How does sayit tell a real accent apart from a real mispronunciation?

sayit tells a real accent apart from a real mispronunciation by checking whether a sound difference is a near-merger — the kind of small, systematic variation real English accents produce, like R-coloring or a reduced final vowel — or a genuine substitution that changes which sound was intended. Near-mergers score as clear, with a quiet note that something was softened; real substitutions get flagged. The two are never treated the same way, even when the raw acoustic difference looks superficially similar.

30-second version: sayit keeps a list of known near-merger sound pairs — the kind of systematic differences that separate real English accents rather than separating a correct word from an incorrect one. When a detected substitution matches one of those pairs, it's penalized only lightly and still counts as a clear word, just marked as "reduced." A substitution that isn't on that list — the kind that would actually cost you the word with a real listener — gets the full weight of a genuine error. The product's own accent note puts it plainly: common accent differences are not marked wrong.

What actually counts as a "near-merger," concretely?

Sound pairs that real, standard varieties of English differ on systematically, without that difference ever confusing a native listener about which word was said. R-coloring is the clearest example: American English pronounces the /r/ in words like "car" and "hard," while most British and Australian English doesn't — dropping it isn't an error, it's simply how a huge population of fluent English speakers actually talks. Schwa neutralization — a reduced, relaxed vowel replacing a fuller one in an unstressed syllable — is another: it's how connected, natural English is supposed to sound, not a sign something went wrong.

How does that show up in my actual score?

A near-merger substitution is penalized only lightly rather than at full weight, and the word it's part of typically still reads as clear overall — you'll see it marked as "reduced" rather than flagged as wrong, a visible note that something was softened without it costing you the word. A genuine substitution that isn't on the near-merger list gets the full penalty, because unlike R-coloring or schwa reduction, it's the kind of sound swap that actually changes what a listener hears.

Why does this distinction matter so much?

Because an app that flags every deviation from one specific accent as an error punishes people for speaking correctly, just not the way the app expected — and that's a genuinely demoralizing, inaccurate kind of feedback. A learner deliberately speaking a non-rhotic variety of English shouldn't be told their "car" is wrong; a learner who substituted an entirely different consonant for one that changes the word absolutely should be told, because that's the kind of error that actually costs them being understood. Collapsing those into one flat "match or don't" rule gets the second case right at the cost of getting the first case badly wrong.

Is this the same thing as sayit adapting its scoring target to my accent?

No, and this distinction matters. The near-merger tolerance described here is about interpreting your output generously when it falls inside a real accent's normal range — it's live, and it's part of how every take is scored by default. It's a different mechanism from changing what target you're scored against based on a chosen accent setting, which — as covered in how the TTS voice works — is currently still American English across the board, regardless of your selected playback voice. Tolerance for natural variation and a fully accent-matched target aren't the same feature, and only the first is live today.

Accent handling at a glance

Does this mean I should stop worrying about my accent entirely?

For the features that are genuinely just regional color, yes — that's the point. What's still worth attention is anything that collapses two real English words into one sound, the kind of error covered in per-sound feedback, because that's the category that actually costs you being understood, accent aside.

Try it

Read a sentence with a natural non-rhotic or reduced pronunciation in sayit and see whether it scores as clear — the goal is that your accent, on its own, never becomes the thing being corrected.