What it costs, which videos work, how Stories, Paraphrase and Practice work, where your words live, and what happens to what you type and say.
No. You can use everything without one: where you reached in a video, the transcripts you have used and your settings are kept in your own browser, not on a server, which also means they stay on the machine you practiced on.
If you want the same progress on a phone and a laptop, press Connect devices and scan the code with your other device's camera. It is optional, there is no password and no e-mail. While devices are connected, your progress (places in videos, saved words, fixes and what you typed in them) is also kept on our server so your devices can read it, and you can delete that copy from the same box. The privacy policy has the detail.
Yes, and it is built for it rather than bolted on. You say which language you came to practice, and that choice stands even when a video's captions turn out to be in another one. On a lesson taught in Vietnamese about Chinese, you are offered the Chinese rather than quietly handed the narration. English, German, Chinese, French, Spanish, Italian, Portuguese, Vietnamese and Japanese each have explicit support: for Chinese that means checking character by character, word boundaries from a real lexicon, and a proper dictionary; for German, knowing which capitals carry meaning and which only start a sentence.
Speaking is the one part that is not everywhere. All 34 languages can be listened to, typed and looked up; ten of them — English, Chinese, Spanish, French, German, Japanese, Korean, Portuguese, Vietnamese and Italian — can also be spoken and scored. In the others the Say it switch is not offered, rather than being offered and doing nothing.
Yes. There is no paid tier, no trial that expires, and nothing about the dictation itself is held back. Nothing here is sold and nothing is advertised.
Almost any. Most videos already have a transcript, because YouTube writes captions automatically for anything with clear speech, and those work straight away. For the ones that do not, there are two more routes: DictaBooth can listen to the audio and transcribe it, or, where the subtitles are printed on the picture instead of supplied as a caption track, read them off the screen. The only videos it cannot help with are the ones with no words in them at all, and it says so plainly rather than letting you practice against nothing. A video that mixes two languages is usable too: you are asked only to type the sentences in the language you came for.
Popular videos in the language you chose, shown under the paste box and in an archive page for each of English, German, Chinese, Spanish and French. Pick a level, then tap a story to read it. Hard is the video's own words. Easy (A1–A2) and Medium (B1–B2) show a version an AI service has rewritten for reading, which your device's voice reads aloud. Press Practice this video to work on it. If you opened the story from the archive, closing it takes you back there.
You hear a part of the video, then say it in your own words. Hint opens a guide map: how the part is built (a cause and its effect, a contrast, steps), with the few words that carry each piece and some alternatives. Write or speak your version and press Check. Each piece of the map turns green, yellow or red for whether your version still says it, wrong words are underlined with the correction under them, and two signals say how it went: the key ideas, and the grammar.
It is a machine's judgement and can be wrong: it can miss a subtle slip, underline something that is fine, and be strict about small details in the meaning. If a suggestion sounds odd, trust your ear and look it up. It works in eleven languages, and better in some than others. What is sent when you press Check is covered under “Is what I type sent anywhere?”.
The full report has one Practice card. It lists each kind of practice that has something in it for you: vocabulary cards, the fixes from Paraphrase, words you missed in dictation, and words and lines to say again. Tap a row to practice that kind, or press Practice all, one after another: each finish screen then offers the next kind. The say-it drills use the microphone, so skip any kind you would rather not do.
Yes, free. Email us the video link, its language and the format you want; the details are on the About page. For anything else, get in touch there too.
They stay in your browser. Each one keeps its meaning, the sentence you met it in, and which second of which video that was, and they come back to you on a schedule until you know them: a day, three days, a week, three weeks, starting again when you miss one. Nothing about them is sent anywhere, which is also why they live on the machine you saved them on, so export them if you want them somewhere else. The flashcard deck, Anki, Notion, CSV and Markdown are all one click, and the deck in particular is a single file you can keep, open on any machine, and practice from without an account. If you save words in more than one language, the language you pick above the list is the one Practice deals and the one an export holds, and the menu says which before you choose.
Not when you practice dictation: the checking happens in your browser and what you type never leaves it, unless you have connected devices, when your progress is also kept on our server. Paraphrase is another exception, and it asks first. When you press Check, your version and the video sentence go to our helper, which asks AI language models to judge it. It is not kept, and nothing about you goes with it. If you speak your version with the microphone button, the recording goes to an AI speech model to be turned into text, and is not kept either. YouTube sees that you asked for the video, because it plays it, and nothing is requested from YouTube until you allow it. A dictionary is asked only when you click a word, and is sent that one word.
Two features use the microphone, and only after you press for them. Press Shadowing and there is a short count-in, then it records until you stop talking — five seconds of quiet ends it, and so does the Stop button. The tone check on a Chinese word never leaves your device, because the sound becomes a pitch line in the browser and is then dropped. Shadowing does leave: the recording goes to our helper and on to a pronunciation-assessment service, Microsoft Azure Speech, which returns the scores. It is not written to disk and not kept afterwards. If you would rather your voice stayed on your own machine, simply do not press it, because nothing else on the page uses the microphone. The privacy policy sets all of this out in full.
Shadowing proper means speaking along with a voice you can hear. Reading aloud means saying the line from the page with nothing playing. Both are worth doing and they are not the same exercise.
The Speaker setting beside the practice switches is which one you get. It starts on Silent, because most people practice on a laptop with no headphones, and there the microphone hears the speaker as well as you — which flatters the score. Set it to Original and the speaker plays under you, a beat ahead of the word lighting up, which is the shadowing version. Headphones settle it either way.
Because the service that scores it does not score that language. The pitch line under the row is drawn from your own recording and is there whatever language you are in — it is the number beside it that comes from elsewhere, and for several languages there is none to have. A dash says so rather than inventing one.
It depends where it came from, and the page says so beside the practice switches, on every sentence. There are five things that badge can say:
Wrong words and wrong timings both happen. If a line looks odd, suspect the transcript before you suspect yourself — and if a sentence is plainly broken, the fastest fix is to skip it rather than argue with it. We are not publishing an accuracy figure until there is one worth publishing: that takes a set of clips checked by people, per language, and it does not exist yet.
Fetching a transcript needs a helper that can reach YouTube, and it is not always up. When it is not, the page says so and shows three ways to bring the transcript yourself: a one-click bookmarklet, a browser extension, or copying it from YouTube's own transcript panel. Everything else works exactly the same afterwards.