Pitch Detection

Pitch is a repeat spacing

Making a voiced sound means the vocal cords opening and closing rapidly, pushing air out. This opening and closing repeats on a regular beat, and the more times it repeats per second, the higher the sound registers to the ear. A low voice repeats roughly a hundred times a second; a high voice repeats about two or three times that.

Instruments work the same way. A string that's thicker and longer vibrates slowly and produces a low sound; one that's shorter and tighter vibrates fast and produces a high one. Pitch, in the end, is just this repeat rate.

That makes pinpointing pitch a different job from measuring a sound's loudness or brightness. Sing loud or soft, in a rich voice or a thin one — if the repeat spacing is the same, it's the same note.

Sliding to find the overlap

The most familiar way to measure the spacing is exactly this business of laying stamped patterns on top of each other. A slice of sound gets copied and slid a little at a time, checking how well it matches the original. Slide it a tiny bit and it's off; slide it exactly one repeat's worth and it lines up cleanly. That amount of sliding is the repeat spacing.

There's a way to find it on a sound picture too. A voice shows up in that picture as a set of parallel stripes stacked at even intervals. Read the lowest stripe, or the spacing between stripes, and out comes the base pitch.

Either way, a cleaner sound matches better. Mix in noise, or overlap several sounds, and the pattern you're laying on top gets blurry, so the value jumps around.

Miscounting one cycle throws the octave off

The repeating pattern hides a trap. It lines up if you slide it one cycle, but it lines up just as well at two cycles, or three. Mistake a two-cycle shift for a single repeat and the pitch reads a full octave lower than it really is. Mistake a small ripple inside the pattern for a whole repeat, and it reads an octave too high.

This is called an octave error, and it's the single most common mistake in pitch detection. When a singing-score app suddenly drops points on what felt like a clean note, or a pitch graph plunges and snaps back mid-note, this mistake is usually behind it.

The fix is to look at what comes before and after. A human voice doesn't leap a full octave from one note to the next out of nowhere, so a value that jumps out alone from the surrounding flow gets treated with suspicion and corrected.

Some sounds have no pitch

Not every sound has a height to it. A clap, a door closing, a hissing or breathy consonant — none of these have a repeating wave. Force a value out of a stretch like that and you get a nonsense number. That's why pitch detection first judges whether there's any pitch worth measuring here before it tries to measure one.

Several notes ringing together are hard too. When the sounds making up a chord overlap, their repeating patterns tangle into several layers, making it much harder to pin down a single thread. In cases like that, the sounds sometimes get split apart first, with pitch pinned down for each one afterward.

Where it's used

An instrument tuner is the most familiar use — pluck a string and it tells you right away whether the note is sharp or flat against the reference. Singing-score features work the same way, continuously tracking the pitch sung and comparing it against the original melody.

On the music-making side, it's used to nudge a sung pitch slightly into tune, or to turn a hummed melody into sheet notation. It's used in speech too — pitch rising at the end of a sentence signals a question, falling signals a statement, so pitch values are needed for both reading and shaping intonation.