Real-Time Translated Subtitles:
How They Actually Work

Real-time translated subtitles are captions generated from live audio and translated while the speaker is still talking. They appear on screen for a few seconds and then they're gone. That's the opposite of a subtitle file such as SRT or WebVTT, which is a finished artifact: a list of timecoded cues aligned to a recording that already exists.

Those two things get treated as the same product constantly, and that's why so much advice on this topic is useless. Search for real-time translated subtitles and you'll mostly be handed subtitling editors: tools that want an uploaded video file, a transcript pass, and twenty minutes of your afternoon. Great software. Wrong problem. Nobody uploads a file to a video call that's happening right now.

This guide covers what live-generated subtitles actually are, what they can't do, the one scenario the rest of the internet skips, and how to get them running on a browser tab in a couple of minutes. If you want to skip the theory, you can open MirrorCaption in your browser and point it at a tab.

What Real-Time Translated Subtitles Actually Are

A subtitle, in the traditional sense, is a timed artifact. The W3C WebVTT specification defines a cue as text bound to a start time and an end time, expressed against the timeline of a media file. That definition assumes the media is finished. You can't bind a cue to a timeline that hasn't happened yet.

So "real-time subtitle" is, strictly, a contradiction. What people mean when they search for it is: text on my screen, in my language, matching what I'm hearing right now. That's a live pipeline, not a document. Audio streams in, a transcription engine emits partial words that self-correct as more context arrives, a translation layer rewrites them into the target language, and the result paints on screen a beat behind the voice.

The distinction isn't pedantry. It determines what you can and can't ask for.

Real-time subtitles vs. the three things they get confused with

Column three is the one people conflate most often, and we won't relitigate it here: our guide to live captions vs. transcripts already works through that terminology properly.

Which of these three readers are you?

Three quite different people type this phrase into a search box.

1. The attendee. You're watching audio you don't control: a livestream, a conference broadcast, a webinar where you're a guest. Keep reading, this article is mostly for you.
2. The meeting participant. You're in a call where someone speaks a language you don't. Also keep reading, sections three and four are yours.
3. The person who needs a deliverable subtitle file. You want an SRT to attach to a video for other people to watch later. Real-time tooling is the wrong shelf entirely, and no amount of cleverness will change that. Use a dedicated subtitling editor on the recording. We'd rather tell you that in paragraph twelve than waste your afternoon.

The Case Nobody Writes About: When You Don't Control the Audio

Here's the gap. Almost every article on live translation assumes you're the host. Enterprise captioning vendors want to sell your events team a contract. Platform documentation tells you to ask the meeting organiser. Subtitling blogs tell you to wait for the recording to be posted.

None of that helps the person who is simply listening. And that person is common: an engineer watching a hardware launch stream from Shenzhen, a policy analyst following a parliamentary session, a student sitting in on a guest lecture, an investor on an earnings call where they're firmly in listen-only mode.

In every one of those cases you have no permission to change anything about the broadcast, and no relationship with whoever runs it. What you do have is a browser that is already decoding the audio. That's the leverage. The getDisplayMedia API lets you grant a page access to the audio of another tab, with an explicit permission prompt each time. The sound never leaves your machine's audio graph on its way to being captured; you're subtitling what you're already hearing.

Two honest caveats, because this is where a lot of marketing copy quietly lies. First, tab-audio capture is a desktop browser feature: it works in Chrome and Microsoft Edge, and it isn't available on mobile browsers. Second, no bot joining the meeting is a technical statement, not a legal one. Whether you may transcribe a given call depends on your jurisdiction and your employer's policy, and those questions deserve a real answer before you press start on anything external.

How to Get Real-Time Translated Subtitles on a Browser Tab

The setup is short enough that describing it takes longer than doing it.

1. Open the audio source in desktop Chrome or Edge. A Zoom or Google Meet call in the browser, a YouTube livestream, a webinar player, a news broadcast, anything that plays sound in a tab.
2. Open MirrorCaption in a second tab and choose Meet mode, which is the mode built for capturing meeting-tab audio.
3. Pick source and target language. There are 50+ selectable languages, and the pair is bidirectional, so Mandarin to German is as valid as English to Spanish.
4. Grant tab audio when the browser asks. The share picker appears; choose the tab playing the audio and make sure the audio toggle is on. This prompt is browser-enforced and appears every session.
5. Read. Original and translation sit side by side on desktop, stacked on mobile, with each voice labelled separately.

There's no install, no browser extension, and nothing joins the call as a participant. That last part is the difference between this and the meeting-bot category: tools like Fireflies need an assistant invited to the meeting, which is exactly the thing an attendee cannot arrange.

When the other person needs to hear it, not read it

Subtitles are a reading interface, which is fine when you're the only one who needs the translation. It stops being fine the moment the conversation is two-way.

Speak Translations covers that. It reads your own translated speech aloud in the target language, close enough to real time to keep a conversation moving. Playback can go through your laptop speaker, through a phone paired by QR code so the sound comes from the phone, or, on the Mac client, through a virtual microphone that Zoom, Meet, and Teams see as a normal input device.

On a phone, Talk mode handles the face-to-face version. It's a continuous session rather than a push-to-talk button: you start it once, both people speak in turns, and the transcript and translation context carry across the whole exchange instead of resetting after every sentence.

What to Expect from Live Translation Quality

Live translation is genuinely good now and it is not flawless, and any page that tells you otherwise is selling something. Here's the honest shape of it.

Partial results correct themselves. Streaming transcription emits words before the sentence is finished, then revises as grammar resolves. Text visibly rewriting itself mid-sentence is the system working, not glitching. It's also why a screenshot of live subtitles taken at the wrong instant always looks worse than the experience.

Audio quality dominates everything else. A clean headset mic in a quiet room is a different problem from a laptop speaker in an echoing meeting room with three people talking over each other. Clear audio is the single biggest lever you control.

Idiom is where machines still lose. When a Japanese client says ちょっと難しいです, a translator will render it as "it's a little difficult," which is linguistically correct and commercially a polite refusal. The translation isn't wrong; the reading is. This is precisely why MirrorCaption shows the original alongside the translation and lets you tap any translated word to reveal the source. If you negotiate across languages, keep one eye on the left column.

Expect strong results on clear speech in well-supported languages, and expect to verify anything that carries money or legal weight. We go deeper into this in our write-up on real-time translation accuracy.

What the Built-In Platform Options Cover

Every major conferencing platform now ships some form of live translation, and for a lot of teams that's the right answer. If your organisation lives inside one platform and your host has the right licence, using what's already there beats adding a tool.

• Zoom Translated Captions translate speech into another language during a Zoom meeting. Availability depends on the account's plan tier and on the host enabling the feature.
• Microsoft Teams live translated captions do the equivalent inside Teams, again gated by licensing and tenant-level admin settings.
• Google Meet translated captions cover a set of language pairs within Meet, subject to the Workspace edition in use.

Check each vendor's own current documentation for the exact language lists, because they change and any number printed in a blog post ages badly. The structural limits, though, don't move:

• They only work inside their own platform. A livestream, a webinar player, a browser-based radio broadcast, or an in-person conversation is out of scope by definition.
• They depend on somebody else's plan. As an attendee, you can't switch on a feature that the host's licence doesn't include.
• Captions are usually ephemeral. Exporting varies by plan; the running text often isn't yours to keep.

MirrorCaption sits outside all three of those constraints because it captures the tab rather than integrating with the platform. Tool-agnostic is the design goal: use whichever conferencing tool your host picked. If you're weighing the alternatives properly, our roundup of the best meeting translators in 2026 compares the category side by side, including where competitors are genuinely stronger.

Frequently Asked Questions

Can I get real-time translated subtitles on a livestream I'm only watching?

Yes, if the stream plays in a browser tab on your own machine. Tab-audio capture reads the sound your browser is already playing and subtitles it locally, so you don't need the broadcaster to enable anything. It's the standard route when you're an attendee rather than a host.

Can I download real-time translated subtitles as an SRT file?

Not as a timecoded SRT. Real-time subtitles are produced from audio as it plays, so nothing is aligned to a finished media file. MirrorCaption exports the running transcript and its translation as Markdown or plain text. For a distributable SRT, use a subtitling tool on the recording.

Do I need the host's permission to caption a meeting I'm in?

Nothing joins the meeting, so there's no bot to approve, but that isn't the same as a legal green light. Recording and transcription rules vary by country and by employer. Check your local law and your company policy before you caption a call, especially an external one.

Does real-time translation work in both directions during a conversation?

Yes. Talk mode runs as one continuous session, so both people take turns inside the same conversation instead of restarting per phrase. Speak Translations can optionally read your translated speech aloud, so the other side hears the target language rather than only reading it.

What do real-time translated subtitles cost?

MirrorCaption gives every account 1 free hour, one-time, with no credit card and no monthly reset. Annual is 54.99 euro per year with 100 hours of hosted transcription credit; Premium is 99 euro once with 200 hours included. Extra hours come from Voice Packs, sold separately.

The Bottom Line

Real-time translated subtitles are a live pipeline, not a document, and once you accept that the whole category makes sense. There's no file, so stop searching for one. What there is instead is a way to read audio you're hearing right now, in a language you actually speak, without waiting for a recording and without needing anyone's permission to flip a switch.

Four things worth carrying away. Real-time and subtitle-file tooling solve different problems, and mixing them wastes time. Tab-audio capture is what unlocks the attendee case that platform features structurally can't reach. Built-in translation from Zoom, Teams, and Meet is a fine answer when you're inside that platform and your host has the licence. And live translation is strong on clean audio but still misses idiom, which is why reading the original alongside the translation matters when the stakes are real.

If you're the attendee, the participant, or the person across a desk from someone you can't understand yet: the next hour is free, and it takes about ninety seconds to find out whether this solves your problem.