The MIDI File Format

MIDI File Structure

Chunks

MIDI files are structured into chunks.

Each chunk consists of:

• A 4-byte chunk type (ASCII)
• A 4-byte length (32 bits, msb first)
• length bytes of data

There are two types of chunks:

Header Chunks

which have a chunk type of "MThd"

Track Chunks

which have a chunk type of "MTrk"

A MIDI file consists of a single header chunk followed by one or more track chunks.

Since the length-field is mandatory in the structure of chunks, it is possible to accomodate chunks other than "MThd" or "MTrk" in a MIDI file, by skipping over their contents. The MIDI specification requires that software be able to handle unexpected chunk-types by ignoring the entire chunk.

Value Representations

Within a MIDI file, there is a variety of information in addition to the basic MIDI data, such as delta-times and meta-events.

Where such information contains numeric values, these are generally represented in one of two formats:

• binary
• variable length quantity

Binary

Binary values are stored:

• 8-bits per byte
• MSB first (Most Significant Byte first)

(unless otherwise specified)

Variable Length Quantities

The variable-length quantity provides a convenient means of representing arbitrarily large integers, without creating needlessly large fixed-width integers.

A variable-length quantity is a represented as a series of 7-bit values, from most-significant to least-significant. where the last byte of the series bit 7 (the most significant bit) set to 0, and the preceding bytes have bit 7 set to 1.

Examples:

As you can see from the above examples, small values (0-127) can be represented by a single byte, while larger values are also accomodated.

The largest value allowed within a MIDI file is 0FFFFFFF. This limit is set to allow variable-length quantities to be manipulated as 32-bit integers.

Header Chunks

The data part of a header chunk contains three 16-bit fields. These fields specify the format, number of tracks, and timing for the MIDI file.

The length of the header chunk is 6-bytes. However, software which reads MIDI files is required to honor the length field, even if it is greater than expected. Any unexpected data must be ignored.

<length>

length in bytes of the chunk data part.
This is a 32-bit binary number, MSB first.
This will be exactly 6 (bytes) for any MIDI file created under the MIDI 1.0 specification. Nevertheless, any MIDI file reader should be able to cope with larger header-chunks, to allow for future expansion.

 

<format>

The MIDI file format.
This is a 16-bit binary number, MSB first.
The only valid formats are 0, 1 and 2.

 

<tracks>

The number of track chunks contained in this MIDI file.
This is a 16-bit binary number, MSB first.

 

<division>

This defines the default unit of delta-time for this MIDI file.
This is a 16-bit binary value, MSB first.

This may be in either of two formats, depending on the value of MS bit:

 

Bit:	15	14 ... 8	7 ... 0
<division>	0	ticks per quarter note
	1	-frames/second	ticks / frame

 

bit 15 = 0:

bits 0-14

number of delta-time units in each a quarter-note.

 

bit 15 = 1:

bits 0-7

number of delta-time units per SMTPE frame

bits 8-14

form a negative number, representing the number of SMTPE frames per second. Valid values correspond to those in the MTC Quarter Frame message.

 

   -24 = 24 frames per second
   -25 = 25 frames per second
   -29 = 30 frames per second, drop frame
   -30 = 30 frames per second, non-drop frame

MIDI File Formats

MIDI files come in 3 variations:

• Format 0
  ...which contain a single track

   

• Format 1
  ... which contain one or more simultaneous tracks
  (ie all tracks are to be played simultaneously).

   

• Format 2
  ...which contain one or more independent tracks
  (ie each track is to be played independently of the others).

   

Format 0 MIDI Files

Format 0 MIDI files consist of a header-chunk and a single track-chunk.

The single track chunk will contain all the note and tempo information.

Format 1 MIDI Files

Format 1 MIDI files consist of a header-chunk and one or more track-chunks, with all tracks being played simultaneously.

The first track of a Format 1 file is special, and is also known as the 'Tempo Map'. It should contain all meta-events of the types Time Signature, and Set Tempo. The meta-events Sequence/Track Name, Sequence Number, Marker, and SMTPE Offset. should also be on the first track of a Format 1 file.

Format 2 MIDI Files

Format 2 MIDI files consist of a header-chunk and one or more track-chunks, where each track represents an independant sequence.

Track Chunks

The data part of a track chunk contains one or more <delta_time> <event> pairs. The <delta_time> is not optional, but zero is a valid delta-time.

<delta_time>

is the number of 'ticks' from the previous event, and is represented as a variable length quantity

 

<event>

is one of:

• <midi_event>
• <sysex_event>
• <meta_event>

Events

Notice that there are no explicit delimiters between the <delta_time> and <event> instances.
This is possible because both the delta-times and events have clearly defined lengths:

• The last byte of a delta-time is identified by having MSbit=0;
• MIDI Channel messages have a pre-defined length (even when running status is used);
• and sysex-events and meta-events contain an explicit length field.

MIDI Events

A <midi_event> is any MIDI Channel message. This includes:

• Channel Voice messages
• Channel Mode messages

Running status is applicable within MIDI files in the same manner as usual. Running status is cancelled by any <sysex_event> or <meta_event>

Sysex Events

Where is is desirable to include messages other than MIDI Channel messages in a MIDI file (System Exclusive messages in particular), the <sysex_event> can be used.

Sysex events come in two flavors:

In both cases, the End-Of-Exclusive message, F7, is not sent automatically. It must be specified explicitly within <sysex_data> or <any_data>.

Most System Exclusive messages are quite simple, and are sent as a single packet of bytes, starting with F0 and ending with F7. These are easily accomodated using the F0 form of the <sysex_event>

However, some System Exclusive messages are used to control device parameters in real-time, and what is syntactically a single System Exclusive message may consist of a series of small parts which must be sent with appropriate delays. Hence it is necessary to be able to break up a single message into a number of events, with appropriate delta-times. This can be accomodated by using the F7 form of the <sysex_event>.

In fact, the F7 Sysex Event can be used to included any data into the MIDI stream, such as MIDI System Real-Time messages (though this is generally not desirable).

Example 1:

To generate the message:
F0 7E 09 03 01 01 F7
(Sample Dump Request - Device 9, sample number 257)
the <sysex_event> would be:
F0 06 7E 09 03 01 01 F7
or alternately:
F7 07 F0 7E 09 03 01 01 F7

Example 2:

The MIDI System Real-Time message 'stop' can be inserted using
the <event> F7 01 FC, while 'continue'. would be F7 01 FB.

Suppose you wanted to pause an external drum machine by sending a 'stop', followed 48 delta-time units later by 'continue'. The complete delta-time/event sequence would look something like this:

  00 F7 01 FC 30 F7 01 FB

Meta Events

Meta Events are used for things like track-names, lyrics and cue-points, which don't result in MIDI messages being sent, but are still useful components of a MIDI file.

Meta Events have the general form:

FF <type> <length> <data>

where:

 

<type>

is a single byte, specifying the type of meta-event.
The possible range is 00-7F. Not all values in this range are defined, but programs must be able to cope with (ie ignore) unexpected values by examining the length and skipping over the data portion.

 

<length>

is the number of bytes of <data> following.
This is a variable length quantity
0 is a valid <length>

 

<data>

zero or more bytes of data

Footnotes

Note 1

In the case where the value of <division> (in the header chunk) defines delta-time units in 'ticks per quarter note' (MSbit=0), a change in tempo means a change in the length of a unit of delta-time.

In the case where <division> MSbit=1, and the 'ticks' are defined in absolute terms (ticks/frame and frames/second), it is not clear from the specification what effect a new tempo should have.