How to assess speaking skills: listen before you sign

How to assess speaking skills: score each criterion from a recording, use the transcript to navigate, and let an examiner who listened sign the result.

Here is how to assess speaking skills consistently: score each criterion of a speaking rubric — pronunciation, fluency, coherence, grammar, vocabulary — from a recording rather than from memory, and have a named examiner listen before the result is signed. That is the workflow NUADU runs for spoken responses: transcription with playback, criterion-based scoring, CEFR-referenced where applicable, and a marker who reviews at audio level. For language centre managers and heads of assessment, one distinction does most of the work: the transcript helps the examiner find their way; the audio is the evidence.

How do you assess speaking skills?

Assess speaking against written criteria, from a recording, with a named examiner deciding. Speaking assessment is the evaluation of a learner's spoken language against defined criteria, usually through a structured task — an interview, a short presentation, a response to a prompt. Four steps make it consistent:

  1. Write the criteria and their descriptors before the exam, so every examiner scores against the same rubric.
  2. Record every response, so a score never depends on what someone remembers.
  3. Score each criterion separately against its descriptor, instead of forming one overall impression.
  4. Listen again before signing wherever the evidence is thin or a score sits on a level boundary.

The usual failure is the opposite of all four. Written papers can be second-marked and objective sections scored by rule, but a speaking exam is often one examiner, one room and one impression, written down after the candidate has left. When a candidate asks why they were placed at B1 and a classmate at B2, the honest answer is sometimes that two examiners heard two different things — and with no recording, nobody can show which.

Calibrate the examiners before the session as well. Have each of them score the same few recorded responses, compare the scores criterion by criterion, and agree what each descriptor sounds like in practice. A disagreement found in a calibration meeting costs an hour; the same disagreement found in an appeal costs far more.

What are the criteria for assessing speaking?

A typical speaking rubric scores five criteria, each with descriptors for every level:

  • Pronunciation — whether sounds, stress and intonation let the listener understand without effort.
  • Fluency — pace, hesitation and self-correction across the response.
  • Coherence — whether ideas connect and the response holds together.
  • Grammar — the range and accuracy of the structures used.
  • Vocabulary — the range and precision of words for the task.

With AI assistance on, NUADU applies your speaking rubric across the same five; the descriptors stay yours. Where scoring is CEFR-referenced, the descriptors are tied to the levels of the Common European Framework of Reference for Languages, which describes what a language user can do at each level. That makes a result readable beyond the centre that produced it — the point of a CEFR-aligned test. A CEFR reference is not a CEFR certificate.

Replay, review the transcript, adjust scores, sign off.

Can AI assess speaking?

It can draft an assessment; it should not be the one who listens. With AI assistance on, NUADU transcribes each spoken response with playback, applies your speaking rubric criterion by criterion, and returns draft scores with rubric-anchored explanations. The marker then replays the response, reviews the transcript, adjusts any score the recording does not support, and signs off. The AI never finalises a speaking grade; the rest of the marking, rubric and moderation workflow is on the Grade page.

The order matters most for speaking, because a transcript flattens exactly what the rubric measures. Pronunciation is not in the text. Fluency — hesitation, self-correction, pace — is only partly there. Coherence can read better on the page than it sounded in the room. So the draft score is a starting point, the transcript is a way to find the moment that matters, and the audio is the evidence. NUADU's AI Transparency Notice adds a practical limit: speaking and pronunciation scores depend on microphone quality and background noise. One more reason the signature belongs to someone who listened.

The written side of the same workflow — draft score, note per criterion, override, sign-off — is described in AI-assisted marking, and there is a live demo of spoken-response marking.

How do you score a speaking test?

Score every criterion, keep the reason with the score, and release a profile under the examiner's name. A speaking exam a centre might run — a per-criterion rubric, AI assistance drafting scores, the examiner signing off — ends with a per-criterion profile released to the candidate: where they stood on each criterion, not one band with no explanation.

Score analytically, one criterion at a time, rather than with a single overall band. An overall band is quicker to give, but it cannot tell the candidate — or a reviewer — which criterion held the result back. A profile can, and it gives a second examiner something specific to check.

That profile is also what a dispute is about. If the candidate contests the grade, the file they see is the one the examiner signed: feedback anchored to the rubric, the marker's notes and the result for each criterion. The institution's appeals process takes it from there, with the audit trail of AI outputs, overrides and sign-offs behind it.

Keep one tool out of the score. NUADU's pronunciation feedback gives phoneme-level feedback to language learners during practice and is never used in high-stakes scoring; teachers decide whether to act on it. A learner should be able to practise sounds without every attempt becoming evidence in a grade, and the live demo of pronunciation feedback shows that practice side.

Two things to confirm before you schedule. Language coverage, including for AI-assisted marking, is confirmed per deployment — name the languages your centre tests. And the boundary: NUADU delivers and scores English assessments for placement, progress, preparation and institutional requirements. It is not a national certification body and does not issue state-recognised language certificates — it complements them.

→ See the English exams path