is judged at its end, the after-mute
  check high-passed at 20 Hz.
• tools/hil_scan.py on MARIMBA: a failing row that matches one of tools/marimba_scan.py's known, user-decided
  items (KNOWN / KNOWN_CARDS) is reported KNOWN with its reason (listed after the suite's table and in the
  json), not FAIL; a USB-jack mismatch still flags it.
• tools/hil_scan.py measures fixed (the sound was right, the measure was not): PHASER SPEED 0 and CHORUS SPEED 0
  on a long take against a DEPTH 0 reference (VIBES and MARIMBA), MARIMBA's CHORUS SPEED 50 / 127 by the same band;
  MARIMBA's NOISE contact by the host gate's measure; MARIMBA's DRIVE TONE read clear of the 44.1 -> 48 kHz
  resampler's image, and on the jack only above the jack's own harmonic floor; MARIMBA's DELAY DIG vs ANA judged by
  the tape's wow and flutter in the echoes' timing (a marimba's echoes have too few overtones for ANA's darker loop
  to show in their spectrum).
• tools/hil_scan.py's USB recorder: the 13–27 ms "gaps" in long runs' USB recordings were the recorder's own stall
  (cutting a take copied the whole block log under a lock), not the FM-1; a cut now copies only its own part.
• tools/bar_scan.py --type 5 scans MARIMBA against its own spec; tools/gen_bars.py --marimba-offset-add DB --out DIR writes an A/B header for listening to another MARIMBA level (never the default tables).
• tools/manual_shots.py makes the MARIMBA screenshots; the emulator's ?selftest&marimba.
• tools/release.py's FM-1 step tests MARIMBA too: hil_cpu --engine marimba (<= 4500 us, nothing shed),
  hil_scan --marimba --source both (the voice suites every row, KNOWN rows passing; the effect suites their clicks,
  takes, drift, bypass and level rows), hil_jack --engine marimba (the jack connected, the tails' silence gate, the
  idle floor) and vib_sampleback --engine marimba (Gate VD-M on every key). hil_scan's MARIMBA level row now
  expects the measured −1.5 LU re the Rhodes (±1).
• tools/bar_scan.py: the strike windows (and S1's guards) start at the audio block a strike sounds in; S8 / S9
  (the cards) are MARIMBA-aware (its short notes); MARIMBA's S10 also measures G3, C4 and E4.

Tested

• Host: make test, make pytest (the MARIMBA tests: Gate VA-M, every card and control audible, the scan and the
  musicality checks, the compatibility with 0.9.3 settings, the held-key fix), every Rhodes and VIBES render byte for
  byte as in 0.9.3 (tests/baselines/renders.sha256), Gate VB-M (the firmware's MARIMBA against its reference model,
  23 cases: held notes from C1 to C8, MUTE, no tube, a roll, the pedal; this caught the top note C8 playing about
  5 dB too quietly, fixed before release).
• FM-1 (the final development build of this branch; the release build is checked again before publishing):
  tools/hil_cpu.py --worst, nine low notes at full velocity with the pedal and every effect on: 4.1–4.2 ms of the
  5.8 ms per audio block on the Rhodes, 3.4 ms on VIBES, 3.4–3.5 ms on MARIMBA, under the 5.0 ms where the voice
  guard would act. The Rhodes and VIBES scans as for 0.9.3; the MARIMBA scan over USB and the headphone jack (every
  voice, decay, roll and effect row; the remaining rows are the known, decided items); the jack: hiss and idle floor
  as the Rhodes', no audible gate on MARIMBA's tails; every key C2–C7 recorded back and compared with the model
  (pitch, decay, bloom, overtones).

Full manual: https://github.com/leonardoMirabal/frodhes/blob/main/docs/MANUAL.md · Built on Dinghy. GPL-3.0.