How do I actually read the per-word IPA verdict on my result?

By treating the three verdicts as confidence levels, not just a pass/fail grade — good means clearly on target, check means a specific, serious error was caught with enough confidence to name it, and unclear covers everything honestly in between, which is most of real speech.

30-second version: Every scored word gets a verdict, and the target-vs-produced IPA strings underneath it are the actual evidence. good means the word scored above a high confidence threshold. check means the system found a genuinely serious error — a clear wrong sound or a dropped consonant — and is confident enough to flag it directly; it's deliberately rare, because it's the only verdict that amounts to "you mispronounced this." unclear is everything that isn't confidently one or the other: not a flat accusation, but not a clean pass either.

Why "check" is rarer than you'd expect

It's tempting to read unclear as a soft version of check — a lighter shade of wrong. It isn't. check specifically requires a serious error (a clear substitution, not a minor sound-merger, or a dropped consonant) combined with either a short word or a low overall score, and even then, if the underlying acoustic confidence was low, the system pulls back from check to unclear rather than making an accusation it isn't sure about. That last part matters: a low-confidence "wrong sound" on fast, natural speech is more often the recognizer being unsure than a genuine mispronunciation, so the scoring is built to stay quiet in that case rather than flag something that might not be real. unclear isn't a lesser insult — it's usually the honest answer.

This is also why short function words — a, the, to, of — almost never get flagged as check even when something about them sounds slightly off: words under a small phoneme-count threshold are excluded from that verdict entirely, because they're too short and too noisy in fast speech for a confident, specific accusation to be fair.

Reading the target-vs-produced comparison

Underneath the verdict, a flagged word shows two IPA strings side by side: the target (what the word should sound like) and the produced (what your recording's acoustic evidence actually matched), aligned phone by phone. Four things can happen at each position:

• Match — the sound landed.
• Substitution — a different sound than the target was detected (target /θ/, produced /s/, for instance).
• Deletion — a target sound has no matching evidence at all (a dropped consonant).
• Insertion — an extra sound appears with no target to match it (an added vowel, for instance).

These four categories are why the feedback can be specific rather than vague — "unclear" alone would tell you almost nothing; "target /θ/, produced /s/, at position two" tells you exactly what to drill.

The "sounded like" hint

When a substitution or deletion happens to produce a real English word, the result shows that word directly — "sounded like 'sink'" on a "think" that came out as /s/-initial. This is one of the fastest ways to understand an error, because hearing that your mispronunciation spelled an actual different word makes the stakes of the fix concrete in a way an abstract phoneme comparison sometimes doesn't.

What to actually do with this

Don't chase a single check on one take — chase a pattern. If the same substitution (say, target /θ/ produced as /s/) shows up across several different words in a session, that's your real curriculum, distinct from a one-off flag that might just be a noisy recording. The features overview covers how this rolls up into focus areas across sessions, which is the more reliable signal than any single word's verdict.

Try it

Open sayit free — no install, no card — and record any sentence. Open a flagged word specifically and read the target-vs-produced comparison rather than just the headline verdict; that's where the actual, actionable information lives.