The scene is familiar from every pronunciation app of the last decade. You read a sentence into the microphone and a number comes back: 78%. You read it again, more carefully: 81. Once more, trying very hard: 76. At this point an honest question presents itself — what, exactly, are you supposed to do? The number tells you how it went. It does not contain a single instruction. It is a verdict where you needed a diagnosis.
A sentence of ordinary length holds thirty or forty individual sounds. A sentence score takes everything that happened across all of them and averages it into one figure — and averaging is exactly the wrong operation, because it destroys the one thing you need: which sound failed. Two learners can post the same 78% with nothing in common. One is losing points every time “th” slides into “s”; the other’s consonants are immaculate but the short “ih” keeps stretching into “ee”. Same number, different problems, different fixes. The average that makes the number tidy is the same average that makes it useless.
Worse, a headline number invites the wrong kind of practice. When the score is the goal, the natural move is to read the same sentence again, a little slower, a little more carefully — polishing everything a millimetre instead of fixing one thing properly. The score drifts up a few points and plateaus, because the actual substitution, the one doing the damage, was never named and never addressed.
Feedback you can act on has three properties. It is stated at the level where change happens — the individual sound, because “improve your pronunciation” is not a movement your mouth can make, but “let the tongue tip touch between the teeth” is. It is tied to your evidence — the actual words in your actual recording where the sound slipped, not a generic list of errors that other people somewhere tend to make. And it comes one thing at a time, because attention is the scarcest resource in the room, and forty simultaneous corrections are the same as none.
“Seventy-eight! And what is a seventy-eight — an opera reviewed by a spreadsheet? Tell me which note went flat, caro, and then we have something to rehearse!” — Nunzio, of course, has opinions.
This is why the Nunzio app never leads with a number. The verdict on a sentence is a hand gesture — one, two or three 🤌 — and the real payload sits underneath: every word coloured by the sounds inside it, each miss named in plain respelling (“ih” as in ship, never a bare phonetic symbol), and one fix chosen as worth your next ten minutes. The percentages still exist, tucked in an expander for the curious. They are simply never the headline, because a number has never once told anyone what to do with their tongue.
Read five sentences into the Nunzio app and it will name the sound that is actually costing you — scored on your own phone, one diagnosis at a time. The full cast of sounds it listens for is in the sound library.