Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Table of Contents

Python MIDI

Python, for all its amazing ability out of the box, does not provide you with an easy means to manipulate MIDI data. There are probably about ten different python packages out there that accomplish some part of this goal, but there is nothing that is totally comprehensive.

This toolkit aims to fulfill this goal. In particular, it strives to provide a high level framework that is independent of hardware. It tries to offer a reasonable object granularity to make MIDI streams a painless thing to manipulate, sequence, record, and playback. It's important to have a good concept of time, and the event framework provides automatic hooks so you don't have to calculate ticks to wall clock, for example.

This MIDI Python toolkit represents about two years of scattered work. If you are someone like me, who has spent a long time looking for a Python MIDI framework, than this might be a good fit. It's not perfect, but it has a large feature set to offer.


  • High level class types that represent individual MIDI events.
  • A multi-track aware container, that allows you to manage your MIDI events.
  • A tempo map that actively keeps track of tempo changes within a track.
  • A reader and writer, so you can read and write your MIDI tracks to disk.


Follow the normal procedure for Python module installation:

python install

Examine a MIDI File

To examine the contents of a MIDI file run

$ mary.mid

This will print out a representation of "Mary had a Little Lamb" as executable python code.

Example Usage

Building a MIDI File from scratch

It is easy to build a MIDI track from scratch.

import midi
# Instantiate a MIDI Pattern (contains a list of tracks)
pattern = midi.Pattern()
# Instantiate a MIDI Track (contains a list of MIDI events)
track = midi.Track()
# Append the track to the pattern
# Instantiate a MIDI note on event, append it to the track
on = midi.NoteOnEvent(tick=0, velocity=20, pitch=midi.G_3)
# Instantiate a MIDI note off event, append it to the track
off = midi.NoteOffEvent(tick=100, pitch=midi.G_3)
# Add the end of track event, append it to the track
eot = midi.EndOfTrackEvent(tick=1)
# Print out the pattern
print pattern
# Save the pattern to disk
midi.write_midifile("example.mid", pattern)

A MIDI file is represented as a hierarchical set of objects. At the top is a Pattern, which contains a list of Tracks, and a Track is is a list of MIDI Events.

The MIDI Pattern class inherits from the standard python list, so it supports all list features such as append(), extend(), slicing, and iteration. Patterns also contain global MIDI metadata: the resolution and MIDI Format.

The MIDI Track class also inherits from the standard python list. It does not have any special metadata like Pattern, but it does provide a few helper functions to manipulate all events within a track.

There are 27 different MIDI Events supported. In this example, three different MIDI events are created and added to the MIDI Track:

  1. The NoteOnEvent captures the start of note, like a piano player pushing down on a piano key. The tick is when this event occurred, the pitch is the note value of the key pressed, and the velocity represents how hard the key was pressed.
  2. The NoteOffEvent captures the end of note, just like a piano player removing her finger from a depressed piano key. Once again, the tick is when this event occurred, the pitch is the note that is released, and the velocity has no real world analogy and is usually ignored. NoteOnEvents with a velocity of zero are equivalent to NoteOffEvents.
  3. The EndOfTrackEvent is a special event, and is used to indicate to MIDI sequencing software when the song ends. With creating Patterns with multiple Tracks, you only need one EndOfTrack event for the entire song. Most MIDI software will refuse to load a MIDI file if it does not contain an EndOfTrack event.
You might notice that the EndOfTrackEvent has a tick value of 1. This is because MIDI represents ticks in relative time. The actual tick offset of the MidiTrackEvent is the sum of its tick and all the ticks from previous events. In this example, the EndOfTrackEvent would occur at tick 101 (0 + 100 + 1).

Side Note: What is a MIDI Tick?

The problem with ticks is that they don't give you any information about when they occur without knowing two other pieces of information, the resolution, and the tempo. The code handles these issues for you so all you have to do is think about things in terms of milliseconds, or ticks, if you care about the beat.

A tick represents the lowest level resolution of a MIDI track. Tempo is always analogous with Beats per Minute (BPM) which is the same thing as Quarter notes per Minute (QPM). The Resolution is also known as the Pulses per Quarter note (PPQ). It analogous to Ticks per Beat (TPM).

Tempo is set by two things. First, a saved MIDI file encodes an initial Resolution and Tempo. You use these values to initialize the sequencer timer. The Resolution should be considered static to a track, as well as the sequencer. During MIDI playback, the MIDI file may have encoded sequenced (that is, timed) Tempo change events. These events will modulate the Tempo at the time they specify. The Resolution, however, can not change from its initial value during playback.

Under the hood, MIDI represents Tempo in microseconds. In other words, you convert Tempo to Microseconds per Beat. If the Tempo was 120 BPM, the python code to convert to microseconds looks like this:

>>> 60 * 1000000 / 120

This says the Tempo is 500,000 microseconds per beat. This, in combination with the Resolution, will allow you to convert ticks to time. If there are 500,000 microseconds per beat, and if the Resolution is 1,000 than one tick is how much time?

>>> 500000 / 1000
>>> 500 / 1000000.0

In other words, one tick represents .0005 seconds of time or half a millisecond. Increase the Resolution and this number gets smaller, the inverse as the Resolution gets smaller. Same for Tempo.

Although MIDI encodes Time Signatures, it has no impact on the Tempo. However, here is a quick refresher on Time Signatures:

Reading our Track back from Disk

It's just as easy to load your MIDI file from disk.

import midi
pattern = midi.read_midifile("example.mid")
print pattern


If you use this toolkit under Linux, you can take advantage of ALSA's sequencer. There is a SWIG wrapper and a high level sequencer interface that hides the ALSA details as best it can. This sequencer understands the higher level Event framework, and will convert these Events to structures accessible to ALSA. It tries to do as much as the hard work for you as possible, including adjusting the queue for tempo changes during playback. You can also record MIDI events, and with the right set of calls, the ALSA sequencer will timestamp your MIDI tracks at the moment the event triggers an OS hardware interrupt. The timing is extremely accurate, even though you are using Python to manage it.

I am extremely interested in supporting OS-X and Win32 sequencers as well, but I need platform expert who can help me. Are you that person? Please contact me if you would like to help.

Scripts for Sequencer

To examine the hardware and software MIDI devices attached to your system, run the script.

] client(20) "OP-1 Midi Device"
]   port(0) [r, w, sender, receiver] "OP-1 Midi Device MIDI 1"
] client(129) "__sequencer__"
] client(14) "Midi Through"
]   port(0) [r, w, sender, receiver] "Midi Through Port-0"
] client(0) "System"
]   port(1) [r, sender] "Announce"
]   port(0) [r, w, sender] "Timer"
] client(128) "FLUID Synth (6438)"
]   port(0) [w, receiver] "Synth input port (6438:0)"

In the case shown, qsynth is running (client 128), and a hardware synthesizer is attached via USB (client 20).

To play the example MIDI file, run the script. 128 0 mary.mid

Website, support, bug tracking, development etc.

You can find the latest code on the home page:

You can also check for known issues and submit new ones to the tracker:


I originally wrote this to drive the electro-mechanical instruments of Ensemble Robot, which is a Boston based group of artists, programmers, and engineers. This API, however, has applications beyond controlling this equipment. For more information about Ensemble Robot, please visit: