About

Standard MIDI Files (SMF) store musical performance data in a compact binary format defined by the MIDI Manufacturers Association. The format encodes delta times as variable-length quantities (VLQ), packs channel messages into status bytes with running status optimization, and nests everything inside MThd and MTrk chunks. Misinterpreting a single byte offset corrupts the entire event stream. This tool performs real binary parsing of .mid files into human-readable JSON, preserving every meta event, control change, and timing value. The reverse operation reconstructs valid MIDI binaries from JSON with correct chunk lengths and VLQ encoding. Note: this parser handles SMF Format 0, 1, and 2 files, but Format 2 (independent sequences) is rare and some DAWs may not load it correctly.

Formulas

MIDI files encode timing deltas using Variable-Length Quantity (VLQ) encoding. Each byte uses 7 data bits and 1 continuation bit.

VLQ(value) → bytes where b_i[7] = 1 if more bytes follow, b_i[6..0] = data bits

Tempo conversion from microseconds per quarter note to BPM:

BPM = 60,000,000μs/QN

Chunk structure follows the pattern: 4-byte type identifier (MThd or MTrk) followed by a 4-byte big-endian unsigned integer for chunk data length, then the raw chunk data.

Chunk = type_4B + length_4B + data_lengthB

Where type = chunk identifier, length = size of data in bytes (big-endian 32-bit unsigned), data = chunk payload. Pitch bend value is a 14-bit number assembled from two 7-bit data bytes:

pitchBend = LSB + (MSB × 128)

Reference Data

MIDI Event Type	Status Byte	Data Bytes	Description
Note Off	0x8n	2	Release note on channel n
Note On	0x9n	2	Press note (velocity 0 = note off)
Poly Aftertouch	0xAn	2	Per-note pressure change
Control Change	0xBn	2	Controller number + value (0 - 127)
Program Change	0xCn	1	Instrument/patch selection
Channel Aftertouch	0xDn	1	Channel-wide pressure
Pitch Bend	0xEn	2	14-bit value, center = 8192
SysEx Start	0xF0	Variable	System Exclusive message begin
SysEx End	0xF7	Variable	System Exclusive continuation/end
Meta: Sequence Number	0xFF 0x00	2	Track sequence identifier
Meta: Text Event	0xFF 0x01	Variable	Arbitrary text annotation
Meta: Copyright	0xFF 0x02	Variable	Copyright notice string
Meta: Track Name	0xFF 0x03	Variable	Name of the track
Meta: Instrument	0xFF 0x04	Variable	Instrument name for track
Meta: Lyric	0xFF 0x05	Variable	Lyric text synchronized to events
Meta: Marker	0xFF 0x06	Variable	Rehearsal marker or section label
Meta: Cue Point	0xFF 0x07	Variable	Cue for external synchronization
Meta: Channel Prefix	0xFF 0x20	1	Associate channel with following meta events
Meta: End of Track	0xFF 0x2F	0	Mandatory track terminator
Meta: Set Tempo	0xFF 0x51	3	Microseconds per quarter note
Meta: SMPTE Offset	0xFF 0x54	5	Starting SMPTE time code
Meta: Time Signature	0xFF 0x58	4	Numerator, denominator power, clocks, 32nds
Meta: Key Signature	0xFF 0x59	2	Sharps/flats count + major/minor flag
Meta: Sequencer Specific	0xFF 0x7F	Variable	Proprietary sequencer data

Frequently Asked Questions

Format 0 stores all channels in a single track, making it the simplest but hardest to edit per-instrument. Format 1 uses multiple simultaneous tracks (one per channel typically), which is the most common format exported by DAWs. Format 2 contains independent single-track patterns (rarely used). This converter handles all three formats. The format type is stored in the first two bytes of the MThd chunk data.

VLQ encodes an unsigned integer using 7 bits per byte. The most significant bit (bit 7) is a continuation flag: 1 means more bytes follow, 0 means this is the final byte. A delta time of 127 fits in one byte (0x7F). A value of 128 requires two bytes (0x81 0x00). The maximum VLQ value in MIDI is 0x0FFFFFFF (4 bytes). This encoding is critical to preserve - incorrect VLQ reconstruction produces corrupted MIDI files that DAWs will reject.

SysEx messages (status 0xF0) are preserved as byte arrays in the JSON output. The converter stores the raw payload bytes (excluding the 0xF0 start and 0xF7 end markers) as an array of integers. When converting back to MIDI, these bytes are re-wrapped with the correct markers and VLQ length prefix. Multi-packet SysEx (using 0xF7 continuation) is also handled.

The converter preserves all event data faithfully, but running status optimization may differ. The original file might use running status (omitting repeated status bytes) while the reconstructed version writes explicit status bytes for every event, or vice versa. Both are valid per the SMF specification. If timing sounds different, check that the ticksPerBeat value in the header was preserved correctly. Some DAWs also interpret pitch bend center differently (8192 vs 0).

Yes. The header's division field is parsed as a signed 16-bit value. If bit 15 is set (1), the upper byte represents negative SMPTE frames per second (−24, −25, −29, or −30) and the lower byte is ticks per frame. This is stored in JSON as a timeDivision object with framesPerSecond and ticksPerFrame properties instead of a simple ticksPerBeat integer.

The converter processes files up to 50 MB in the browser. Files larger than 100 KB are parsed inside a Web Worker to prevent UI freezing. Typical MIDI files range from 5 to 500 KB. Files exceeding 50 MB likely contain embedded audio (not standard MIDI) and should be checked for corruption.