Why My Shoulders Give Away My Breathing

A good vocal teacher does not diagnose your breathing by listening. They watch. Specifically, they watch your shoulders on the inhale, because shoulders that climb toward your ears the moment you breathe in are the single most reliable visible tell that something is off. Yesterday's post introduced the camera layer inside Coach Listens. Today is the part underneath it: why shoulders are the signal, and how a browser turns that signal into a number.

Clavicular breathing, and why it matters

There are two broad ways to take a breath before a phrase. A low breath drops the diaphragm, the ribs widen out to the sides, and the belly and lower back expand. Almost nothing above the collarbone moves. A clavicular, or chest, breath does the opposite: the shoulders and upper chest lift to pull air in, while the lower torso barely participates. It is the breath most people default to under stress, and it is also the shallower, less efficient one. Singers who breathe this way tend to run out of air mid-phrase, carry visible tension into the neck and jaw, and lose the low support that keeps tone steady on a long note. It is one of the first habits most voice teachers correct, and they correct it almost entirely by eye.

How a shoulder rise becomes a number

The camera coach watches the same thing a teacher would, just continuously and with more patience. Once you turn it on, an on-device pose model tracks your shoulders and hips, dozens of times a second, entirely inside your browser. A few design choices keep the reading honest.

It filters out normal drift. Everybody sways a little and slouches a little over a few minutes of singing. The coach compares your shoulder height against a slow-moving baseline, not a fixed starting point, so ordinary postural drift over a long session never gets miscounted as a breath.

The threshold is physical, not arbitrary. A visible chest breath typically lifts the shoulders somewhere around one to three centimeters on an average torso. The model scales that expectation to your own frame, using the distance from your shoulders to your hips (or, if your hips are out of frame in a tight seated shot, your shoulder width instead), so it reads the same whether you are sitting close to a laptop camera or standing further back.

Breaths need space between them. Two rises less than about a second apart get treated as one breath, not two, which keeps a shaky signal from inflating the count.

It says nothing until it has seen enough. A couple of frames of movement is noise, not data. The model waits for a few seconds of usable tracking before it will render any verdict at all.

What the report actually tells you

Three things come out the other side of a session, and each one maps to something a teacher would say out loud. Your breaths per minute across the take. Whether your session reads as a shoulder breather, meaning most of your detected breaths were the bigger, clavicular kind rather than the small, mostly invisible kind. And, when you are following a real melody instead of a drill, whether any of your breaths landed mid-phrase instead of at a natural gap between lines, which is usually the moment a singer runs out of air mid-word rather than choosing where to breathe.

What it will never tell you

A shoulder is not a lung. The model has no way to measure how much air you actually moved, how efficiently you used it, or anything about the health of your vocal folds, and it never claims to. It reports a visible movement pattern and nothing else. That boundary is not a limitation we are apologizing for, it is the whole design: the coach only ever talks about what it has actually measured, and a webcam has genuinely measured your shoulders moving, not your lungs filling.

The point of watching

One session tells you what your breathing pattern looked like today. The value shows up over several: does the shoulder-breathing flag fade as you get more comfortable with a song, or does it show up reliably right before the hardest phrase? That is the same question a teacher would ask across a semester of lessons, just answered from your own webcam, on your own machine, one session at a time.