Tracking in a DAW: Sessions, Takes and Comping

A tracking session is everything between a placed microphone and a folder of finished files. The rate and depth you record at, the buffer size that decides how far behind your playing you hear yourself, the takes you keep, the names you hang on them. None of it is the performance. All of it decides whether the performance survives contact with next month.

The common mistake is treating setup as paperwork and the recording as the real work. So the sample rate gets picked from a forum argument. The buffer gets set once and never touched. Eleven takes pile up under names the software invented, and the session gets reopened in the spring by someone — usually you — who can no longer tell which guitar pass was the keeper.

44.1 kHz Already Reaches Past Your Hearing

Sampling captures a frequency faithfully as long as the rate is more than twice that frequency. So 44.1 kHz reaches 22.05 kHz — past the top of anybody's hearing, with 2.05 kHz of transition band left for the anti-alias filter. The number itself is a historical accident, inherited from the PCM adaptors that were the cheapest way to get audio from a studio to a CD plant. 48 kHz has one practical claim on you: video and broadcast expect it.

Above those two, the case thins out fast. For a finished file, 24-bit/192 kHz buys nothing audible, takes about six times the space, and can make playback slightly worse, because real gear intermodulates the ultrasonic content back down into the audible band. Two production arguments survive. A higher rate cuts latency at the same buffer size: 256 samples is 5.8 ms one way at 44.1 kHz and 2.7 ms one way at 96 kHz. And halving playback speed halves every frequency with it. A 44.1 kHz file carries nothing above 22.05 kHz to begin with, so slowed by half it has nothing left above 11 kHz — where a 96 kHz file still reaches past the top of hearing.

The third argument concerns saturation and amp-simulation plugins. A nonlinearity generates harmonics above the note it is fed, and anything above half the sample rate does not vanish — it folds back down as a lower frequency unrelated to whatever produced it. Whether the session rate is the right lever for that is a plugin question, not a recording one.

Twenty-Four Bits Is Headroom, Not Resolution

Bit depth does not describe how finely the waveform gets drawn. It describes how far down the quiet end you still hear music rather than quantising. The headroom lesson already did that arithmetic — 6.02 dB per bit, with real 24-bit converters landing near 120 dB A-weighted rather than the theoretical 144. Call it twenty usable bits, the bottom four holding the converter's own hiss.

What it buys at the moment of recording is permission to be conservative. Broadcast parks ordinary programme material 18 or 20 dB below full scale. Track a peak at −20 dBFS in 24-bit and you still have roughly 100 dB underneath it, which is more range than a 16-bit file gets when it is slammed all the way to the ceiling. The loudest moment has not happened yet, and leaving room for it is free. How hot to push belongs to the recording basics lesson; the format is not what is pushing you.

Twenty-four bits is where you record, sixteen is where plenty of material still gets delivered, and that last step wants dither. Truncating the spare bits leaves an error that tracks the signal, which the ear reads as distortion. Dither scatters the same error into noise, which the ear mostly ignores — the one decision in the chain that belongs at the very end rather than the beginning.

Your Buffer Setting Is Really Two Settings

Your interface hands the computer audio in blocks, and the block size — the buffer — sets how long each hand-off takes. Divide it by the sample rate for one pass: 128 samples at 44.1 kHz is 2.9 ms, 256 samples is 5.8 ms, 512 samples is 11.6 ms. Those are one-way figures. Treating them as what a player feels is the standard error, because a musician monitoring through the computer waits for two passes, one in and one back out, plus about a millisecond in each converter. A 128-sample buffer at 44.1 kHz therefore puts roughly 7.8 ms between the string and the headphone. At 512 samples it is about 25 ms.

Turn those round trips into distance and they stop being abstract. Sound covers 343 m every second, so 7.8 ms is like standing 2.7 m (9 ft) from your amp and 25 ms is like standing 8.6 m (28 ft) back. Exactly where it becomes unplayable is disputed, and the published numbers do not all measure the same path. Ten milliseconds is a common professional target. Singers hearing their own voice start noticing somewhere above 15 ms. A listening test across several instruments and monitoring types put the acceptable figure anywhere from 42 ms down to under 1.4 ms, with wedges far more forgiving than in-ears. There is no threshold worth memorising. Ask the drummer.

Which is why the buffer is two settings wearing one name. While tracking, the session is nearly empty and the computer has little to do, so run it small and spend the processor on being playable. While mixing, nobody is playing into it, so the delay costs nothing and you want the largest buffer that still lets you work. If the round trip is still too long at the lowest setting, stop fighting the computer and monitor directly through the interface. Hardware monitoring returns the input to your headphones at close to no delay — and costs you only the sound of your plugins while you play.

Keep the Take You Are About to Delete

A comp is an edited recording assembled from the best parts of several takes. Every DAW supports it under a different word — lanes, playlists, take folders — stacking passes on one track so you can audition them against each other and stitch a keeper out of the pieces. The mechanic is the same everywhere. The discipline around it is not.

Record whole passes rather than fragments. Four complete takes give you four readings of the phrasing; a singer stopped and restarted eleven times gives you eleven anxious openings and no ending. Resist reviewing between takes while the performance is still warm — listening back cools a room faster than almost anything else in the building.

Cut on rhythmic boundaries rather than mid-sustain, and crossfade every join, even where it already sounds seamless. Then do the part most people skip: play the finished comp against the single best take, start to finish. A comp drawn from six takes can be correct in every bar and still be worse than take three, because the small inconsistencies you edited away were the ones carrying the phrasing. Keep the source takes either way — a comp is an edit decision, and edit decisions get reopened.

Punch In on a Boundary, Not on a Note

Punching means re-recording one defined stretch of a take and nothing else. You set an in point and an out point, and the software plays the original outside that window and records inside it. On tape this was genuinely dangerous, because the pass overwrote what was already there. In a DAW it can be non-destructive — worth confirming yours is set up that way before the moment you need it.

Put the boundaries in silence where there is any, and on a downbeat where there is not. A punch landing mid-word, mid-cymbal-decay or halfway up a bent note shows its seam however good the new performance is, because the sound either side of the join is at a different point in its life. Give the player pre-roll too — a couple of bars of the original running in beforehand, so they arrive in tempo rather than cold on the beat you need.

Everything else has to match, and that is what actually makes punches fail. Same mic in the same place, same gain, same headphone balance, same distance from player to source. It is a strong argument for punching the same day rather than next week, and for photographing mic positions before you strike anything. Where the mic goes belongs to the placement lesson and the gain chain to gain staging; a punch just holds you to what you already decided.

Ten Minutes at the Start, or an Afternoon in Six Months

• Set rate and depth before the first take — changing later means resampling everything already recorded, and files at the wrong rate play back at the wrong pitch and speed.
• Name every track before you arm it — "Audio 7" tells the next person nothing, and the next person is usually you.
• Run a small buffer while tracking and a large one while mixing — one is a playability setting, the other a stability setting, and they were never the same number.
• Judge the round trip, not the buffer — two passes plus two converters, then hand it to whoever has to play against it.
• Reach for direct monitoring before you reach for a lower buffer — it costs you the plugins, not the performance.
• Drop a marker at every section as you record — a keystroke now, against scrubbing for the second chorus later.
• Keep every take a comp was built from — the comp is an opinion and the takes are the evidence.
• Save a dated version at each milestone — a session you can roll back is a session you can take risks in.
• Consolidate the audio into the project folder before moving it — then open the copy elsewhere and check it plays before deleting anything.
• Three copies, two kinds of media, one somewhere else — a second folder on the same drive is one copy, and drives fail whole.
• Leave five lines of notes beside the session — tempo, key, mic list, what is a keeper, what still needs redoing.