English Shadowing with Anki: Build Cards That Make You Speak
Build Anki shadowing cards that make you speak before the answer: use a cue, reveal model audio, shadow it, record yourself if useful, then vary the line.
Put the communicative cue on the front, hide the full English target until the answer, require an audible attempt, then reveal model audio for shadowing and finish with one changed version.
If the full English sentence is already on the front, your eyes may finish the card before your mouth even clocks in.
Anki can schedule the card. It cannot decide whether you speak.
Current Anki already has most of the mechanics you need. Card templates control what appears on the question and answer sides. Notes can contain audio media. During review, Anki can replay card audio, and the current desktop review workflow includes Record Own Voice and Replay Own Voice. The Anki Manual documents those review actions here.
But none of those buttons automatically turn a recognition card into a speaking card. The review event has to be designed so silence is not the easiest path through it.
Anki schedules the return. Your card design decides whether you actually speak.
Build one simple note with five fields
Technically, Anki has you add notes, and templates generate cards from those notes. For this workflow, keep the note simple:
- Cue: the situation, intention or meaning prompt.
- Target: one useful model sentence.
- Audio: the model recording you have the right to use.
- Focus: one speaking feature — stress, chunking, linking, intonation or a tricky phrase.
- Variation: one instruction for changing the line.
Keeping these as separate fields matters. Anki’s own manual recommends separating information into fields because it makes card layouts easier to change later, and it recommends keeping cards simple enough to review comfortably. See Anki’s adding/editing guidance.
Front: give the job, not the answer
The front needs an awkward little silence built into it: the moment before Show Answer when you actually have to produce English.
A good front might say:
Situation: Politely ask a colleague to send the latest file before 12:00.
Your job: Say one natural sentence aloud. Then show the answer.
Do not put the full English target on the front. If your translation is so literal that it gives away the exact English wording, it is doing nearly the same thing.
Original practice example
Could you send me the updated file by noon?
Could you …? is a polite request frame. By noon means no later than noon. A useful speaking target is keeping Could you lighter while giving more prominence to send, updated file and noon.
Use this for workplace requests when you need an action completed by a deadline.
Front cue: “Ask a colleague to send the latest file before 12:00.” Say your response first. Then reveal this model and compare.
Explore more language-learning guides in Media-Based Language Learning.
If you said, “Can you send the new file before noon?”, that is not automatically a failure. It is a natural sentence with the same basic intention. Whether exact wording matters depends on what the card is testing.
Back: reveal the model, then shadow it
After your attempt, the answer side should give you the model you were missing:
- the target sentence;
- the model audio;
- one short Focus instruction;
- one Variation prompt.
Audio attached to Anki notes can be replayed during review. Current deck options also let you decide whether card audio plays automatically; if you disable automatic playback, you can trigger replay manually. See Anki’s current audio/deck options.
Now shadow the model. Do not spend the whole review analysing it. Listen for the one feature named in Focus, speak with the audio, and move on.
Original practice example
I’m not completely sure, but I think the meeting starts at three.
I’m not completely sure, but I think … marks uncertainty instead of presenting uncertain information as fact. The small pivot before but I think helps the listener hear the change from caution to your best answer.
Use it when you need to answer about a schedule or fact you have not fully confirmed.
Front cue: “Give the meeting time, but make your uncertainty clear.” After reveal, shadow the model, then replace the meeting and time with your own example.
Record your voice only when the recording answers a question
Anki’s current review screen includes Record Own Voice and Replay Own Voice. The manual is explicit that this review recording is temporary: it disappears when you move to the next card. That makes it useful for an immediate comparison, not for building a permanent portfolio of recordings. See the current Studying documentation.
Use it when you have a concrete question:
- Did I stress the contrast?
- Did I keep the phrase together?
- Did my question sound like the same communicative intention?
- Did I add a long pause where the model flowed?
If you intentionally want a recording attached permanently to the note, the current editor has a microphone control that records and attaches audio to the note. That is a different feature from the temporary review recording. Anki documents the editor microphone here.
Original practice example
What I meant was that we need a little more time.
What I meant was that … is a natural clarification frame. It lets you correct an interpretation without restarting the entire conversation.
Use it in meetings, messages and disagreements when your first point was misunderstood.
Focus: keep What I meant was as one launch chunk. Record yourself only if you want to check whether you broke that chunk into separate words.
Vary the line before you grade the card
Shadowing gives you a model. Variation tests whether the sentence can leave the museum.
Change one thing: person, time, object, reason, location, degree of certainty. Keep the useful phrase frame.
Original practice example
If it works for you, we could move it to Friday.
If it works for you softens the suggestion by leaving room for the other person. We could presents an option rather than a command.
Use it when rescheduling appointments, meetings or plans collaboratively.
Shadow the model, then change Friday to another day or replace move it with another proposed action.
This final spoken change is what keeps the card from becoming exact-script recitation.
Grade recall; coach pronunciation separately
Anki’s answer buttons are part of a memory scheduler. The current manual describes Again for an incorrect or unrecalled answer, Hard for a correct answer recalled with substantial doubt or delay, Good for correct recall with some effort and Easy for effortless correct recall. See Anki’s current answer-button guidance.
Do not quietly turn those four buttons into an accent score.
How should you grade a speaking card?
I could not produce the intended English at all.
Again. Retrieval failed.
I produced the intended sentence or core phrase correctly, but only after a long struggle.
Hard can fit: the answer was recalled, but with significant difficulty.
I produced a correct or natural-enough response with some effort.
Good. That matches the ordinary idea of successful recall with effort.
The response came immediately and accurately.
Easy may fit if it was genuinely effortless.
My wording differed from the model, but it was natural and communicated the same intention.
Do not automatically fail it. Decide what the card was testing. If the goal was the communicative intention, your answer may pass. If the goal was one specific phrase you deliberately wanted to retrieve, check whether that phrase appeared.
My English was correct, but my rhythm did not match the model.
Keep the memory grade about retrieval. Use the Focus field and optional self-recording for the pronunciation repair. Anki is not automatically scoring your prosody.
My pronunciation made the target word genuinely unclear.
If intelligible production of that word was an explicit target of the card, the target was not met. Repair it before treating the card as mastered.
Keep the deck small enough that you still have time to speak
Speaking cards take longer than silent recognition cards. That is not a bug. You are doing more.
Current Anki deck options explicitly warn that new cards create future review work, and the scheduler exposes daily limits and retention/workload controls. See the current deck-options documentation.
So do not copy someone else’s magical “new speaking cards per day” number. Watch your own review session. If you are starting to skip the spoken attempt, the shadow, or the variation just to clear the queue, your intake is too high for the practice you designed.
Do not distort FSRS settings just to force pronunciation repetitions. Let Anki schedule memory. Use the review event itself to make the due card spoken.
Optional: one note can create two different practice cards
Anki templates can generate different card types from the same note, including separate production and recognition designs. That can be useful later. The current Card Templates manual even gives language-learning production versus recognition as an example.
When would I use a second “shadow-only” card?
If you want one card whose front immediately plays the model audio for pure imitation, you can create a second card type from the same note. Keep the main production card separate so hearing the answer is not always the first event.
Won’t sibling cards give each other away?
They can. Anki’s current deck options include burying controls for sibling cards. If you generate multiple cards from one note, consider whether seeing one should delay the other so the first review does not make the next trivial.
But do not build an elaborate template system before you have one card you actually enjoy speaking through. The simple five-field version is enough to start.
Make the mouth part unavoidable
A giant deck of familiar English can still leave you silent if every review is solved by recognition.
Hide the model long enough to retrieve something. Speak. Reveal. Shadow. Change one detail. Then let Anki decide when the memory comes back.
The front should create a mouth-first event.