Skip to content
MicroEnglish
Search Help Contact us Start free trial Sign in

How speaking feedback works

Updated 16 September 2026

Feedback has three parts (Words understood, Sound clarity and Flow) produced by speech recognition compared with the reference line; it is guidance, not a grade.

After you choose Get feedback, speech recognition compares your take with the lesson's line in the accent you picked and gives three numbers out of 100: Words understood, Sound clarity and Flow, plus one tip for your next try. As the block says, "This is practice feedback. It does not affect your level."

The Speaking Beta block in a lesson with the model recording and the Record my voice button

Read the three numbers

  • Words understood is the share of the line's words the recogniser heard. Skip a word and it drops. In Your response practice there's no fixed line to compare with, so this one shows Not scored.
  • Sound clarity is how closely each word's sounds matched the model accent: the pronunciation number.
  • Flow is about rhythm: the pauses between words and how evenly the sentence moves. Restarts and long gaps lower it.

Below the numbers, each word of the line is shown; one marked Review was missed, added or clearly mispronounced. Tick Show detailed word scores in the block's Settings to see each word's clarity as "n / 100".

Use the tip

The tip picks the one thing most likely to help:

  • "One word was difficult to hear. Listen once more, then include every word." A word was missing.
  • "An extra word was heard. Try the sentence again with a steady rhythm." One crept in.
  • "Focus on “…”. Compare it with the model, then try once more." That word had the lowest clarity.
  • "Clear work. Repeat once more to make the rhythm feel automatic." The take was fine.

Move each number

  1. Words understood: play the Model recording again before you record, keep the text on screen with Listen and repeat, and say every word without adding any.
  2. Sound clarity: in Settings, set Playback speed to 0.75× and listen for the word marked Review. Say that word alone a few times, then the whole line. Check Accent for feedback matches the model you're copying.
  3. Flow: start a moment after the microphone starts, leave a little silence at the end, and don't stop to correct yourself mid-line; record again instead.

Know the limits

  • Noise. A fan, traffic or a second voice lowers every number. If you see "We could not score that recording. Check the room noise and try once more.", move somewhere quieter and stay close to the microphone.
  • Very short lines. With a three-word line, one small slip moves the numbers a lot; compare several tries.
  • Accents. The recogniser compares you with a British, American or Australian model and handles some voices and accents less well than others. A lower number doesn't mean people can't understand you; being understood is the aim, not sounding like the model.
  • Length. A take is one to thirty seconds; under one second, nothing is saved.

What the numbers don't touch

Speaking feedback appears in the Speaking Beta block and in Speaking history under Practice in your account. It never changes your listening score for the lesson, the "Average completed score" in your weekly report, or your level; see How scoring works for the listening side. Each recording you send with Get feedback uses one monthly AI Coach credit; checking the same recording again with Check feedback uses none. When they've run out the block says "Speaking feedback is temporarily unavailable or your allowance has been used. Your recording is still here." Credits refresh at the start of the next month; listening, Practice and Test never use them.

If a number seems wrong for what you clearly said, tell us the lesson and line: Contact us about a lesson.

Did this help?

Ask meAI tutor

AI tutor

Opening your tutor…