# MIDI track splitter Split a multi-track MIDI file into separate single-track .mid files, one per track or per channel. Tempo and time signature are copied into every part. ## Run it - **CLI:** `gizza tool midi-track-splitter "4d546864000000060001000300604d54726b0000002000ff0309436f6e647563746f7200ff510307a12000ff58040402180800ff2f004d54726b0000002000ff03055069616e6f00c00000903c4060903c00009040466090400000ff2f004d54726b0000001800ff03044261737300c12100912464814091240000ff2f00"` - **Web:** https://gizza.ai/tools/midi-track-splitter/ - **Agents:** machine-readable descriptor (parameters JSON Schema) at https://gizza.ai/tools/midi-track-splitter/tool.json ## Inputs - `input` — MIDI file (base64 or hex) _(field)_ - `encoding` — Input encoding _(field)_ - `split_by` — One file per… _(field)_ - `include_conductor` — Copy tempo, time and key signature into every part _(field)_ - `output_format` — Write each part as _(field)_ - `skip_empty` — Skip parts with no notes (drops the tempo-only conductor track) _(field)_ - `select` — Only these parts (track or channel numbers, e.g. 1,3-5) _(field)_ - `filename_prefix` — Filename prefix _(field)_ - `output` — Return _(field)_ ## Output - Split parts (text) ## Query parameters Open the tool pre-filled and auto-run via URL: - `input` — MIDI file (base64 or hex) - `encoding` — Input encoding - `split_by` — One file per… - `include_conductor` — Copy tempo, time and key signature into every part - `output_format` — Write each part as - `skip_empty` — Skip parts with no notes (drops the tempo-only conductor track) - `select` — Only these parts (track or channel numbers, e.g. 1,3-5) - `filename_prefix` — Filename prefix - `output` — Return Example: `https://gizza.ai/tools/midi-track-splitter/?input=4d546864000000060001000300604d54726b0000002000ff0309436f6e647563746f7200ff510307a12000ff58040402180800ff2f004d54726b0000002000ff03055069616e6f00c00000903c4060903c00009040466090400000ff2f004d54726b0000001800ff03044261737300c12100912464814091240000ff2f00&encoding=auto&split_by=track&include_conductor=true&output_format=format-0&skip_empty=true&select=leave%20empty%20for%20every%20part&filename_prefix=part&output=files` --- ## About this tool A multi-track MIDI file is a container: one header, then a stack of **track chunks**, one per part, plus a conductor track at the front holding the tempo, time signature and key signature for the whole piece. This tool takes that container apart and writes each part back out as a complete `.mid` file of its own — so you can import just the bass line into a project, hand one player their part, or feed a single melody to a tool that only accepts one track at a time. Paste the file as **base64 or hex** and choose how it should be cut: - **One file per track** — the usual cut for a Format 1 file exported by a DAW or notation program, where each instrument already has its own track chunk. - **One file per MIDI channel** — every event on channel 3 is gathered from *every* track into one part. This is the right cut for files that put several instruments in one track, and it is what a Format 0 file (which has only a single track) is always split by. The thing naive splitters get wrong is the conductor data. Tempo and time signature normally live in track 0 only, so a part pulled out on its own plays back at the MIDI default — 120 BPM in 4/4 — no matter what the original said. **Copy tempo, time and key signature into every part** is on by default and fixes exactly that; events a part already carried are never duplicated. Turn it off if you are re-importing the parts into a project that supplies its own tempo map. Each part is written as **Format 0** (one merged track — a genuinely single-track file) or **Format 1** (the conductor data kept as its own first track, with the part second). Either way, every note-on, note-off, velocity, controller, pitch bend and program change is carried through at its original tick, and the source division — ticks per quarter note, or SMPTE timecode — is preserved, so nothing is re-gridded or quantised. Filenames are built from what is actually in the file: the track name meta event when there is one, otherwise the General MIDI instrument implied by the part's program change (channel 10 becomes the drum kit), so an import shows `part-02-bass.mid` rather than `track2.mid`. ### Worked example Splitting a small three-track file — a conductor track at 120 BPM in 4/4, a **Piano** part on channel 1 with two notes, and a **Bass** part on channel 2 with one — with every default left alone gives two files: | # | Part | From | Instrument | Notes | Length | File | |---|------|------|------------|-------|--------|------| | 2 | Piano | track 2 · ch 1 | Acoustic Grand Piano | 2 | 1.00 s | `part-02-piano.mid` | | 3 | Bass | track 3 · ch 2 | Electric Bass (finger) | 1 | 1.00 s | `part-03-bass.mid` | The conductor track itself is not exported, because it holds no notes — its tempo and 4/4 time signature were copied into both parts instead, so each one opens at 120 BPM on its own. Every "Try:" button above runs a real example; switch **Return** to *Just the list of parts* to see the same table without producing any files, which is the quick way to look inside a file first. ### Limits and edge cases - The decoded file may be up to **4 MiB**, and one split may produce at most **64 files**. Use the parts selector (`1,3-5`) to work through a bigger file in batches. - **Format 0** input has a single track, so a track split would just hand back the input; it is cut by channel automatically and the summary says so. - **Format 2** files hold independent sequences rather than parallel parts, so each track is exported as-is and no conductor data is shared between them. - **SMPTE timecode** files are supported: the division is copied unchanged and part lengths are computed from the frame rate rather than a tempo map. - Parts with no notes — the conductor track, muted placeholders — are skipped unless you untick **Skip parts with no notes**. - Nothing is uploaded. The whole split runs as WebAssembly inside this tab, so unreleased material never leaves your machine. ## FAQ
Will the split parts still play at the right tempo? Yes, as long as **Copy tempo, time and key signature into every part** is left on (it is by default). Tempo, time signature, key signature and SMPTE offset are collected from the conductor track — and from anywhere else in the file that carries them — and written into each part, so a part opened on its own plays at the original tempo instead of the MIDI default of 120 BPM. If a part already contained one of those events, it is kept and not duplicated.
Should I split by track or by channel? Split by **track** when your file came out of a DAW or notation program with one instrument per track — that is the normal Format 1 layout, and it preserves the names you already gave each part. Split by **channel** when several instruments share a track (common in older files, game music and General MIDI karaoke files), or whenever you specifically want one file per MIDI channel. A Format 0 file only has one track, so it is always cut by channel.
What is the difference between Format 0 and Format 1 output? **Format 0** merges the conductor data and the part into a single track — one genuinely single-track file, which is what most importers and simple players expect. **Format 1** keeps the conductor data as its own first track with the part second, which is closer to how DAWs organise a project and makes the tempo map easy to see and edit separately. The notes are identical either way.
Why paste base64 or hex instead of uploading the .mid file? This page runs the whole split as WebAssembly in your browser tab and takes its input as text, which keeps the tool a pure, link-shareable function — every field can be pre-filled from the URL. To get the text, run `base64 -w0 song.mid` (or `xxd -p song.mid | tr -d '\n'` for hex) and paste the result. Encoding is auto-detected, so you rarely need to touch the encoding selector.
Does splitting change my notes, timing or velocities? No. Note positions are stored in **ticks** — musical units relative to the file's division — and every event is copied to its original tick in the output. The division itself is preserved too, so nothing is quantised, re-gridded or rounded. Swing, humanised timing, velocity dynamics, controller curves and pitch bends all come through exactly as they went in.
How are the output files named? Each name is `--.mid`, for example `part-02-bass.mid`. The number is the source track number (or the MIDI channel, when splitting by channel), and the name comes from the track name meta event, falling back to the General MIDI instrument of the part's program change, or to the channel number. Channel 10 is named as the drum kit, per the General MIDI convention. Change the prefix to anything you like; duplicates get a numeric suffix so no two files collide.
Can I export only some of the parts? Yes — put the ones you want in **Only these parts** as numbers and ranges, such as `1,3-5`. The numbers are track numbers when splitting by track (track 1 is the first chunk in the file) and MIDI channel numbers 1-16 when splitting by channel. Leave the field empty to export everything. Setting **Return** to *Just the list of parts* first is the easiest way to see which numbers you want.
## Related tools - [Convert AIFF to FLAC](https://gizza.ai/tools/aiff-to-flac/): Convert AIFF or AIF audio to lossless FLAC in your browser — preserve samples and textual tags, choose FLAC compression level 0-12, and download the result. - [Bleep or Mute Regions of an Audio File](https://gizza.ai/tools/audio-bleep-censor/): Censor an audio file in your browser — bleep, mute, or duck the exact time regions you pick. Set a tone frequency and output format. Nothing is uploaded, free. - [Audio Channel Tool](https://gizza.ai/tools/audio-channel/): Swap left/right channels, downmix stereo to mono, up-mix mono to stereo, or copy one side onto both — right in your browser. Nothing is uploaded, free. - [Compress an Audio File](https://gizza.ai/tools/audio-compress/): Shrink MP3, WAV, M4A or any audio file right in your browser — pick a target bitrate (96 kbps default) and format. Nothing is uploaded, free. - [Compress Audio Dynamic Range](https://gizza.ai/tools/audio-compressor/): Even out loud and quiet parts of an audio file in your browser with threshold, ratio, attack, release and make-up gain. Local and private.