Any app that puts a number on your speech owes you an explanation of what the number means. This is that explanation, including the parts that are limitations rather than features.

What happens to an attempt

You read the articulation cue for a sound, you record when you are ready, and the recording is uploaded. Before anything is scored, the audio is checked for whether there is something usable in it at all. If the recording is silent, or the room drowned you out, the attempt is not counted against your monthly allowance. A false start should not cost you anything.

If the audio is usable, it is analysed and comes back as a score together with a written explanation of what to adjust. The written part is the point. A score on its own tells you that something was off; the explanation tells you which thing, in language you can act on.

The derived result is kept as your practice history, and how far that history goes back depends on your plan. Raw audio is treated differently from the scores, with its own retention window, and you can request deletion of your recordings at any time.

What the score is measuring

It measures how closely the sound you produced matches the target sound you were practising, in this recording, in this room, on this microphone. That is a narrower claim than it might first appear, and all four of those qualifiers matter.

  • It is about one target at a time: the sound, word or sentence you were asked for, not your speech in general.
  • It is about this attempt. One score is a data point, not a verdict. The pattern across days is the useful signal.
  • It is affected by conditions. A noisy room and a poor microphone both lower a score without your production changing at all.
  • It is a comparison to a target, not to other people. There is no leaderboard and no percentile.

The limits worth knowing

A model of speech is trained on speech, and the range of speech it has seen determines what it handles well. Accents that are less represented in training data tend to be scored more harshly, and that is a property of the technology rather than a fact about the speaker. If a sound you know you are producing correctly keeps scoring low, that is worth telling us about rather than absorbing as failure.

Recording conditions matter more than people expect. Background noise, a microphone across the room, and speaking too far from the device all degrade the input before anything is analysed. Most unexplained low scores are environmental.

And a score is a measure of a production, not of a person. It cannot tell you whether you were understood. Being understood involves context, familiarity, lip-reading, gesture and patience on both sides, and no number captures that.

What we deliberately do not measure

Some of these were decisions rather than gaps, and they are worth naming.

  • We do not measure or estimate your hearing. Sign & Talk is not an audiometry tool and it makes no assessment of hearing loss.
  • We do not diagnose anything. It is educational practice software, not a clinical service, and it does not replace a speech-language pathologist or an audiologist.
  • We do not rank you against other users. Your history is compared to your own history and to the target, and to nothing else.
  • We do not score how "normal" you sound. The target is a specific sound, not a voice we have decided is correct.
  • We do not run streaks or send guilt notifications. A missed week is not a thing the product has an opinion about.
  • We do not judge whether you should be speaking at all. That is your decision, and it is not a question a pronunciation tool has any business answering.

A score tells you how close this attempt was to a target. It does not tell you how well you communicate, and it was never going to.

How to read your own history

Look at the shape rather than the last number. A sound is going well when the low attempts are getting rarer, not when a single attempt is high. Progress in motor learning is uneven by nature, and a dip after a good week is ordinary rather than a sign of going backwards.

When a sound falls apart inside a sentence after it was reliable on its own, that is not lost progress. It is the next stage of difficulty arriving on schedule. Take that sound back to word practice in the position where it failed, then bring it forward again.

And treat any single attempt as disposable. The reason there is no countdown timer on a recording, and no penalty for an unusable one, is that an attempt you can throw away is an attempt you can make without dread. That is the whole design intent behind the numbers.

Sign & Talk is educational practice software from Vinxign Media Private Limited. It does not diagnose or treat anything and it is not a substitute for a speech-language pathologist, a teacher of the deaf, or an audiologist.