Guitar Tab View — How it Works

The short version

A MIDI file describes which notes are played at which times, and how loud. A guitar tab describes where on the fretboard those notes should be played, with which finger, and how (palm-muted, bent, slid, etc.). The gap between the two is big, and a lot of it cannot be recovered from the MIDI alone.

We bridge the gap with heuristics: pick reasonable fret positions, detect palm-mutes from short note durations, infer hammer-ons from adjacent same-string notes, etc. The result is readable and playable, but it will rarely match a tab a human transcriber would write for the same song.

What survives the MIDI round-trip

These things are encoded in standard MIDI and we use them as-is:

• Pitches and timing. The actual notes of the song.
• Tempo changes. Including gradual ramps in songs that change BPM.
• Time-signature changes. 4/4 to 7/8 to 21/32 alternations all render with the correct stacked signatures.
• Section markers. "Intro", "Verse 1", "Chorus", "Solo A" labels above the bars come from MIDI marker meta-events.
• Track names and instruments. Shown in the sidebar.
• Velocity. Used to render dynamics letters (f, mf, etc.) and to infer accents.
• Pitch bends. When the source MIDI contains pitch-bend events, the tab renders bend curves on the affected notes.

What we infer (best effort)

None of these are recorded directly in MIDI, so we guess from the note pattern. The guesses are pretty good for typical rock / metal guitar but they will sometimes miss or false-positive:

• Fret & string choice. A given MIDI pitch can be played in several places on the neck. We pick a comfortable hand position per phrase using a "Viterbi" search that minimises hand movement.
• Palm-mute brackets ("P. M."). Inferred from short note duration + low pitch + mid velocity + tight rhythmic context. Works well on chugging riffs; misses single-note palm-mutes.
• Hammer-on / pull-off. Same string, adjacent attacks, very tight gap, both notes fretted. alphaTab draws the slur.
• Slides. Same string, adjacent attacks, sustained first note, fret delta of 1–5.
• Dead notes ("x"). Very short duration + low velocity.
• Accents (">"). Velocity significantly above the local median.
• Grace notes. Very short notes immediately preceding longer ones.
• Ties. A sustained note that crosses a beat boundary becomes two beats joined by a tie curve.

What MIDI cannot give us

These are the gaps. The information was created by the transcriber/composer, not by the music itself, so it's not recoverable from a MIDI file no matter how clever the inference:

• The specific fingering a real guitarist would use. Two players will choose different strings/frets for the same notes based on the surrounding phrase, the song style, the tuning, the fingering they've practised. We pick one reasonable option; a human transcriber picks the one that matches the actual recording.
• Text annotations. "let ring", "neck pickup for the solo", "rake", "12th-fret harmonic" — these are notes the transcriber wrote for the reader, not MIDI data.
• Bend curves with intent. Even when MIDI carries pitch-bend events, the original transcriber's label ("full", "½", "release") is gone.
• Whammy-bar dives. Some are encoded as pitch bends but the visual labels aren't.
• Vibrato wavy lines. MIDI vibrato (if any) is a continuous modulation, not a marker.
• Harmonic noteheads. No MIDI signal — natural / artificial / pinch harmonics all sound similar enough that no encoder writes them differently.
• Pickstroke directions (∨ / ∧). Not MIDI data.
• Triplet "3" brackets. Triplet feel is a notational choice — when the source MIDI represents triplets as straight ticks at 1/24-of-bar positions, we can sometimes detect them, but the detection is unreliable, so it's currently disabled to avoid putting "3" brackets on bars that aren't actually triplets. The notes still appear; they just lose the "this is a triplet group" annotation.
• Lyrics. Not in our MIDI.

Why is this view still useful?

Even with the gaps, a generated tab is faster to read than a piano roll for guitarists, and good enough for:

• Sight-reading and practice along with a known recording.
• Picking up the riff structure of an unfamiliar song quickly.
• Learning rhythm-guitar parts where exact fingering is less critical than the rhythmic pattern (the tempo, time signatures, ties, palm-mutes, and section markers usually come through correctly).
• Comparing across multiple instruments — bass, lead, rhythm, drums — that came from the same MIDI.

For pixel-perfect lead-guitar transcriptions you should still go to a hand-authored tab or Guitar Pro file.

Knobs and trade-offs

A few quick notes on the heuristics' design choices:

• Fret position is chosen per phrase to fit the lowest comfortable window that covers most of the pitches. Outlier high pitches escape onto the thinner strings. A different choice (e.g. "always play this lick at the 12th fret") would also be musically valid; we have no way to know which one matches the recording.
• Palm-mute inference requires both short note duration AND tight rhythmic context, to avoid putting "P. M." over every short note in a sparse passage.
• Triplet bracket rendering is disabled (see above) — it's not worth the false-positive cost on songs that aren't actually in a triplet feel.
• Tab clef ("T A B") is hidden because it adds visual noise without telling you anything new.

If something looks wrong

File a bug with:

• The MIDI filename.
• The track and bar number (visible at the top-left of each bar).
• A short description of what the tab shows and what you'd expect instead.

Fingering inaccuracies are usually inherent limitations; rhythmic or pitch errors usually point to a real bug.