The method
Learn to speak by speaking.
Say Something in Welsh proved that adults learn to speak by speaking. Cymanfa turns that loop into something you carry in your pocket, not a quiz to tap through or a streak to protect.
Four steps, on repeat
Hear it, say it, mean it.
Hear the cue
An everyday phrase, in English, spoken by a native voice.
Say it aloud
A short gap. You reach for the words and say them, out loud, before you are shown.
Hear it right
The answer, in a real first-language voice, never a robot. You judge how you did.
Meet it again
Spaced repetition brings each phrase back just as you are about to forget it.
Practice, your way
Say it. Hear it. Fill it.
One phrase, more than one way to make it stick, because speaking, spelling and reading are different muscles. Every mode runs on your device, and a near-miss still counts. No timer, no score, no penalty.
Speak aloud
Reach for the words and say them before you are shown. You judge how it went.
Dictation
Listen, then type what you heard. The one drill that trains your ear and your spelling at once.
Cloze
A sentence with one word missing. You supply it, in context, and a small typo still counts.
Every word, one tap
Tap a word. Know it.
Meet a word you do not know? Tap it. A plain definition appears right there, offline, drawn from Wiktionary, with a link to the full entry if you want to go deeper. No leaving the lesson, no hunting through a browser tab.
Real voices
A voice for every language.
Native recordings come first, always. Where a language has none yet, an open voice speaks the words on your device, clearly marked as such. And a community can bring its own voice by link, verified on the spot before anyone hears it.
Native first
Real first-language recordings lead. A machine voice never pretends to be a person.
On your device
The voice runs in your browser, offline, so a language with no recordings can still be heard.
Community voices
Add an open voice by link. We fingerprint it, check it is real, and you test it before it ships.
Never a grade
Feedback is a mirror, not a judge. No machine scores how you speak.