Perio by voice · 7 of 9

When Mesio mishears

It will mishear things. What matters is what happens next — and why a chart that occasionally stops is safer than one that never does.

Every voice system mishears. The question a practice should ask is not whether it happens but what the software does about it, because the failure modes are not equally dangerous.

The dangerous failure is the quiet one

A misheard word that produces obvious nonsense is harmless. You see it, you fix it, you move on.

The failure that matters is the one that looks fine. A value lands on the wrong tooth, or a call is swallowed as though it were conversation, and everything after it shifts by one. Nothing looks wrong on screen, because a chart full of plausible numbers looks exactly like a chart full of correct ones.

Mesio is built around that specific risk. Almost every design decision in the perio work exists to turn a quiet error into a loud one.

What you actually see

• It stops and names a tooth. The most common signal. Mesio heard something it could not place, so it holds the cursor and tells you which tooth it needs. You say the tooth and carry on. It costs you three seconds and saves you a quadrant.
• It marks a surface as unsure. When someone is assisting, or when you have told it not to interrupt you, an uncertain site is flagged rather than guessed. Somebody clears the flag later with the audio in front of them.
• It leaves a gap. If it is confident a value was spoken but cannot read it, it leaves that tooth empty and moves on, so the rest of the arch stays aligned. One visible hole beats a silent shift.

It gets a second chance before it gives up

A short phrase heard on its own is much harder to understand than the same phrase heard in context. That is true of people and it is true of software. So something Mesio cannot make sense of is not discarded on the first attempt — it gets reconsidered with more to go on before anything is declared lost.

In practice you see this as calls that simply arrive correctly a moment later, and as a section that fills itself in as you work through it.

What it will not do

It will not invent a measurement. It will not move a value it is unsure about to a tooth that would make the sequence look tidy. It will not decide that a word it does not recognize was probably a number.

All three of those would make the chart look better and make it less true, and the whole point of charting by voice is that the record is worth as much as the one typed by hand.