Bebop demo
soloscribe transcription audit · clip.wav · compared against the original · 01 Aug 2026, 04:02
MEDIUM CONFIDENCE
The shape of the solo is right; the details need a human. Work through the weakest bars listed below.
- 100% note agreement at ±100 ms and 88% of audible attacks accounted for.
- The resynthesis had to be stretched by up to 0.88 s to line up with the recording, so the tab slips out of time somewhere.
Listen
Play these against each other. Your ears will find the errors faster than the table below will, and the table exists mainly to tell you which bar to rewind to.
Originalwhat you gave us
Transcription, resynthesizedwhat the transcriber heard, voiced with a sampled guitar at the times it was played — not the quantized tab, and not what the numbers below measure
What was heard, what was written
Metrics
| Measurement | Value | What it means |
|---|---|---|
| Note agreement, ±100 ms | 100.0% | Of everything the transcriber heard, how much still lines up after being written onto a musical grid. This is the headline number: how much of the playing survived being written down. |
| Note agreement, ±50 ms | 98.2% | The same test with a stricter stopwatch. A big drop from the ±100 ms figure means the notes are right but the rhythm is written a little loose. |
| Precision, ±50 ms | 98.2% | Of the notes written in the tab, how many correspond to something actually played. Low means the tab has notes the recording doesn't. |
| Recall, ±50 ms | 98.2% | Of the notes actually played, how many made it into the tab. Low means notes are missing. |
| Harmonic alignment cost | 0.091 | How different the resynthesis sounds from the recording, pitch-class by pitch-class. 0 is identical. Above ~0.45 usually means wrong notes rather than wrong rhythm. |
| Worst timing drift | 0.88 s | The furthest the resynthesis had to be shifted to line up with the recording. Small means the tab follows the take; a large number means it slips out of time somewhere. |
| Typical timing drift | 0.05 s | The same measurement taken across the whole solo rather than at its worst point. If this is small and the figure above is large, the problem is one spot, not the whole transcription. |
| Onset coverage | 88.3% | Share of the pick attacks audible in the recording that have a note within 80 ms in the tab. Low means the tab is missing notes you can hear. |
| Onset spuriousness | 5.4% | Share of notes in the tab with no audible attack anywhere near them. High means the tab is inventing notes. (Some of this is normal: tied and legato notes have no fresh attack to detect.) |
Bar by bar
Timing error is how far the played notes had to move to land on the grid, averaged over the bar. A bar with high error and low confidence is where to start editing.
| Bar | Notes | Mean confidence | Timing error (RMS) | |
|---|---|---|---|---|
| 1 | 8 | 0.91 | 33.4 ms | |
| 2 | 9 | 0.90 | 31.0 ms | |
| 3 | 3 | 0.91 | 32.2 ms | |
| 4 | 9 | 0.90 | 33.3 ms | |
| 5 | 8 | 0.91 | 34.3 ms | |
| 6 | 9 | 0.90 | 33.6 ms | |
| 7 | 5 | 0.89 | 35.6 ms | |
| 8 | 5 | 0.90 | 35.4 ms | |
| 9 | 0 | — | — | no notes written |
Caveats
- transcribed in single-note mode — if this solo uses octave doubling (Wes Montgomery-style), the doubling will have been written as a single line; re-run with 'Chords too' to keep it
- everything pitched in the clip gets written down — if another instrument shares it, its notes are in the tab too, and the checks below cannot tell whose notes are whose
- No isolated guitar track was available, so the audit compares against the full mix. Bass, drums and keys all count against the harmonic and onset figures, which makes those two numbers pessimistic — usually by a lot.
- The resynthesis you can play above is voiced with a sampled guitar and plays the notes at the times the transcriber heard them, before they were written onto the grid. Every number in this report is measured from a plucked-string render of the quantized score instead, so that player cannot tell you what quantization changed — the piano roll and the bar-by-bar table can.
- Articulation is approximate. Slides, hammer-ons, pull-offs and legato phrasing are written as separately picked notes; the tab will say the right pitches but not always the right left hand.
- Bends and vibrato are written into the .gp5 but are NOT reproduced in the resynthesis used for this audit. A heavily bent phrase therefore scores worse here than the transcription actually deserves.
- Swing is notated straight with a feel marker, which is the convention. Swung eighths will look like even eighths on the page and should be played with the marked feel, not as written.
- This is a first pass, not a finished transcription. Open the .gp5 in Guitar Pro or TuxGuitar and fix it by ear — this report is a map of where to look, not a verdict on what to write.