All postsPost 1 of 7

Why we tag by ear, not by filename

Every sample library ends up named by whoever was in a hurry. Here is what the analyser listens for instead — and the three places it still gets things wrong.

6 min read
spectrogram · kick_808_punch_124.wav · 0–1.84 s[COVER ART PLACEHOLDER]

Open any producer's drive and you will find the same archaeology: a folder called kicks FINAL, another called from laptop 2019, and three thousand files whose names were typed in the ninety seconds before a session started.

Filenames are the worst possible index for audio. They record what you intended at the moment you saved, not what the file sounds like now — and they are written by the person least able to be objective about it. So Loopbox does not read them. It listens.

The filename problem

A name has to be decided once and then carries forever. Two years later snare3.wav is not a snare, it is a rim hit you nudged through a tape emulation and never renamed. The library disagrees with itself and you are the only one who can settle it.

A sample library is half tooling, half taste. Naming is where the two get confused.

What the analyser listens for

Four things, in order of how reliable they are. Onset shape tells us percussive from sustained. Spectral centroid and rolloff separate a kick from a tom and a hat from a shaker. Harmonic content gives a key when there is one. Tempo comes last, and only when there is enough material to be sure of it.

Onset windowHarmonic contentNoise floor
The first 40 ms carry most of the decision. Everything after it mostly confirms.
Listenkick_808_punch_124.wav · 1.84 s · MONO
0:00.63 / 0:01.84

Where it still gets things wrong

Three failure modes come up often enough that we now show confidence next to every detected field rather than pretending the guess is a fact.

  1. Processed one-shots. Heavy saturation moves the centroid up and a kick starts to measure like a tom.
  2. Two-second sources. Tempo detection assumes a loop. Most of a sample library is not loops.
  3. Genre on anything atonal. A dry rimshot belongs to every genre, so the confidence stays low and we say so.

What we are changing next

Confidence is going on the sample page next to each detected field, so a low number is visible before it costs you a bad search. After that: letting a correction on one file propagate to everything that arrived in the same folder, since those are usually wrong together.