Introducing Camera Coaching: Mouth Shape, Posture, and Breathing, On Your Device Only

Your ears can tell when a note is flat. They cannot tell you that your jaw never opened, that your head has drifted eight degrees off level for the last verse, or that your shoulders have been creeping up on every inhale for the whole song. A camera can. Coach Listens, the scored-singing side of the AI Vocal Coach, now has an optional camera layer that watches exactly those three things while you sing, and scores them alongside your pitch.

It is off by default. Check one box, "Add camera coaching (mouth, posture and breathing)," and it turns on for that session only. Leave it unchecked and Coach Listens works exactly as it always has, audio only.

What the camera actually watches

Three things, all of them things a camera can genuinely see, nothing more.

Mouth shape. How far your jaw actually drops on open vowels, tracked as a ratio against your own face so it works regardless of how close you are sitting. Sing a whole take through a mostly-closed jaw and the report will say so plainly, the same note a teacher would make watching you in a room.

Posture. Head tilt, measured off the line between your eyes. A little drift is normal; the coach only flags a sustained lean, and the report shows what percentage of the take your head spent tilted.

Breathing. Not your lungs, your shoulders. A visible shoulder rise on the inhale is clavicular, or chest, breathing: the first mechanical habit most vocal teachers correct by eye, because it is a real, visible tell and it is exactly what a camera can measure. The coach counts your breaths per minute and, when you are following a real melody, flags any breath that landed mid-phrase instead of at a natural break, which is usually the moment a singer runs out of air mid-line.

Nothing leaves your device

This is not a policy we wrote down after the fact, it is how the feature is built. Your pitch has always been analyzed locally, in the browser, note by note. The camera layer works the same way: face and body tracking run as on-device models inside your own browser, frame by frame, and the video never travels anywhere, not to HumMatch, not to any third party. There is no upload step to skip, because there was never a path for the video to leave. When the session ends, the camera stream is simply gone. What survives is a handful of numbers: how open your jaw tended to be, how often your head tilted, how many breaths you took and where. Nothing leaves your device. Pitch and camera analysis run entirely on this machine; only your scores are kept. If your browser cannot use the camera, or you decline the permission prompt, the session just continues audio-only. Nothing breaks, nothing is stored, nothing was ever at risk.

Yes, this coaches breathing. No, we didn't quietly change the rule.

When we introduced the AI Vocal Coach in July, we said the coach would never diagnose your breathing or your vocal folds, and meant it as a boundary against exactly the kind of overreach a lot of "AI coach" apps make: guessing at things a phone cannot actually know, like lung capacity or vocal fold health. That boundary has not moved. What is new is coaching, not diagnosis: the camera watches something genuinely visible, your shoulders, and tells you when the pattern looks clavicular instead of low and quiet, the same distinction a voice teacher makes on sight. It will never say a word about your lungs, your vitals, or your vocal folds, because a camera cannot see those, and the coach only ever talks about what it has actually measured. We have gone back and sharpened the wording on the original launch post to make that boundary explicit, since "breathing" now means two different things on this product: a habit the camera can coach, and a medical question it never will.

How to turn it on

Camera coaching lives inside Coach Listens, which needs an account and the Coach tier, $19.95 a month or $119 a year, the same subscription that unlocks unlimited coaching sessions and song coaching across the 66,000+ song catalog. Sign in, pick a sustained-note drill or a real melody to follow, check the camera box, and allow the permission prompt when your browser asks. The rest works exactly like an audio-only session, just with two more scores in your report.

Try it once, then watch the pattern repeat

One session tells you whether your jaw opens, whether your head drifts, whether you are breathing from your chest. A few sessions tell you whether it is a habit. That is the actual point of putting a camera in the loop: not a single verdict, but a pattern you can watch change, on the same measured foundation as everything else HumMatch does, and none of it leaving the room you are standing in.