PyMidiInstrumentDefs

What a particular model of MIDI instrument answers to - its controls, its voicing, its note range - read from definitions transcribed from the manufacturers' own manuals.

It is the companion to PyMidiDefs, which carries the MIDI specification's own constants. PyMidiDefs says CC 7 is Volume; this package says a Matriarch's CC 94 switches it between one, two and four voices.

Two kinds of fact

PyMidiDefs holds specification facts. CC 7 is Volume; note 60 is C4. They are true for everybody, permanently, they are transcribed from published standards, and they cannot be wrong about the world.

This package holds model facts. A Minitaur ignores notes above 72; a Matriarch's CC 94 switches it between one, two and four voices. They are true for everybody who owns that model, and they change when the manufacturer ships firmware.

The difference matters when you are deciding how much to trust a number. A model fact is a report, and reports are wrong in the wild. For one parameter on one common synth - the Minitaur's key priority - the manual, the firmware addendum, and the .midnam file everybody shares give three different answers. So every instrument definition carries a source saying which manual and which page it came from, and one imported automatically stays marked unverified until a person has checked it. Sources, below, says where each of the bundled ones came from and gives that Minitaur disagreement in full.

Rig facts are not here and never will be. Which MIDI channel your Minitaur is on, which port it is plugged into, which notes you have chosen to play - those belong to your own project, not to a definition shared by everyone who owns the same box.

And specification facts stay in PyMidiDefs. The MIDI channel mode messages, CC 120 to 127 - All Sound Off, Local Control and the rest - mean the same on every instrument that has them, so a definition does not list them as controls even where a manufacturer's chart prints them. The validator refuses one, and names the PyMidiDefs constant it would have duplicated.

Installation

pip install pymidiinstrumentdefs

Requires Python 3.10 or later. It depends on PyMidiDefs and on PyYAML, since every definition is a YAML file.

To install the latest unreleased code straight from the repository:

pip install git+https://github.com/simonholliday/PyMidiInstrumentDefs.git

Usage

import pymidiinstrumentdefs

matriarch = pymidiinstrumentdefs.load("moog/matriarch")

matriarch.voice.voicing_modes            # (1, 2, 4)  switchable voicing
matriarch.voice.plays_note(60)           # True
matriarch.controls["glide_type"].kind    # 'choice'

# What to send to put a banded control into a named state. Definitions record
# the band boundaries as manuals print them; working out the number is this
# library's job.
matriarch.controls["glide_type"].value_for("exp")   # 106

pymidiinstrumentdefs.available()         # what is on the search path

Some instruments are several instruments at once, each answering on its own MIDI channel. A definition says so with parts, and a control names the part it belongs to - so the same controller number can mean two different things and still be unambiguous:

streichfett = pymidiinstrumentdefs.load("waldorf/streichfett")

streichfett.parts["solo"].channel_for(1)   # 2 - one channel above the strings
streichfett.parts["solo"].takes("notes")   # True
streichfett.parts["solo"].polyphony        # 8 - its own voices, not the strings'
streichfett.controls_by_part()[""]         # the controls on the base channel
streichfett.controls_reaching("strings")   # all 19 - the strings sit on the base channel

A definition also says what its maker calls each group of controls, which is what a page can head a table with. The groups are what the maker's own document groups them into, in the order to show them:

take_5 = pymidiinstrumentdefs.load("sequential/take_5")

take_5.groups["mod_1"]                   # 'Mod 1' - the mod matrix slot, as the panel numbers it
take_5.groups["envelope_1"]              # 'Env 1 (Filter)' - the guide's own words for which is which
list(take_5.grouped_controls())[:3]      # ['performance', 'oscillators', 'glide']

A group with no label is shown by its own name, which is honest: a label is only ever there because a document says it.

Names, and where the files go

A definition is named for its maker and its model, and the name is also where its file sits: moog/matriarch is moog/matriarch.yaml, in a folder named for the maker. Each half is lower case, digits and underscores.

Definitions are looked for in three places, nearest first:

an instruments/ folder beside your project,
your own library, one per person, so a definition you write once is found by every tool you run:
- Linux: ~/.local/share/pymidiinstrumentdefs/ (or under $XDG_DATA_HOME)
- macOS: ~/Library/Application Support/pymidiinstrumentdefs/
- Windows: %APPDATA%\pymidiinstrumentdefs\
the set bundled with this package.

So a file you drop always beats one we shipped. That is the whole answer to adding your own synth: no index to edit, no registration, no pull request. To describe your own synth, or to correct one of ours, write instruments/<maker>/<model>.yaml beside your project. A definition is one self-contained YAML file, and you share it by sending it.

The bundled definitions

Name	Instrument
`ableton/move`	Ableton Move: no controls, because its maker states the absence - "MIDI CC, and MIDI mapping are not supported" - and the one controller it sends is All Sound Off on stop; built against its own manual, which is three years of releases out of date and says nothing about being so
`access/virus_ti`	Access Virus TI: four controls, because the maker's two manuals publish no control map at all - the 2006 one says the list is on its website and the 2013 one deletes even that sentence
`akai/mpc_key_37`	Akai MPC Key 37: one controller number, which is the mod wheel's factory default - the third definition out of one 530-sheet guide about fourteen machines, and the first to find anything in it
`akai/mpc_live`	Akai Professional MPC Live: two controls, 1 and 11, both received, because one user guide covers ten machines and every other controller number in it is one the owner assigns
`akai/mpc_live_iii`	Akai Professional MPC Live III: two controls, 1 and 11, both received, the first MPC Live's answer read again from its own guide - two machines in one book, every other controller number the owner's both ways, and a guide that says a fixed map exists three times without printing it
`akai/mpc_sample`	Akai Professional MPC Sample: no controls at all, because its maker publishes no controller number and the absence is proved rather than assumed
`akai/mpc_xl`	Akai Professional MPC XL: two controls, 1 and 11, both received, read from the MPC Live III's own book - the same answers in every field, and a panel of its own with four MIDI outputs, eight CV/Gate outputs and the one setting only it has, its footswitches
`alesis/sr_16`	Alesis SR-16: no controls, because a 72-page reference manual never names a controller - twelve pads on twelve notes it gives as defaults, a hundred drum sets by program change, and clock and start both ways with stop sent
`arturia/astrolab`	Arturia AstroLab: 37 controls off a table that gives a direction for every row in both directions - and a Section column that cannot be read as a grouping, because the rows are in controller-number order
`arturia/astrolab_37`	Arturia AstroLab 37: the AstroLab 61's table less its two looper rows, 35 controls unchanged from its 1.0.0 manual to its 2.0.0 one - and three pedal rows on an instrument with one pedal input
`arturia/drumbrute`	Arturia DrumBrute: publishes no controller number - its seventeen drum notes are one screenshot that both editions of its manual print, and its program change is in the firmware release notes and in no manual
`arturia/drumbrute_impact`	Arturia DrumBrute Impact: publishes no controller number at all - its drum map is the one set of numbers it states, and it states them inside a screenshot
`arturia/matrixbrute`	Arturia MatrixBrute: 80 controls on 78 numbers - two of them shared by an expression pedal and a macro knob, as its manual says - and 32 that can send 14-bit values on fine halves no page numbers
`arturia/microbrute`	Arturia MicroBrute: thirteen controls, every one of them a setting of an editor whose manual ships inside a software archive rather than on the downloads page
`arturia/microfreak`	Arturia MicroFreak: four paraphonic voices, and 21 controls out of a manual with no MIDI chart in it
`arturia/minifreak`	Arturia MiniFreak: six voices, twelve when paraphonic, and one control its manual never mentions that its firmware release notes do
`arturia/polybrute`	Arturia PolyBrute: 74 controls off a chart set as eighteen small tables three across, whose column positions shift from one block of rows to the next
`arturia/polybrute_12`	Arturia PolyBrute 12: the same 74-number chart as the PolyBrute, asserted against it number for number - plus controller 74, the MPE Slide dimension, which that chart does not list
`asm/hydrasynth_explorer`	ASM Hydrasynth Explorer: 112 controls off a chart its maker prints twice, once by module and once by number - and four more the maker puts on numbers the MIDI specification reserves for channel mode, which this format cannot hold
`asm/leviasynth`	ASM Leviasynth: 109 controls off a complete 128-row chart the maker prints three times across two manuals, all three agreeing row for row - and one definition covers the Desktop and the Keyboard because of it
`behringer/deepmind_12`	Behringer DeepMind 12: 269 controls from a retailer's copy of Behringer's own manual, which Behringer does not serve - 105 controllers and 177 NRPNs, on one control only where both tables print one name, thirteen times
`behringer/edge`	Behringer EDGE: no controls and deliberately not `none` - everything its maker publishes for it is a quick start guide, and the 16-voice poly chain its page sells is documented nowhere but a diagram
`behringer/lm_drum`	Behringer LM DRUM: 17 controllers, all received, and sixteen drum notes from the manual for firmware 1.1.5 - the quick start guide in the box numbers six of them for other drums, from before every drum could be tuned - and each drum playable at pitch on a channel of its own
`behringer/model_d`	Behringer MODEL D: no control changes at all; its remote surface is SysEx
`behringer/neutron`	Behringer NEUTRON: two controllers, and the maker counts them for you - everything else this semi-modular answers to is system exclusive, and a note outside its stated range is clamped rather than silenced
`behringer/pro_800`	Behringer PRO-800: 70 controls out of a table whose six columns say what each number is - 34 of them 14-bit, paired to a fine half by name alone, at three different offsets and none of them the MIDI specification's
`behringer/rd_6`	Behringer RD-6: eight drums on eight fixed notes and eight voices at once, out of a quick start guide's MIDI message table that names no controller - `none` by enumeration, as for the TD-3
`behringer/rd_9`	Behringer RD-9: eleven drum notes, ten voices at once, a start message and a SysEx dump, and no controls - a search of its whole 38-page manual finds no controller, the route the Model D's `none` takes
`behringer/td_3`	Behringer TD-3: no control changes either, established from the maker's own complete MIDI message table - its "Modded Out" sibling adds exactly one and needs a definition of its own
`behringer/ub_xa`	Behringer UB-Xa: no controls, because the maker advertises a comprehensive MIDI implementation and publishes no part of one - two quick start guides covering two products name a controller number nowhere, and the keyboard's leaves its last two pages blank
`dirtywave/m8`	Dirtywave M8: no controls, because every continuous controller it answers to is one its player assigned - the maker prints "10 user defined CCs per instrument" on its own specification list, and the 128 CCs a third party publishes for it is the count of mapping slots a song can hold
`dreadbox/artemis`	Dreadbox Artemis: 84 controls out of a chapter of the manual rather than out of the standalone CC list, which is a firmware behind it - and one of them, number 31, decides what all the others mean
`dreadbox/nymphes`	Dreadbox Nymphes: 82 controls read off two pages that are pictures, on an instrument whose maker ships with control change switched off
`dreadbox/typhon`	Dreadbox Typhon: 98 controllers out of a three-column list on two pages of the manual, which is the only place its maker publishes them - the standalone CC chart on the same download page belongs to a different instrument
`elektron/analog_four`	Elektron Analog Four: 229 controls off an appendix whose maker publishes it three times over, in which 157 parameters can be reached by NRPN and by nothing else
`elektron/analog_rytm_mkii`	Elektron Analog Rytm MKII: 99 controls of the 319 its appendix prints, because the other 220 are 32 machines' names for the same eight numbers
`elektron/digitakt`	Elektron Digitakt: eight audio tracks and eight that only send, with thirty controller numbers meaning more than one thing
`elektron/digitakt_ii`	Elektron Digitakt II: 144 controls off an appendix that numbers two of its own sections twice, and gives four pairs of parameters the same NRPN
`elektron/digitone`	Elektron Digitone: nine parts on nine MIDI channels, where one CC means three things
`elektron/digitone_ii`	Elektron Digitone II: 164 controls over sixteen tracks that each sound or send, sharing sixteen voices - and eleven NRPNs that mean one thing on an audio track and another on a MIDI track
`elektron/machinedrum`	Elektron Machinedrum: 71 controls, which is what its maker prints - sixteen tracks over four MIDI channels, of which it enumerates one track and abbreviates the other fifteen with an ellipsis
`elektron/model_cycles`	Elektron Model:Cycles: the Model:Samples' manual with a synth in it, where four of its 30 controls mean something different on each of six machines
`elektron/model_samples`	Elektron Model:Samples: 29 controls off a one-page appendix headed CC MSB that gives an LSB for exactly one of them
`elektron/octatrack`	Elektron Octatrack: two controller maps where 51 numbers mean different things, one file for the MKI and the MKII because their appendices agree row for row - and sixteen rows the maker puts on numbers the MIDI specification reserves for channel mode, which this format cannot hold
`elektron/syntakt`	Elektron Syntakt: twelve tracks and an FX track, where 28 controller numbers mean one thing on a track and another on the FX track
`elektron/tonverk`	Elektron Tonverk: 242 controls over sixteen tracks in four kinds, where one controller number can name a different parameter on each kind - and 343 more numbers its appendix generates by a rule rather than printing
`ensoniq/esq_1`	Ensoniq ESQ-1: two third-party scans of Ensoniq's own manual, read by eye - four controllers for the keyboard and the same four for each of eight sequencer tracks, which in MULTI mode each answer on a channel of their own
`erica_synths/hexdrums`	Erica Synths HexDrums: ten voices on ten consecutive notes and no controllers at all - the one instrument here whose maker says it chose not to have any, and says why
`erica_synths/perkons_hd_01`	Erica Synths PĒRKONS HD-01: 44 controls off two pages of a manual reached through a news item, whose numbers a player can rewrite in a file on the SD card
`expressive_e/osmose`	Expressive E Osmose: an MPE instrument whose 24 voices each take a MIDI channel of their own, and which ignores velocity entirely
`groove_synthesis/third_wave`	Groove Synthesis 3rd Wave: 514 controls in two maps a setting chooses between - 101 controllers that name the keyboard's own knobs and buttons, and 413 NRPNs that name the settings a program holds - from a specification that covers three models and gives each its own controller table
`groove_synthesis/third_wave_24m`	Groove Synthesis 3rd Wave 24M, the 24-voice desktop module: 517 controls in the same two maps - 99 controllers from a table that grew from the panel's knobs to numbers for each oscillator and envelope, and 418 NRPNs, five of them the desktop module's own
`groove_synthesis/third_wave_8m`	Groove Synthesis 3rd Wave 8M, the 8-voice desktop: 476 controls - 61 controllers and 415 NRPNs - two parts, and the only one of the three whose manual says it takes program change
`intellijel/cascadia`	Intellijel Cascadia: four controllers, and four is all there are - a semi-modular whose MIDI becomes a voltage at eight jacks, two of them driven by whichever controller its owner last taught it
`korg/drumlogue`	Korg drumlogue: 66 controls out of the first of its manual's two complete implementation charts - the alternate map shares 67 of its numbers and 65 of those mean something else
`korg/electribe`	Korg electribe: 18 controls out of a plain text file that prints every number twice, with sixteen parts on one MIDI channel and no way given of choosing one
`korg/kronos`	Korg KRONOS: 72 controls, every one travelling both ways, out of a MIDI document that is prose rather than a chart - and 28 of them switched off on a new instrument until somebody loads them
`korg/m1`	Korg M1: four controls, because everything else about this 1988 workstation is reached by system exclusive - read by eye out of a 138-page scan with no text in it at all
`korg/microkorg`	Korg microKORG: 41 controller numbers that are the factory assignment and not fixed, each meaning one thing in a synth program and another in a vocoder one
`korg/microkorg2`	Korg microKORG2: 181 controls out of a maker who prints the same map three times, where the NRPN table gives six parameters one address and numbers two more twice over, and the manual's own pages are what put all eight right
`korg/minilogue`	Korg minilogue: 39 controls, thirteen of them switches whose named bands the maker gives twice over - as the values it sends and as the bands it reads them in - from an implementation its maker publishes as a plain text file and again, five years later, as a chart in the manual
`korg/minilogue_xd`	Korg minilogue xd: 58 controls and 61 NRPNs, out of a document that inverts two of them between sending and receiving
`korg/modwave_mk_ii`	Korg modwave mk II: one manual covers three models, and the only difference is the voice count - with a chart whose two direction columns had to be proved rather than read in order
`korg/monologue`	Korg monologue: 24 controls out of an implementation printed twice, one direction each, where two numbers are received only and one control's bands are the footnote the maker forgot
`korg/ms_20_mini`	Korg MS-20 mini: no controls, because the only MIDI this 1978 reissue accepts is notes on MIDI channel 1 with velocity disabled - and the maker says exactly that in one sentence
`korg/multi_poly`	Korg multi/poly: nine fixed controllers and twelve that are factory defaults, off a chart page three of whose spans are enciphered
`korg/nts_1`	Korg NTS-1 digital kit: 29 controllers both ways, off a text implementation and a chart that agree on every number, eight of them types whose every value sent falls in the band received
`korg/opsix`	Korg opsix: 30 controls off one chart page whose text is enciphered, five of them recognised and never sent
`korg/triton`	Korg TRITON: 50 controls, the 39 controller numbers among them all travelling both ways, out of a text implementation shaped like the KRONOS's - one definition for the 61-, 76- and 88-key models, sixteen tracks, and an arpeggiator reached by NRPN
`korg/volca_bass`	Korg volca bass: 12 controls that only arrive, from an instrument with no MIDI Out jack - and three of them reach parameters its panel cannot, two of which its owner's manual never names
`korg/volca_beats`	Korg volca beats: 20 controls and ten parts off a plain text file, where the chart beside it draws its yes and no marks and names only seven of the ten
`korg/volca_drum`	Korg volca drum: six parts on six MIDI channels, and the one of its maker's two charts that the instrument answers to out of the box
`korg/volca_fm`	Korg volca fm: eleven controls that only travel inwards, out of the first implementation in this corpus that is a plain text file - and a value table that disagrees with itself
`korg/volca_keys`	Korg volca keys: sixteen controls, all received, and a chart whose ticks and crosses are drawings - its own legend prints `: Yes` and `: No` with the symbols missing
`korg/volca_sample`	Korg volca sample: eleven controls off a chart whose ticks and crosses are drawings rather than text, so the only column that matters is invisible to a reader that extracts the page
`korg/wavestate`	Korg wavestate: 41 of its 50 controller numbers are defaults a player can reassign, and its maker publishes no control-change chart
`korg/wavestation`	Korg Wavestation: eight controls read by eye from three scans, from a chart and two appendices that disagree only about controller 7, which a 1991 addendum settles; sixteen channels in its multi mode
`make_noise/zero_coast`	Make Noise 0-COAST: 19 controllers off a list whose maker says most of them are ignored unless the instrument is in a particular mode - and one row of which is not text at all but a drawing
`modal/carbon8m`	Modal CARBON8M: 106 controls, and a voice count set per patch
`moog/dfam`	Moog DFAM: no MIDI at all, and the file says so
`moog/grandmother`	Moog Grandmother: one voice, five controls in 14-bit pairs, and a value table of 24 clock divisions
`moog/labyrinth`	Moog Labyrinth: answers to notes, clock and transport, and nothing else
`moog/matriarch`	Moog Matriarch: 37 controls, one of which only its firmware notes mention, and a voice count you can switch over MIDI
`moog/messenger`	Moog Messenger: 56 controls, 31 of them 14-bit pairs the specification's own way round, with every band named
`moog/minimoog_model_d`	Moog Minimoog Model D: one control, the fourteen-bit mod wheel on CC 1 and CC 33 - the only controller number in 84 pages, from a maker whose manual points at a SysEx document it no longer offers
`moog/minitaur`	Moog Minitaur: plays notes 0-72, with the firmware v2.1 corrections
`moog/mother_32`	Moog Mother-32: seven controllers, four of which reach no sound at all but a voltage at a jack the player patches
`moog/muse`	Moog Muse: 102 controls off one appendix that four documents print identically, two of whose rows are misprinted - and two timbres on two MIDI channels out of the box
`moog/slim_phatty`	Moog Slim Phatty: 52 controls off one table that gives every stepped setting's exact values - and the manual is older than the firmware Moog ships, whose notes add a controller this file leaves out
`moog/sub_37`	Moog Sub 37: 114 controls, every one of them the same as the Subsequent 37's, established by reading both charts rather than assumed
`moog/subharmonicon`	Moog Subharmonicon: reads a note as an offset from C4, not as a pitch, and answers to one controller its manual does not list
`moog/subsequent_37`	Moog Subsequent 37: 114 controls in CC and NRPN pairs, out of a chart whose empty cells were proved empty rather than unread
`native_instruments/maschine_plus`	Native Instruments MASCHINE+: two controller numbers in 243 pages, both of them ones the MIDI specification had already spoken for - every other one belongs to the owner, and the document that would hold them ships inside an application
`novation/bass_station_ii`	Novation Bass Station II: 94 controls across three products that share one guide, and one controller number the guide prints that the MIDI specification says is something else
`novation/circuit`	Novation Circuit: 299 controls over two synths, four drums and a session, from a guide that prints seventy-four of its rows twice and leaves thirty-nine out - two of the eight macro knobs among them
`novation/circuit_rhythm`	Novation Circuit Rhythm: 74 controls over eight sample tracks and a project, out of the third Circuit here - where 64 of its numbers are also the Circuit's, only eleven of those mean the same thing, and the master filter moved
`novation/circuit_tracks`	Novation Circuit Tracks: 358 controls, the most here, over two synth tracks, four drum tracks, two MIDI tracks and a project - including eight the maker's contents page files as a table of values rather than of controls
`novation/mininova`	Novation MiniNova: 557 controls off a nine-sheet chart, where 105 answer to a controller number and the other 452 only to an NRPN, and no parameter to both
`novation/peak`	Novation Peak: 246 controls off two documents three firmware releases apart, 33 of which carry no default, because the maker's default column does not agree with itself
`novation/summit`	Novation Summit: 189 controls over two parts, and sixty more its guide names with no controller number at all - the fifteen Modulation Matrix slots after the first, blank in three editions of the guide and numbered in full in the sibling Peak's manual
`oberheim/ob_x8`	Oberheim OB-X8: 147 controls out of a user's guide that contains its own MIDI implementation, and contradicts itself about more things than any document here - including printing one page number twice
`oberheim/teo_5`	Oberheim TEO-5: 198 controls off an implementation document that says in its own words the map belongs to a Sequential synth
`polyend/tracker`	Polyend Tracker: 80 controls in two maps that are live in different modes - the performance effects and mixer whenever it listens, the instrument's own parameters only in synthesiser mode - plus one controller the manual's table drops and the instrument's own screen shows
`pwm/malevolent`	PWM Malevolent: from its quick-start guide alone, and says so
`roland/alpha_juno_2`	Roland αJUNO-2: five controllers off a 1986 implementation read by eye from Roland's own scan, carried as the implementation numbers them because its chart prints six controller numbers beside five lines
`roland/d_50`	Roland D-50: the manual twice sends you to a MIDI implementation chart that is not in it, so this carries no controls - only the two ranges its pedals may be set to send
`roland/fantom_6_7_8`	Roland FANTOM-6/7/8: 31 controls, every one of them the MIDI specification's own assignment under its own name, out of a 67-page implementation whose transmit section names five fewer than its receive section and says why
`roland/jd_08`	Roland JD-08: 85 controls over two parts, from the implementation chart at the back of its Reference Manual - which the HTML edition prints again row for row - and a system channel that selects patterns and passes notes to the selected part
`roland/jd_800`	Roland JD-800: seven controls read by eye from two scanned 1991 manuals, whose three implementation charts agree on everything but pan - and the special part's chart prints its controller numbers against the wrong names
`roland/jd_xa`	Roland JD-XA: 176 controls on three kinds of part - four analogue, four digital and the program - each on a channel of its own, where the same controller number means one thing on an analogue part and another on a digital one
`roland/jd_xi`	Roland JD-Xi: 86 controls over four parts on four fixed MIDI channels, 21 of them addressed by NRPN rather than by a controller number - the first Roland here whose NRPN is more than a crossed box
`roland/jp_8000`	Roland JP-8000: 50 controls on two parts and a performance channel, read by eye from a scanned manual - forty knobs at the numbers its second edit mode gives them from the factory, twelve of which its first mode sends as system exclusive instead
`roland/ju_06a`	Roland JU-06A: 39 controls off one page of an implementation chart whose other page is a second whole map - fifteen numbers in both and eleven of them meaning something else
`roland/juno_106`	Roland JUNO-106: two controller numbers in the whole instrument, and everything else it can be told is system exclusive
`roland/jupiter_x`	Roland JUPITER-X: 22 controllers, 37 controls over two kinds of part, from a 90-sheet implementation that holds none of its tone parameters - those are in the Parameter Guide, one table per model, where 21 numbers name more than one thing
`roland/jx_08`	Roland JX-08: 49 controls over two parts and a system channel that selects patterns, from the implementation chart at the back of its Reference Manual - which prints two names twice each, and has no row for one of the panel's two TUNE knobs
`roland/jx_8p`	Roland JX-8P: five controls read by eye from a scanned 1985 manual, behind four MIDI function switches that no page pairs with a message - and the switch table's footnote says a switch only stops sending, while the implementation says it stops receiving too
`roland/mc_101`	Roland MC-101: 28 controls over four tracks and a fifth MIDI channel that makes no sound, from a chart its maker publishes in numbered editions - of which the middle one is missing, so one firmware step was never printed
`roland/mc_707`	Roland MC-707: three editions of one chart served from one path, a control channel that makes no sound, and an effect send two documents put on two different numbers
`roland/p_6`	Roland P-6: 40 controls out of a list its implementation chart only points at - the chart names no controller number at all - and every one of them moved MIDI channel in the firmware this describes, from the auto channel alone to the granular sampler's as well
`roland/s_1`	Roland S-1: 54 controls out of two tables that each hold half the answer - a chart with ranges and no names, a list with names and no directions - and two of them told apart only by a drawn waveform
`roland/sh_4d`	Roland SH-4d: 32 controls off the first of three whole implementation charts its manual prints, where the other two cross control change entirely - so these are the four tone parts' controls and the rhythm part answers to none of them
`roland/sp_404mkii`	Roland SP-404MKII: 27 controls, seven of which the implementation chart publishes only inside a footnote - and one controller number that selects a different effect on each of three buses, so the same value means three things
`roland/tb_3`	Roland TB-3: thirteen controls off a one-sheet chart, seven of them confirmed by a second document that was written about system exclusive - the only controller numbers in this corpus published twice
`roland/tr8s`	Roland TR-8S: eleven voices of four controls, and two it only sends
`roland/tr_08`	Roland TR-08: 38 controls off a one-sheet chart, every number one the TR-8 uses and eleven of them under another name - 86 to 88 reach this machine's cowbell where the TR-8 has `RC` - and sixteen instruments over twenty-three notes that no single range describes
`roland/tr_1000`	Roland TR-1000: 66 controls off a real chart, which is nine firmware releases behind the instrument it describes
`roland/tr_6s`	Roland TR-6S: 34 controls, 33 of which are the TR-8S's - and one the TR-8S has not, named BEAT and explained nowhere
`roland/tr_8`	Roland TR-8: 50 controls off a chart five firmware releases old, eleven instruments over seventeen contiguous notes, and three controls it obeys and never reports
`roland/tr_909`	Roland TR-909: no controls, because its maker's whole published MIDI is one page of a scanned manual that names seven kinds of message and not one number
`sequential/fourm`	Sequential Fourm: 100 controls out of two tables for one set of parameters, a controller and an NRPN apiece, which disagree about six of them and about which neither is reliably right
`sequential/prophet_10`	Sequential Prophet-10: the Prophet-5's 85 controls, because its maker offers the Prophet-5's implementation for it, read again from the same file; ten voices, and since OS 2.0 two layers that nothing published says how to reach over MIDI
`sequential/prophet_5`	Sequential Prophet-5: 85 controls off a real MIDI implementation, which is also the Prophet-10's and is older than two of the instrument's own operating systems - 50 by controller number and NRPN both, 21 by NRPN alone
`sequential/prophet_6`	Sequential Prophet-6 keyboard and desktop module: 113 controls, most reachable by CC and by finer NRPN, from an appendix its maker prints three times over - twice in the English manual and once again in the German - with no two printings agreeing
`sequential/prophet_x`	Sequential Prophet X: 598 controls, because its NRPN table gives every one of 241 program parameters a number for each of its two layers - and its 90 controllers are not paired with them, since no page says which layer a controller reaches
`sequential/take_5`	Sequential Take 5: 170 controls, most reachable by CC and by finer NRPN
`soma/pulsar_23`	Soma Pulsar-23: every note and controller assigned by MIDI learn
`synthstrom_audible/deluge`	Synthstrom Audible Deluge: no fixed controller or note map at all, read from both of the guidebooks its maker publishes - one per display edition - whose last chapter turns out to be a reprint of a user-written guide
`teenage_engineering/ep_133_ko_ii`	teenage engineering EP–133 K.O. II: four controllers and 48 pads, out of a web guide whose implementation chart says yes with a picture and no with a letter
`teenage_engineering/op_1`	teenage engineering OP-1: publishes no controller number at all - four incoming control changes are routed by the player, per sound
`teenage_engineering/op_1_field`	teenage engineering OP-1 field: 52 controls out of two forms of one guide, where the web page carries the whole MIDI reference and the downloadable PDF carries none of it - and the four controllers the page omits are in the PDF
`teenage_engineering/op_xy`	teenage engineering OP-XY: its maker publishes one guide twice, and the two editions give different numbers for the same row
`teenage_engineering/op_z`	teenage engineering OP-Z: 53 controls, every one of them from the guide's incoming MIDI table - what it sends is a number the player sets, parameter by parameter, and no page prints one before they do
`ten_ten_music/blackbox`	1010music Blackbox: sixteen pads on a note map, each pad on a channel of its own if set, three named controllers and the rest learned - and a 16-note pool that two of its five pad modes stand outside
`udo_audio/super_6`	UDO Audio Super 6: 86 controls off the most complete implementation a new maker has brought here - every number from 0 to 127 given a row, and 41 of them carrying an NRPN that is the controller number plus 1024
`vermona/drm1_mkiv`	Vermona DRM1 MkIV: a drum machine that ignores controller data
`vermona/perfourmer_mkii`	Vermona PERfourMER MKII: Vermona's own manual, from a retailer's copy - four synthesiser channels, each one voice and each assigned a MIDI channel, and CC 1 to the pulse width
`voce/electric_piano`	Voce ELECTRIC PIANO: 16 or 32 voices, depending on the chorus
`waldorf/blofeld`	Waldorf Blofeld: a chart with a row for all 128 controller numbers, seventeen of which are not controls - and one manual covering a Desktop with no MIDI out and a Keyboard with one
`waldorf/iridium`	Waldorf Iridium: no control map at all - almost every parameter is reached by MIDI learn, so what is here is the fifteen controller numbers the maker fixes, out of a manual two product pages serve as the same bytes
`waldorf/iridium_core`	Waldorf Iridium Core: the Iridium's fifteen fixed controllers, most other parameters by MIDI learn, two layers on one pool of twelve voices, and program change sent from a Macro button and received through the `Favorites` list on one FAQ entry the three Iridiums share
`waldorf/iridium_mk2`	Waldorf Iridium MK2: the Iridium's fifteen fixed controllers and Master Volume on 7, most other parameters by MIDI learn, four layers on one pool of sixteen voices, program change through the `Favorites` list on one FAQ entry the three Iridiums share, and MIDI 2.0 over USB, which the format has no field for
`waldorf/protein`	Waldorf Protein: fifteen fixed modulation sources and MIDI learn for everything else, out of a manual whose MIDI chapter is a glossary of what MIDI is
`waldorf/streichfett`	Waldorf Streichfett: controls whose values are exact numbers, not bands
`yamaha/dx7`	Yamaha DX7: sends one set of controllers and answers to another, out of a 1983 manual that only exists as a scan
`yamaha/montage_6_7_8`	Yamaha MONTAGE6/7/8: 29 controls over 16 parts, out of a Data List whose maker lists ten editions of it and serves two of those rows from one file - the map is unchanged across the three editions read, and half the controls are the factory assignment rather than a fixture
`yamaha/reface_cp`	Yamaha reface CP: 18 controls out of the same data list as the reface DX, and the only one of its four instruments with a sostenuto and a soft pedal - it does no program change at all, where its sibling does
`yamaha/reface_cs`	Yamaha reface CS: 21 controls out of a Data List that covers four instruments in four self-contained sections - 17 of them answer only while a setting is on, and its pedal socket sends one of two controllers and never both
`yamaha/reface_dx`	Yamaha reface DX: 29 controls out of a data list covering four instruments, told apart by the system exclusive model identifier - and 25 of them answer only while MIDI Control is on
`yamaha/seqtrak`	Yamaha SEQTRAK: 40 controls over eleven tracks, each on a fixed MIDI channel it cannot be moved from - the numbers come from the Data List's implementation chart and the names from the User Guide, because neither document carries both

The set is a starting point rather than a catalogue. moog/dfam is five lines, because the DFAM has no MIDI at all and saying so is worth more than saying nothing. Three others - the Labyrinth, the MODEL D and the DRM1 - are nearly as short for the same reason: somebody read the whole manual and found no control changes, and the file records that it looked.

Starting from a MIDNAM file

If your instrument has a .midnam file - Ardour bundles several hundred - pymidiinstrumentdefs.midnam will start a definition from it:

import pymidiinstrumentdefs.midnam

draft = pymidiinstrumentdefs.midnam.read_file("Moog_Minitaur.midnam")

pymidiinstrumentdefs.midnam.suggested_name(draft)   # 'moog_music/minitaur'
print(pymidiinstrumentdefs.midnam.to_yaml(draft))

What that lands is a draft, and it says so in its own first line. MIDNAM carries a control map and nothing else - no polyphony, no note range, no velocity response - and the numbers it does carry are worth checking against the manual. It stays marked unverified until you replace the source line with what you checked it against.

This is not how the bundled definitions were made. Those came from manuals, and none of them is an import. Sources measures how far a .midnam can be trusted, on the one instrument where both can be compared.

Sources

A definition is only as good as its source line, and that line is the authority - not this README.

matriarch = pymidiinstrumentdefs.load("moog/matriarch")

matriarch.source        # the manual and the pages, in the file's own words
matriarch.sources       # the same, in fields: edition, address, SHA-256, page offset
matriarch.is_unverified # True if nobody has checked it yet
matriarch.warnings      # what the validator thought worth saying

The definitions bundled here were read out of manufacturers' own documents, page by page, and each names the document and the pages it came from. Usually that is the user manual. For the TR-8S and the Take 5 it is the maker's MIDI implementation document, and for the Malevolent a quick-start guide, because that is all there is, and its file keeps to what the guide says. Two were checked against a second source as well: the Minitaur against Moog's firmware v2.1 addendum, and the DRM1's note map against a working implementation of the same machine. The MiniFreak's second source is the maker's own firmware release notes, which carry one control - CC 7, volume - that its manual never mentions, and the file says so where it records it.

None of them was imported from a .midnam file, and none ever will be. The importer is a tool for starting a definition of your instrument; it is not where ours come from.

Three tests hold that line, so it is a property of the package rather than a promise in a README: one refuses to ship a definition that does not name a document and cite pages, one refuses anything still marked unverified, and one refuses anything the importer wrote.

The Minitaur is the worked example of why that second source matters. Its manual prints the key-priority bands as 0-42, 43-84, 87-127 - leaving 85 and 86 assigned to nothing - and the firmware addendum corrects the third band to 86. This package carries 86, and the file says why it differs from the printed table. That is the level of care every bundled definition is held to, and it is why a source line records pages rather than saying "the manual".

Anything imported is a draft, and is not held to that at all. pymidiinstrumentdefs.midnam starts a definition from a .midnam file, and what it lands says imported from <file>, unverified in its own source line, keeps saying it, and is reported by the validator until a person replaces it.

That is not a reason to avoid importing - it is a good way to start and a poor place to stop, and it is worth being precise about which. Comparing the widely shared Moog_Minitaur.midnam against the same instrument's manual and firmware addendum, on the one instrument where this package holds both:

all 37 control-change numbers agree, and
all 17 14-bit pairings agree, which is the tedious half of a definition and the half most easily mistyped by hand;
4 of the 11 comparable band boundaries differ, including the key-priority band it gives as 85 where the manual says 87 and the addendum says 86.

None of those four changes what gets sent, because a band is transmitted as its midpoint rather than its edge. They do change what gets read back: ask that imported definition what a value of 85 means on key priority and it says last, where the corrected file says high. So an import is reliable about which controller does what, and not yet reliable about what its values mean.

And a definition you supply yourself is yours. Files you drop beside your project or into your own library beat the ones bundled here, by design, and this package makes no claim about where their numbers came from.

Writing one, and want the numbers to be right? docs/adding-an-instrument.md is the long answer: where makers actually publish, which kinds of source can be the only source for a fact and which can do no more than disagree with you, how to tell which firmware a document describes when its own date will not tell you, and how to cite it so that a stranger can check every number in the file.

History

These definitions, the loader and the importer were first published inside PyMidiDefs 0.4.0, as pymididefs.instruments, and moved here so that specification constants and instrument reports can be released separately. PyMidiDefs' own history up to that point is in its repository, ending at commit 4abf367.

License

MIT - you are free to use, copy, modify, merge, publish, distribute, sublicense, and sell copies of this software in any project, including commercial and closed-source applications. The only requirement is that you include the LICENSE file when redistributing the software. See LICENSE for the full text.

Metadata

Release files for pymidiinstrumentdefs 0.2.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for pymidiinstrumentdefs 0.2.0
File	Size	Uploaded
pymidiinstrumentdefs-0.2.0.tar.gz	1.5 MB	Oct 10, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for pymidiinstrumentdefs 0.2.0
File	Interpreter	ABI	Platform	Reset
pymidiinstrumentdefs-0.2.0-py3-none-any.whl 1.4 MB Oct 10, 2026	Python 3	none	any	Details

Total release size: 2.9 MB

Release files / pymidiinstrumentdefs-0.2.0.tar.gz

Download URL	pymidiinstrumentdefs-0.2.0.tar.gz
Size	1.5 MB
Tags	Source
SHA-256 checksum How to use checksums	`30b1a37b147b596839da1184d7db68fbf29b3d2ac1b033b5f8f9862ac90bb39c`
BLAKE2b-256 checksum How to use checksums	`5e633bac7a3664e7c62a8f92d6eb47742d88b29f73a557e78049067cdfda0ef1`
Upload date	Oct 10, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/simonholliday/PyMidiInstrumentDefs
Publishing commit: github.com/simonholliday/PyMidiInstrumentDefs/tree/7f661a7f07f3e1885cc3299985fd24a3b3bfd9f5

Workflow

Publishing configuration: github.com/simonholliday/PyMidiInstrumentDefs/blob/7f661a7f07f3e1885cc3299985fd24a3b3bfd9f5/.github/workflows/publish.yml
Publishing logs: github.com/simonholliday/PyMidiInstrumentDefs/actions/runs/38036979535/attempts/1

Artifact

Subject: pymidiinstrumentdefs-0.2.0.tar.gz
SHA-256 checksum: 30b1a37b147b596839da1184d7db68fbf29b3d2ac1b033b5f8f9862ac90bb39c Verifying attestations

Release files / pymidiinstrumentdefs-0.2.0-py3-none-any.whl

Download URL	pymidiinstrumentdefs-0.2.0-py3-none-any.whl
Size	1.4 MB
Tags	Python 3
SHA-256 checksum How to use checksums	`3dfc452e67023212f661d1f1e2a2a1468cf80ba0c6a38a9a3a3a4358e35e4d3c`
BLAKE2b-256 checksum How to use checksums	`4a521305719ece1a90b691843c1a41c6dc80a6123b489f2e1f49face09432f10`
Upload date	Oct 10, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Provenance

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/simonholliday/PyMidiInstrumentDefs
Publishing commit: github.com/simonholliday/PyMidiInstrumentDefs/tree/7f661a7f07f3e1885cc3299985fd24a3b3bfd9f5

Workflow

Publishing configuration: github.com/simonholliday/PyMidiInstrumentDefs/blob/7f661a7f07f3e1885cc3299985fd24a3b3bfd9f5/.github/workflows/publish.yml
Publishing logs: github.com/simonholliday/PyMidiInstrumentDefs/actions/runs/38036979535/attempts/1

Artifact

Subject: pymidiinstrumentdefs-0.2.0-py3-none-any.whl
SHA-256 checksum: 3dfc452e67023212f661d1f1e2a2a1468cf80ba0c6a38a9a3a3a4358e35e4d3c Verifying attestations