Most guitarists can only describe their timing as a feeling. You can put a number on it, but only if you record carefully enough. A late note should be your hand's fault, not the fault of the audio buffer, the click or a zoom level that makes you look better than you are.
Six matching takes make the test trustworthy
A part can look glued to the grid at one zoom level and still feel unsteady when you listen back. To measure timing in a DAW (your recording software), you need a fixed task, a known target for every note and a way to separate the delay of your audio interface from the delay of your hands.
This method works in any DAW. You make six recordings that can be compared: three at 80 BPM and three at 120 BPM. It checks timing, not groove. A chorus played deliberately behind the beat can be musically right and still measure late. The test answers one question: how consistently does each picked note land against a fixed target? A performance rubric marked by ear can only estimate that.
Play eighth notes on one fretted note; if alternate picking is still new, start with alternate picking from scratch. Strummed chords give several attacks at once, and scales add position shifts and string changes. Alternate-pick the fifth fret on the D string with a clean or edge-of-breakup sound, loud enough to see the pick attack without clipping. The note does not matter here. Doing the same thing every time is what matters.
Session settings for six comparable takes
Create a blank project:
- Set 48 kHz and 24-bit. Any stable rate works, but do not change it between takes.
- Use 4/4, a fixed tempo map, and a quarter-note click with an accent on beat one only.
- Make one mono guitar track with no quantisation, stretching, transient replacement, or post-recording editing.
- Record two count-in bars, then eight bars of continuous eighth notes: 64 target onsets per take.
- Leave two silent bars between takes. Record three takes at 80 BPM, change only tempo to 120 BPM, then record three more.
- Keep the same pick, pickup, gain, monitoring route, buffer, and direct-monitoring choice for all six.
Name the files 80_T1, 80_T2, 80_T3, 120_T1, 120_T2, and 120_T3. This naming prevents comparing the best slow take with the worst fast one and calling it a tempo trend.
At 80 BPM, eighth-note targets are 375 ms apart; at 120 BPM they are 250 ms apart. Play one clear attack on every eighth-note grid line. Do not add accents, swing, palm muting, or expressive pushes in the measurement pass, because each of those choices moves the target.
Check your recording setup before you judge your playing
A steady delay often comes from the recording setup, even when it looks like late playing. Buffers, drivers, hidden plug-in delay and the DAW's own correction can all shift the recorded waveform. A loopback test finds that delay before you blame your technique.
Connect a spare line output to a spare line input with a line-level cable. Do not send a headphone output into a microphone input, do not switch on phantom power, and do not listen to the returning signal through the same output: feedback helps nobody.
Make a very short click, such as a one-sample spike. Send it to the spare output and record what comes back on a new mono track, starting at bar 3 so there is space before it. Record three passes without changing the sample rate, buffer, driver or plug-ins. Zoom in, count the samples between the original click and the first sign of the returned one, and convert that to milliseconds.
Use the middle value of the three. If the returns are 8.2, 8.3 and 8.1 ms late, use 8.2 ms rather than trusting one pass. If the results keep changing, something in the routing, driver or background programs needs fixing before the guitar test means anything.
Do not simply subtract the whole loopback figure from your guitar notes. Loopback measures the way out and the way back in, while the guitar only uses the way in. DAWs also handle manual offset settings differently. Check your DAW's manual, set a recording offset only if it means the same thing as your measurement, then run loopback again. The aim is for the returned click to line up with the original.
If your DAW has no clear setting for this, keep the loopback figure in your notes as a separate baseline. You can still compare take 1 with take 3, or 80 with 120 BPM, because the same delay is in all of them. What you cannot honestly do is call results under 10 ms pure hand timing until you know how the input is placed.
Decide where a note starts before you look at waveforms
The first wiggle in the waveform is not always the start of the note. Pick scrape, string noise, a compressor, a note that fades in and distortion can all move what looks like the start. So choose one rule for marking notes and stick to it.
For a picked note, mark the first lasting movement of the waveform after the pick attack, and ignore any tiny spike before it. Use this rule for all six takes. With a clean direct signal it is usually obvious. With heavy distortion, put the timing markers on the clean direct signal even if you play through the amp sound. Distortion smears the attack and can make notes played at the same time look different.
Three practical onset methods are worth knowing:
| Method | Best use | Failure mode | Sensible use |
|---|---|---|---|
| Manual markers | Small sets and ambiguous attacks | Slow; placement can drift | Mark all notes in one take before moving on |
| DAW transient detection | Clean, percussive guitar | May trigger on pick noise or miss soft notes | Generate markers, then audit visually |
| MIDI or audio-to-MIDI conversion | Rough scan of recurring drift | Detection makes its own timing decisions | Find suspect areas, not final evidence |
The quickest way is automatic detection, then correcting by hand. Accuracy comes from using one rule everywhere. More markers add nothing if the rule keeps changing. Check at least ten random notes per take even when detection looks reliable. If it misses every soft upstroke, you may think you have a picking problem that is not there.
Measure in samples or milliseconds, but keep the musical grid visible. The grid is the target. The click is only a sound to play along with, and its waveform may start slightly before the grid line.
Turn 64 notes into useful numbers
For each note, work out:
error in ms = recorded onset time minus target grid time minus calibrated recording offset
Negative means early, positive means late. Keep it that way throughout. A plain average can hide notes that are alternately 20 ms early and 20 ms late, so for each take note three figures:
- Median signed error shows where the take sits: whether you tend to play ahead of the grid or behind it.
- Median absolute error ignores direction and shows how far a typical note is from the target.
- Spread is the gap between the 90th and 10th percentile (P90 minus P10). In plain terms, it is how wide the middle 80% of your notes are spread, so one accidental missed note does not distort the picture.
The six-take report below is illustrative: each row summarises 64 eighth-note onsets measured with the same calibration rule. It shows how to read the data, not what counts as good enough.
| Tempo | Take | Median signed error | Median absolute error | Spread, P90 minus P10 | Reading |
|---|---|---|---|---|---|
| 80 BPM | T1 | -7 ms early | 12 ms | 37 ms | Slight lead, compact placement |
| 80 BPM | T2 | -3 ms early | 10 ms | 31 ms | Most even slow take |
| 80 BPM | T3 | +2 ms late | 11 ms | 34 ms | Centred, similar control |
| 120 BPM | T1 | +9 ms late | 18 ms | 55 ms | Faster tempo adds late drift |
| 120 BPM | T2 | +13 ms late | 22 ms | 64 ms | Largest timing scatter |
| 120 BPM | T3 | +7 ms late | 17 ms | 52 ms | Better recovery, same direction |
The 80 BPM takes sit close to zero, with narrow spreads. All three 120 BPM takes move late, and the typical error and the spread grow together. That points to a repeatable problem with speed. One random bad take would not look like this. If only one 120 BPM take had a 64 ms spread, check for a missed stroke, an interruption or a detection error before deciding anything about your time feel.
Do not treat median absolute error as a score of musical worth. It only helps when you compare like with like: the same guitarist, task, setup, marking rule and change of tempo.
The shape of the errors tells you more than their size
A decent median can still hide a warning sign. Plot the errors in the order you played the notes, or just look along the markers from left to right. The shape often shows the cause faster than any average.
If the notes drift steadily later over eight bars, think fatigue, a hand that tightens up or attention fading after the count-in. If notes go early, late, early, late, your downstrokes and upstrokes may be uneven. If the first half is clean and things scatter after bar five, the cause may be muting: weak notes give the markers less to work with.
Compare positions in the beat. If downbeats are close but every offbeat is late, you may hear the quarter notes clearly while your inner eighth-note count lags. That is a different problem from every note being 9 ms late. Once you know the cause, the fix is musical work on rhythmic phrasing and subdivision. Grid numbers alone will not solve it.
Do not look only at the notes that look bad. Mark all 64, or decide in advance which ones you will check: beat 1, the "and" of 2, beat 3 and the "and" of 4 in every bar. Picking only the ugly notes turns measurement into a mood report.
A histogram of the errors helps too. One tight hump at +8 ms means you play steadily late. Two humps, one early and one late, mean you are inconsistent even if the average is zero. A long tail to the right shows occasional late notes that the median barely reflects. Each view answers a different question, which is why one "average timing" number is too blunt.
Waveforms measure attacks; feel is a separate judgement
Milliseconds look more exact than they are when targets and note starts are fuzzy. A guitar attack has width, and listeners may hear it later than the first movement of the waveform. An amp simulator, a noise gate or a room microphone widens it further. Moving a marker a few samples at high zoom looks precise, but the musical judgement is not that precise.
Clicks differ too. A sharp digital tick, a cowbell and a soft shaker each seem to land at a slightly different moment. Keep one click sound for the whole test. If you change it, treat it as a new test and keep the results apart.
Feel is a different matter. A guitarist can sit slightly behind the beat in a chorus, then push a fill ahead, and make a track breathe. The waveform records those differences without knowing whether they suit the song. Use this dry one-note task to check consistency, and use the song, the drums, the bass and your ears to decide whether a part feels right.
The same test works for strummed parts: the six strumming mistakes tell you what to listen for. The test cannot find every cause. Your timing may slip only against a busy backing track while you stay tight with the click alone. A very large buffer may cause a monitoring problem that looks like a picking problem. Write down your monitoring setup with the results, so they still make sense six weeks later.
Keep simple notes for every test
After the first clean run, save the session as a template. Add project notes for sample rate, buffer, driver, direct-monitoring status, click sound, guitar input, input processing, calibration median, onset rule, and date. A spreadsheet needs six columns: take name, target sample, onset sample, signed error in milliseconds, absolute error, and marker confidence.
If you are unsure about a note, flag it instead of guessing. A pick scrape before a note, an accidental string hit, or a transient below a noise-gate threshold each deserves a flag. Exclude a note only with a stated reason, and keep both the raw and the valid-note counts. Quietly deleting awkward notes makes comparisons worthless.
Before trusting a report, check:
- The tempo stayed fixed within each three-take block; click sound, sample rate, buffer, and monitoring route stayed the same across all six takes.
- Three loopback passes were captured and their median offset recorded.
- No take was edited, warped, quantised, or copied into a comp lane.
- Every onset used one documented marker rule.
- Signed error, median absolute error, and spread used the same valid events.
- A result called hand timing carries no unexplained recording offset.
One precise question per recording
A repeatable DAW test turns vague frustration into evidence: you play steadily early or late, your spread grows, your downstrokes and upstrokes differ, or the delay comes from your gear and not your hands. The six-take template only has to be built once. Its numbers stay comparable only while the recording path stays the same, so every change of interface or monitoring calls for a fresh calibration. Within that limit the test answers one question at a time. Change a single variable (tempo, pick, monitoring route, or picking motion), record three fresh takes, and see whether the number you were chasing actually moved. Change two at once and you are back to the feeling you started with.