Flashcards for pronunciation: spaced repetition that hears your voice

30-second version: Traditional flashcards self-grade — you decide if you remembered, and the deck trusts you. sayit's flashcards replace that with a real check: say the word out loud, the pronunciation engine scores it, and that score drives the spaced-repetition schedule automatically. A self-grade option still exists for when there's no mic, but the default is your actual voice.

Spaced repetition works because it schedules review right before you're about to forget — not too early (wasted rep), not too late (you've already lost it). That scheduling depends entirely on one input: did you actually know it? Standard flashcard apps get that input by asking you, honestly, right after you've looked at the answer. It's an easy question to answer generously, and a generous answer quietly breaks the whole schedule.

What's different about a pronunciation card

A sayit flashcard front shows a word. Instead of flipping to check a definition and self-rating, you say the word out loud. The same scoring engine that grades a full recording — checking the sounds you actually produced against the target — scores that single word, and the resulting pronunciation score maps directly onto the deck's grade scale (again / hard / good / easy) instead of you choosing it. A weak recording schedules a sooner review. A clean one pushes the interval out. No self-assessment step in between.

If there's no mic available, or a take doesn't score (too little of the word was captured, say), a self-grade fallback still exists — the card stays honest about whether it was actually graded, rather than silently defaulting to a fake pass.

The scheduling underneath

sayit's flashcards use the SM-2 algorithm — the classic spaced-repetition method, not a black box. Each card tracks an ease factor, a current interval, a rep count and a lapse count; a low grade shortens the next interval sharply and increases lapses, a string of good grades stretches the interval out. A card counts as mastered once it's been reviewed at least three times with an interval of three weeks or more — a plain, visible bar you can watch a deck fill toward, not an opaque "mastery score."

Where the cards come from

Three sources feed a deck, and they cover different needs:

• Auto-built from your own mistakes. Any word flagged during normal practice — read-aloud, listening, anywhere — collects into a "My mistakes" deck automatically, deduplicated by word. This is the highest-value deck in the app, because it's built entirely from things you've demonstrably gotten wrong, not from a generic list.
• Built from any word you look up. Define a word anywhere in the app and you can turn it directly into a card, IPA and definition pre-filled.
• Custom decks you build yourself, from a word list, a book you're reading, or anything you paste in.

See flashcards for the review queue and deck view.

A routine that actually uses the voice-grading

1. 1.Clear the due queue daily, even briefly — a handful of cards a day beats a big backlog once a week, because SM-2's scheduling assumes roughly-on-time reviews.
2. 2.Don't second-guess a low score into a manual override. If the engine graded a review "hard," trust it — the whole point of automatic grading is removing the generous self-assessment that breaks the schedule.
3. 3.Let the "My mistakes" deck run itself. You don't need to curate it — every flagged word during regular practice adds itself, so the deck is always current with what's actually giving you trouble right now, not what gave you trouble a month ago.
4. 4.Add words deliberately, too. Auto-collection catches mistakes; it won't catch a word you want to learn but haven't attempted yet — add those from any lookup card or a pronounce list.

Try it: open Flashcards, clear whatever's due, and notice how a card's next interval changes based on how you actually said the word — not on how you felt about it.