Which apps transcribe a language lesson, and is the file yours to keep?

A lesson transcript has an eligibility problem and a filtering problem. Here's who on Zoom, Meet, italki and Preply can produce one, and how to find the words worth keeping. Your lesson ended 20 minutes ago and you want the transcript. Whether you get one was settled before you pressed record. Zoom's audio transcription requires "A Pro, Business, Education, or Enterprise account" plus cloud recording switched on. Free account, no transcript, and no toggle you flip tonight will retrieve it. Google Meet's transcripts run on Business Standard, Business Plus, Enterprise Starter, Enterprise Standard, Enterprise Plus, the Teaching and Learning Upgrade, Education Plus and Workspace Individual. A personal Google account is on none of them. Two language traps sit inside the eligible case. Zoom transcribes in English by default and you change it per file, so skip that and your Spanish lesson comes back in English. Meet's transcript languages are "English, French, German, Italian, Japanese, Korean, Portuguese, Spanish", which leaves a Russian learner with nothing. Clear all that and the filtering starts. This one takes your evening. An hour of talk, speaker labels down the margin, every "sí, sí" and false start preserved. Buried in there are the words you got wrong, and the file marks none of them. So use the search box. Search your own hesitations and restarts. Search for English words inside a target-language sentence. Then scan for any short burst where your tutor's label follows yours immediately, because that shape is a correction. Jump to those timestamps and you've got tonight's cards.

Who can actually make one, platform by platform

Start here, because for a lot of readers this section ends the page. Whether you can get a transcript of tonight's lesson comes down to your platform's billing tier, and you find out in about 30 seconds.

Zoom. The requirements list on Zoom's own transcription article has 3 lines: "A Pro, Business, Education, or Enterprise account", "Cloud recording enabled", "Audio transcription enabled".

A free Zoom account can't produce a transcript at all. Neither can a paid one recording locally to your laptop, because the transcript is generated from the cloud recording.

Clear all 3 and the output is genuinely good. "Audio transcripts are saved in VTT format", which is a plain text file you can open in TextEdit or Notepad, timestamped and attributed by speaker.

The transcript doesn't email itself to you. Sign in to the Zoom web portal, go to Recordings & Transcripts, open Cloud recordings, click into the meeting, and the transcript sits among the recording files.

Zoom supports 19 transcript languages and all 5 of Verbamor's are on the list, but "by default, audio transcripts for cloud recordings are generated in English."

The language is a per-file setting. Change it and "Zoom will reanalyze and transcribe the audio in the newly selected language." Do that before you skim anything.

Google Meet. Same shape of gate, different list. Transcripts need one of these editions: Business Standard, Business Plus, Enterprise Starter, Enterprise Standard, Enterprise Plus, Teaching and Learning Upgrade, Education Plus, Workspace Individual.

Your free gmail.com account is not on it, and neither is a paid Workspace tier below Business Standard.

Meet's transcript languages are "English, French, German, Italian, Japanese, Korean, Portuguese, Spanish." Russian is absent. If you're learning Russian on Meet, this route is closed and no setting reopens it.

Then check who booked the call. "Meeting transcripts are saved in the meeting organizer's Google Drive in the 'Google Meet' folder." Your tutor scheduled the lesson, so you're now asking a person for a document instead of downloading one.

italki. The Classroom's own audio "is saved on italki's private storage" and the separate Record Lesson button hands you a WEBM video in Chrome, Edge or Firefox only, so any transcript of an italki lesson is a job you do yourself. What italki gives you instead is a written summary, pulled apart in turning an italki lesson into flashcards.

Preply. Every sentence Preply publishes about downloading your lesson audio sits inside its Lesson Insights article, and Insights are English-only, so whether a Spanish or Russian lesson gets recorded at all is undocumented. Open Settings > Permissions and look for yourself, then read what a Preply lesson actually leaves you for the parts that survive in every language.

What lands in the file, and why it doesn't help yet

Check one setting first. Zoom is blunt about the default: "By default, audio transcripts for cloud recordings are generated in English."

So your Spanish lesson comes back as an English transcription of Spanish audio. Invented words, confident timestamps, all wrong.

Language is a per-file setting. Change it and, in Zoom's words, "Zoom will reanalyze and transcribe the audio in the newly selected language." Spanish, French, Italian, Portuguese and Russian are all on Zoom's supported list, so the fix exists for every language Verbamor teaches. It never happens on its own.

Now open the file. Zoom saves transcripts in VTT format, so every line arrives with a timestamp and a speaker label before any speech.

Hundreds of lines, in order, no paragraphs.

Some labels read Unknown Speaker, which is what Zoom writes when attribution fails. Any phrase is editable by hand in Zoom's web portal.

Here's the part that stops people. Nearly every word in the file is a word you already knew, because a lesson pitched at your level is comprehensible by design. The few items worth a card sit buried under sí, vale, and your tutor saying "mm-hmm" while you finished a sentence.

A transcript is a haystack a machine built for you, very fast, out of a haystack you were already standing in.

The text is flat. Your tutor's throwaway o sea and the verb you fumbled all lesson sit in the same grey.

I've read one of my own end to end, once. I skimmed the back half, which is the same as not reading it.

What the file gives you is coordinates. Every line is stamped, so you search it for a few things, jump to those timestamps, and listen. The next section is the search terms.

Both routes out of here filter the same file. Your own eyes with the timestamps doing the navigation, or a classifier that ranks what it finds. The manual one works tonight and installs nothing, so it goes first.

Twenty minutes with a transcript and no software

Get the file first. Sign in to the Zoom web portal, open Recordings & Transcripts, go to Cloud recordings, click into the lesson, and the transcript sits among the recording files. Download it and open it in any text editor.

Check the language before you read a word of it. Zoom generates the transcript in English by default and only reanalyses after you change the language on that specific file, so an unchecked Spanish lesson comes back as phonetic slurry.

You'll run 4 searches, write down some timestamps, and close the file for good. Every line already carries the two things that make this work: a timestamp and a speaker label.

1. Search your own stalling. Type um, then uh, then whatever you say when you're fishing for a word: como se dice, comment on dit, your own tic.

Every hit is a moment your mouth stopped ahead of your brain, with a timestamp attached.

2. Search for English sitting inside a target-language sentence. Search the, with a space on either side. In a Spanish or French or Italian transcript a hit is usually you dropping an English noun into a sentence you otherwise had: entonces yo fui al, um, the courthouse

3. Find the short tutor turns wedged between two of yours. Corrections have a shape in a speaker-labelled file: you speak, your tutor speaks briefly, you speak again.

The brief ones sandwiched inside your own turns are where you got fixed.

4. Search your tutor's lines for we say and you can also say. Tutors telegraph, and that phrase sits directly in front of the idiom they've decided you're ready for.

For every hit, write down the timestamp. The word can wait.

Now close the text editor and open the audio.

Jump to each timestamp and listen to the 15 seconds around it. People skip this, and it's the step that makes the whole thing work.

A word on the page is a word you recognise, which is the weakest evidence you have about whether you know it. The audio gives you her delivery, her speed, and the thing you were quietly embarrassed about at 00:23:41.

Say each item out loud before you write it down. In Karpicke and Roediger's 2008 Science study, students who kept being tested on 40 Swahili-English pairs recalled about 80% of them a week later, against 36% for students who kept restudying the same pairs. The testing effect has the long version.

Give the pass 20 minutes and stop when the timer goes, wherever you are in the file. I'd cap the list at 15 items even when the transcript holds more: a list you'll work through beats a complete one you won't.

That list of items with timestamps is where this page ends. Turning it into cards that survive November is a separate craft, and I've written that up in turning lessons into flashcards: where the blank goes, and why you should keep far fewer cards than you found items.

What a machine does to the same file

Ctrl-F gives every occurrence of a word the same weight. It can't tell the word your tutor stopped and fixed you on from the word you produced cleanly and moved past, and that distinction is the whole job.

A classifier reads the speaker turns instead, and tags each candidate word with a reason. The reasons are ranked: struggled, corrected and gap sit above taught, used and personal. The worked example is on the italki notes comparison.

So here's the door, if you've got a transcript open right now.

Paste it. Verbamor's paste screen takes up to 80,000 characters a go and runs the same classifier a recorded lesson does, so the VTT you pulled out of the Zoom portal an hour ago is a first-class input. Strip the timestamp lines or leave them in; it reads speaker turns either way.

If you hand it audio instead, transcription is ElevenLabs Scribe (scribe_v1) with diarization on, and that speaker split is what the reason-tagging stands on. The diarized transcript stays visible in the app, with pills for relabelling a speaker the machine got backwards.

There's no transcript export, so keep your own copy of the VTT.

Then you pick. Nothing becomes a card until you say so, and it's the part I'd defend hardest.

The tagging gets things wrong. A word you produced perfectly well can come back marked struggled because you paused to think.

One honest cost, and it's the real one. Slamecka and Graf (1978) named the generation effect: material you generate yourself is remembered better than the same material read.

Picking a word off a list is weaker encoding than composing the sentence around it, and the card-writing guide makes the case for doing it by hand. My position is that the cards you'll actually make beat the better ones you won't, and you can rewrite the sentence at the first review.

Verbamor is iOS and iPadOS only, 16.4 or later. No Android, no web.

It's $19.99 a month, or $129.99 a year with a 7-day trial on the annual plan. There's no free tier, an awkward sentence to write right after telling you the manual method works.

And it won't teach a beginner their first thousand words. It mines what a lesson contained,

When a transcription pipeline is the wrong answer

Two kinds of learner should close this page and go do something else. Better I say so now than you work it out a month and one subscription later.

So reread the VTT from your last lesson, right now, before the next paragraph. If you didn't, the pipeline was never about the transcript.

The second person takes good notes during the lesson. If you're writing down the word your tutor stopped you on, with the correction next to it, you're filtering by hand while the audio is still in the room. Every one of those notes is a card you can build tonight, and the card guide covers the design.

Two platform-specific answers live elsewhere. On italki, which parts of the summary you already get free and which sit behind Plus: that's here. On Preply, what survives a lesson in a language the Insights feature doesn't cover: that's here.

One limit on my end. Verbamor is iOS and iPadOS only, 16.4 or later, so on Android or at a desktop, one of the routes above is your route.

And it holds on to words you've already met, which makes it a poor way to learn your first thousand. A beginner is better served by a course for a few months, then this.

Questions people actually ask