Miku-FM guide.

Getting started

• Install. Use the installer page in Chrome or Edge with a USB data cable. To go back, the same page restores the official firmware. If anything goes wrong, hold OCT- while switching on: the USB rescue, from which a firmware installs again. After two crashes at boot the FM-1 goes there by itself. See INSTALL.
• First sound. At power-on the FM-1 loads the slot you last used. The first time it is slot 1, the folk song SAKURA SAKURA. Press PLAY and she sings.
• Pages. HOME steps through ten pages and back to the first: STAGE, DANCE, SONG, PATTERN, MIXER, GEN, LYRICS, LOOK, LIBRARY and BANK. The page name is at the top of the screen.
• Keys. The 27 keys play the current track's instrument on every page. On STAGE, DANCE and LIBRARY that is always the voice. OCT- and OCT+ move the keys two octaves either way (not saved).
• Tracks. FX / SCL / ENV / LFO pick the current track (VOCAL, CHORDS, BASS, DRUMS) on most pages; the lit button is the track. On STAGE and DANCE they shape the voice instead. On LYRICS, LIBRARY and BANK they do the jobs listed under each page.
• Play and stop. PLAY plays the song from the cursor row (on PATTERN and LYRICS: it loops the cursor row). SEQ stops.
• Save. A tap on SAVE (stopped) keeps the song and its look in the current slot. Hold it for the slot menu. See Slots and saving.

How a song is built

• A pattern is one bar of 16 sixteenth steps for one track. Each track has 16 patterns.
• The song is a list of up to 64 rows. A row names one pattern per track (or silence) and carries a section label (INTRO, VERSE, CHORUS, BRIDGE, OUTRO). Rows play one after another; reusing a pattern in several rows is the point.
• The mixer has a level, pan, sound, mute and solo per track, plus swing, tempo and master for the song.
• The vocal sings morae (a, i, u, e, o, ka, ki, ... n): each vocal step carries its mora. The mora gives the consonant and the vowel. The consonant is a short sound at the start of the note (30-70 ms, the same at any pitch or tempo), then the vowel. A live key or a legato note articulates the consonant too, without restarting the note.
• A song also carries its voice settings (STAGE), its DX7 patches (BANK) and, on the device, its look (LOOK).

STAGE

STAGE is the voice's home: the keys always sing here, and every control shapes her voice. Her mouth follows the voice; her face blinks and smiles on PLAY.

• SELECT turns the syllable dial under her face: the pink syllable under the pointer is the one the keys sing (and the one PATTERN writes). Selection is silent, including while holding a key.
• SCL (a short press) switches the language: JAPANESE sings the syllables as written; ANIMAL turns every held note into random babble with little pitch jumps, like an animal-game voice; SCAT sings da, ba, bi, do on eighth notes while a note is held; VOWELS drops the consonants. In ANIMAL, hold SCL and turn PRESETS for the speech rate (0 slow, 100 fast, 50 default; not saved).
• ALGORITHM sets her character, the size of the voice: DEEP, LOW, MIKU (the default), YOUNG, TINY. PRESETS is the tempo.
• ARP opens the song LIBRARY.
• The four knobs work in banks. VOICE (the default): breath, brightness, glide, level. FX: chorus, delay (a dotted eighth at the song tempo), reverb, drive. ENV: attack, release, scoop (sliding up into each note), human (small drifts that make her sound alive). LFO: vibrato depth, rate and delay, tremolo. Press the bank's button again to go back to VOICE. Turning a knob shows a ring gauge and a picture of what it does.
• Hold a button (0.4 s) for a deeper layer; its button blinks, and holding it again goes back to VOICE. Hold SCL (SEL): tension, fry (a growl an octave down), whisper, formant. Hold FX: reverb size, delay time, feedback, chorus rate. Hold ENV: consonant length, consonant level, vowel speed, and ON BEAT (the vowel lands on the beat). Hold LFO: a harmony voice at an interval (off, an octave down, a third, a fourth, a fifth, an octave up) with its own mix, detune and character.

All of this is saved with the song. The CPU figure on STAGE shows what it costs: harmony and the effects together roughly double the audio load.

DANCE

The singer full-body, dancing: she sways on the beat, sings while the vocal sounds, throws her arms up in the chorus and bows when the song stops. The knobs, SELECT, SCL and ALGORITHM do what they do on STAGE. EDIT opens LOOK.

SONG

The arrangement: row number, section, and the four pattern numbers (-- = that track is silent in the row). The playing row has a yellow bar.

• SELECT picks the row. FX / SCL / ENV / LFO pick the column (the track). PRESETS sets the pattern of that cell (-- = silent). ALGORITHM labels the section.
• EDIT inserts a blank row before the cursor, ARP duplicates the row, REC deletes it, GLO undoes (eight steps).
• KNOB 1 loops the song (right = on). KNOB 2 tempo. KNOB 3 transposes VOCAL, CHORDS and BASS by up to 24 semitones. KNOB 4 master level.

PATTERN

The current track's pattern in the cursor row: a piano roll for VOCAL, CHORDS and BASS, a six-line grid for DRUMS. A yellow bar marks the playing step.

• SELECT picks the step; a key writes it. VOCAL: the key is the note, ALGORITHM the mora, KNOB 1 the length. CHORDS: the key is the root, ALGORITHM the quality (MAJ, MIN, 7, MAJ7, MIN7, SUS4, SUS2, DIM, AUG, ADD9), KNOB 2 how it plays (block, arps, strum). BASS: the key is the note, ALGORITHM adds accent or slide. DRUMS: keys 1-6 switch kick, snare, clap, closed hat, open hat and perc on and off at the step, keys 7-12 their accents.
• KNOB 2 is the velocity for VOCAL and BASS. KNOB 3 transposes the whole pattern. KNOB 4 moves the cursor row. PRESETS picks the pattern for this track in this row.
• REC clears the step. EDIT duplicates the pattern into a free one so you can vary it. ARP copies (tap: this step; hold ARP and turn SELECT: a range); ARP + EDIT pastes. GLO undoes.

Edits while the song plays are allowed: single steps change at once; changes to the row list or undo pause the sequencer for a few milliseconds.

MIXER

Four tracks: name, sound, level, pan, mute and solo. SELECT picks the track. ALGORITHM picks its sound: CHORDS bells, E.piano, chip arp, pad, organ, pluck, marimba, clav; BASS FM, sub, chip square, acid, sub drop, hard acid; DRUMS 808, chip, punch, digital. On CHORDS and BASS it goes on past those into the ten DX7 voices. KNOB 1 level, KNOB 2 pan (a device setting), KNOB 3 swing, KNOB 4 master, PRESETS tempo, EDIT mute, ARP solo. REC opens the DX7 BANK.

GEN

The generator writes a song for you, as ordinary patterns and rows: everything it makes can be edited on the other pages.

1. KNOB 1 picks the style. Nine of them: HYPERPOP, FUTURE BASS, JERSEY CLUB, DIGICORE, IDOL POP, then POP, BALLAD, DANCE and CHIP. KNOB 2 is the key and KNOB 3 major or minor. KNOB 4 is the seed (ALGORITHM moves it by 100): the same settings always give the same song. PRESETS is the tempo.
2. REC writes a whole song: sections, chords, bass, a sung melody with lyrics, drums. The five newer styles bring their own tempo and their own DX7 chords and bass. The song you had is replaced (it comes back with GLO).
3. Hold EDIT and turn KNOB 2 (ENERGY) or KNOB 3 (COMPLEXITY) to shape the five newer styles: how open or busy the drums are and how many fills and hat rolls there are.
4. Lock a track to keep it: press its button (FX / SCL / ENV / LFO) a second time on GEN. With a track locked, REC keeps the rows and the locked track and rewrites the rest to fit: lock CHORDS and press REC for a new melody, bass and drums over the same chords (also over chords you wrote).
5. ARP writes a new phrase: the pattern of the current track in the cursor row (SELECT picks the row), a different one each press.
6. Each REC or ARP is a take: EDIT keeps it, GLO drops it and brings back the song as it was. Leaving the page keeps it.

The melody stays in the key and ends each section on the chord's root; the lyrics are random morae (sing-able, not words). GEN settings and locks are not saved.

LYRICS

LYRICS edits the sung syllables of the vocal pattern in the cursor row, four at a time, without touching the piano roll. Each line shows the syllable (or REST or HOLD), its length in steps, its note, and its marks.

• SELECT picks the syllable. ALGORITHM changes it: REST, HOLD (the previous vowel carried on to a new note) and the morae. A key sets that syllable's note and sings it.
• KNOB 1 is the length (1-16 steps), KNOB 2 the note by semitones, KNOB 3 the vocal pattern of the row, KNOB 4 the row, PRESETS the tempo.
• FX / SCL / ENV / LFO toggle the marks on the syllable: WORD (the end of a word), PHRASE (the end of a phrase), SEP (a new onset instead of LEGATO) and HOLD.
• EDIT inserts a syllable, REC deletes it, ARP (tap) copies the whole phrase and ARP + EDIT pastes it over another pattern, so one lyric can sing a different melody. GLO undoes.

A phrase has to fit in one bar of 16 steps. If a pattern has long or overlapping notes, LYRICS asks you to fix their lengths on PATTERN first.

LOOK

LOOK is the singer's stage for the song you have loaded. SELECT picks one of 12 outfits (a portrait, a dancing body, a set of stage colours). Then change what you like:

• KNOB 1: the background (ORIGINAL, GRID, HORIZON, CONCERT, LED WALL).
• KNOB 2: the frame (NONE, CORNERS, RAILS).
• KNOB 3 or ALGORITHM: the stage effect (OFF, FIREWORKS, SPARKLES, CONFETTI, SMOKE, BEAMS). REC fires it once.
• KNOB 4: the effect's intensity. PRESETS: STYLE or WARM colouring.
• EDIT goes back to DANCE. Stop the song and press SAVE to keep the look; a mark on the page shows unsaved changes.

The look belongs to the slot on the device: each of the ten slots keeps its own. It is not part of an exported song file.

LIBRARY

LIBRARY lists the ten slots, four at a time, with the name of each song. Slots 1 to 4 start as four Japanese folk songs, written as ordinary songs you can edit: SAKURA SAKURA, TORYANSE, FURUSATO and KOJO NO TSUKI. Slots 5 to 10 start empty. STAGE or DANCE: ARP opens it; HOME gets there too.

• SELECT or PRESETS moves the cursor. PLAY loads the song under the cursor and plays it. SEQ goes back to SONG.
• FX or EDIT loads it (an empty slot says EMPTY SLOT; EDIT starts a new song there). SCL starts a new song in the slot. ENV restores a factory song to how it was shipped (slots 1 to 4). LFO opens its LOOK.
• REC saves the current song into the slot under the cursor. If another song is there, REC asks a second time: OVERWRITE? REC AGAIN.
• Anything that would drop unsaved changes asks first: UNSAVED! AGAIN. Press the same control again within 3 seconds.

Slots and saving

The FM-1 keeps 10 songs. At power-on it loads the slot you last saved or opened. SAVE (tap, stopped) writes the song and its look to the current slot (SAVED, or SAVE FAILED). Saving erases flash, so it only happens while stopped; the sound drops out for a moment.

Hold SAVE (about a third of a second, or turn SELECT while holding it) for the slot menu: SELECT picks the slot, then, still holding SAVE, press FX to load it, SCL for a new empty song or ENV for the demo 「おとのはこ」. Release SAVE: if you pressed none of those, the song is saved into the slot shown.

A song is 4,508 bytes and CRC-checked: a damaged slot is refused, never half-loaded. Songs carry their own copy of the DX7 voices they use, so they play the same whatever is in your user bank.

BANK: the DX7 voices

BANK is the six-operator FM engine. Chords and bass can each use a DX7 voice instead of the classic sounds. The engine has all 32 DX7 algorithms, feedback, envelopes, LFO, pitch bend, sustain and bass glide. Ten original voices come with Miku-FM: HYPER GLAS and HYPER 808, FUTURE SAW and FUTURE SUB, CLUB STAB and CLUB BOUNC, DIGI PLUCK and DIGI GROWL, IDOL KEYS and IDOL BASS.

Open it with HOME, or press REC on MIXER. SCL picks CHORDS and ENV picks BASS (the BANK page works on those two).

Browse and load

BROWSE is the first page. It shows FACTORY n / 10 or USER n / 32 with the voice name, and the level of each operator.

• SELECT browses silently. ARP switches between FACTORY (the ten originals) and USER (your 32, once you have imported a bank).
• REC loads the voice you see into the track. The keys play it. If the track holds an edited patch, REC asks a second time (REPLACE EDIT? AGAIN) so you do not lose it.
• You can also load voices from MIXER: ALGORITHM on CHORDS or BASS goes through the classic sounds, the ten DX7 voices and, once you have edited, a CUSTOM entry that is your edit.

Edit

With a DX7 voice loaded, PRESETS or EDIT steps through the edit pages, and SELECT or ALGORITHM chooses operator OP1 to OP6. The four knobs edit the four fields named on the screen. The voice becomes CUSTOM, and the song keeps it.

ARP on OP FREQUENCY switches the operator between RATIO and FIXED frequency; on MODULATION it steps the LFO waveform. The keys play the voice while you edit. REC goes back to BROWSE. Stop the song and SAVE to keep the patch in the song.

The user bank

The user bank is 32 voices of your own, kept on the FM-1 (not in the songs) until you replace them. It starts empty. You fill it by importing a cartridge or by storing your own patches:

• On any edit page, press GLO: the page shows REC STORE USER n and GLO NEXT. GLO again picks the next slot; REC stores the current patch there, with its name. Storing into a slot replaces what was in it. You have 8 seconds to press REC.
• With no bank yet, the first store creates one and fills the other 31 slots with INIT VOICE.
• Voice names come with the patch (10 characters). To rename a user voice: stop the song, switch BROWSE to USER with ARP, pick the slot with SELECT and press EDIT. SELECT moves between letters, K1 changes the letter, REC saves and GLO cancels.
• Stores need the song stopped.

Export

In BROWSE press GLO: the FM-1 sends its user bank over USB as a standard DX7 32-voice dump (it says USER BANK SENT, or NO USER BANK TO SEND). Receive it with the bank tool or with Dexed or any MIDI librarian that can record a SysEx dump from a port, and keep it as a .syx file.

Importing a cartridge from Dexed or the bank tool

The FM-1 accepts DX7 SysEx over USB MIDI (version 0.7 or newer): a 32-voice bulk dump (a .syx cartridge) or a single voice. The TRS MIDI jack is for notes only.

1. Connect the FM-1 by USB. Stop any song.
2. Send the dump. From the bank tool: open the .syx, press Connect, choose the FM-1 and press Send. From Dexed: send the cartridge (or the current voice) to the FM-1's USB MIDI port.
3. The FM-1 checks the dump and shows what it would do: IMPORT 32 VOICES? REC, VOICE TO USER n? REC, or VOICE TO EDIT? REC. Nothing has changed yet.
4. Press REC within 10 seconds, on any page, to accept. Let it lapse to refuse.

What each does: a bulk dump replaces the whole user bank and selects USER 1. A single voice goes into the edit buffer of the open BANK edit page (VOICE TO EDIT, the patch becomes CUSTOM), and anywhere else into the user slot chosen in BROWSE (VOICE TO USER n; with no bank yet the other 31 are INIT VOICE).

If something is wrong the FM-1 says so and ignores it: BAD CHECKSUM, BAD DX7 LENGTH or INVALID VOICE n (a cartridge with one bad voice is refused whole). Imports need the song stopped (STOP FIRST). The bank tool has a factory file with the ten Miku-FM voices that loads in Dexed too.

MIDI keyboard in

A MIDI keyboard on USB or the TRS jack plays the current track exactly like the keys: the singer on STAGE, DANCE and LIBRARY, otherwise the selected track (drums use the note modulo 12). The sustain pedal (CC 64) holds the notes of the singer, the chords and the bass, classic or DX7. The pitch bend wheel bends DX7 chords and bass; the classic sounds ignore it. CC 120 and 123 release every held note. MIDI does not write pattern steps, and MIDI clock is ignored.

Other firmwares, limits, and when something goes wrong

Songs live in Dinghy's 10 slots of the FM-1's flash; the current slot, the pan settings, the looks and the user bank live in Dinghy's settings sector and blobs. Miku-FM does not read other firmwares' projects or samples, and saving replaces the settings other Felucca-family firmwares keep: back up before installing (see INSTALL).

Limits of this version: songs can't be renamed on the device (a new song is NEW SONG, a generated one is named after its style and seed; user DX7 voices can be renamed); 4/4 only; one bar of 16 steps per pattern; pan is a device setting, not part of the song file; looks are per slot, not in exported songs. USB video output is not in 0.8. MIDI clock sync and WAV export from the device are not there yet (see ROADMAP).

If the firmware misbehaves, hold OCT- while switching on for the USB rescue (no sound, only the updater).

More detail

CONTROLS · DX7 bank engineering notes · Song format · Install and recovery · What has been measured