The oral quiz flow

Some quiz questions ask you to speak your answer. Wisteria transcribes what you say and grades it semantically. This page walks through the flow.

Before you start

Microphone access — your browser needs permission to use the mic. The first time you try an oral question, you’ll be prompted. Tap Allow. Without mic access, you can’t take oral questions.
A quiet environment — background noise hurts transcription. A normal indoor environment is fine; a busy factory floor or open-plan office is harder.
Speak normally — you don’t need to slow down or enunciate unusually. Whisper handles natural speech well.

The flow

1. See the question

The oral question appears with:

The question text
A large mic button below
An empty transcript area (where your words will appear)

2. Tap the mic

The mic icon turns red. Recording starts. A timer counts up.

3. Speak your answer

Talk naturally. There’s no minimum or maximum length — answer the question as you would in real life.

For most questions, 15–45 seconds is the sweet spot. Shorter than 10 seconds might not give Wisteria enough text to grade against. Longer than 90 seconds is allowed but unnecessary.

4. Tap stop

The mic icon turns gray. Wisteria sends the audio to Whisper for transcription.

You’ll see:

A loading spinner
“Transcribing your answer…”

This takes 3–5 seconds depending on length.

5. Read your transcript

The transcript appears in the previously-empty area. Read it. This is what Whisper heard.

If the transcript looks wildly wrong (lots of words you didn’t say, key terms missing or misheard), you have an option to re-record:

Tap Retry recording (only available before submitting)
The transcript clears, mic resets to the start

If the transcript looks mostly right (allowing for minor mishearings), submit.

6. Submit

Tap Submit answer. Wisteria sends the transcript + model answer + keywords + threshold to Claude. Claude returns a score and pass/fail.

This takes another 3–5 seconds.

7. See your result

Passed — green check, your score, encouraging message. You advance.
Failed (attempt 1) — red X, your score, two options: Try again (re-record) or Skip for now.
Failed (attempt 2) — red X, message: “This question will reappear at the end of the quiz for two more attempts.” You advance.

You get TWO attempts on the spot. The question reappears at the END of the quiz as a retry copy, with two MORE fresh attempts. If those also fail, the question is marked wrong in the final results.

Why the retry-at-end pattern

Oral questions cause more anxiety than text ones. The first attempt is often worse than the actual capability — fear of the mic, fear of being heard, novelty of the format.

Wisteria gives you a second chance after you’ve cooled down by working through other questions. If you came back stronger and passed the retry, that reflects your actual capability better than your initial nervous attempt.

Tips for getting a good score

Cover the keywords

Wisteria’s scoring is 40% keyword coverage. If the keywords are greet, apologise, booking system, name, manager, confirm — make sure your answer uses these words (or recognisable variants).

You can’t see the keywords; you have to infer them from the question. A complete, thorough answer naturally uses the keywords.

Don’t read a script

Speak naturally. Reading a memorised script often produces flatter, less complete answers than thinking through the scenario. Claude grades meaning, not memorisation.

Don’t over-talk

90+ seconds of answer is more than needed. A focused 30-second response beats a rambling 2-minute one — more keyword density, clearer meaning, fewer chances for off-topic content.

If you’re stuck on a phrasing

Pause and restart. Whisper transcribes silence as silence — pauses don’t hurt your score.

You can also stop and re-record once, before submitting. Take the time to give a strong answer rather than a fast one.

What if the transcript is wrong

Whisper’s accuracy depends on:

Accent — works well across English accents; can struggle with very strong regional ones
Background noise — quiet beats noisy
Microphone quality — phone mic is fine; laptop mic varies; external USB mic is best
Speed — too fast or too slow hurts; natural pace is best

If Whisper consistently mis-hears a key term in your courses, tell your trainer. They can:

Re-word the model answer to use more transcribable terms
Lower the threshold to be more forgiving
Switch the question to fill-in-the-blank if it’s a vocabulary question

Privacy

Audio is sent to OpenAI Whisper for transcription. Whisper doesn’t retain audio for training.
Audio is stored by Wisteria for 30 days for dispute resolution, then deleted.
Transcripts are stored indefinitely as part of the audit trail (text, small).

If you have privacy concerns about audio leaving your device or jurisdiction, talk to your administrator. There’s no current EU/regional Whisper option but it’s on the roadmap.

Accessibility

Hearing-impaired learners — oral questions might be a barrier. Trainers should provide an alternative (fill-in-blank or written answer) for learners who can’t easily speak. Wisteria doesn’t currently auto-detect or offer alternatives; the workaround is to ask your trainer.
Speech-impaired learners — same. Alternatives can be configured per-learner.
Non-native speakers — Whisper works in many languages but performance varies. If you’re consistently struggling, try speaking in your strongest language and see if transcription is more accurate.