Translate a Meeting Live: What Zoom, Teams and Meet Give You in 2026

Before you buy anything, check what your account already has. Zoom, Microsoft Teams and Google Meet all ship live translation in some form, each behind a different licence, each with a different ceiling. The check takes about thirty seconds per platform, and the answer decides whether this is a purchasing question or a settings question.

The catch that follows is the same in all three: the built-in features translate in one direction, for one listener, on a licence that usually sits with the meeting's organiser. If your problem is a two-way conversation, or you are a guest in someone else's tenant, none of that helps you and the platform will not tell you so.

Check what you already have

Zoom. Join or start a meeting, open the caption controls and look for a language option next to captions. If translated captions are enabled on the account, each participant can pick the language they want to read. If the option is missing, the host has not turned it on in their web settings, or the account does not carry the entitlement.

Microsoft Teams. In a meeting, choose More actions, then Language and speech, then Show live captions. Open caption settings, go to Language settings and look for the Translate to toggle. If it is not there, neither you nor the organiser holds a licence that includes it.

Google Meet. Open More options then Settings then Captions, and look for Translated captions. Speech Translation, the newer feature that produces translated audio rather than text, appears separately and has to be switched on for the meeting by an eligible account.

If you found the option, stop reading the pricing sections below. If you did not, the table says why.

What the built-in options give you in 2026

Every fact in this table was checked against the vendor's own documentation on 22 September 2026. All four features move faster than most vendor pages get updated, so confirm on the day you buy.

On price, only one figure comes from a vendor. Zoom's own announcement lists the Translated Captions add-on at 5 USD per user per month on any paid account, and says the feature is standard on Business Plus, Premier, Enterprise Essentials and Enterprise Plus (zoom.com blog, post dated February 2024, checked 22 September 2026). Microsoft publishes no Teams Premium or Copilot price on the support pages that describe these features, so treat both as a paid add-on and check Microsoft's own pricing for your tenant and region.

The point nobody makes: all of this is one-way, for one listener

Read the "whose licence counts" column again. Three of the five features are gated on the organiser's side, which means the person who most often needs translation, the external guest sitting in someone else's tenant, is the person least able to turn it on. You cannot buy your way into Teams translated captions for a meeting organised by a customer who has not licensed them.

The second constraint is direction. Each of these features gives one reader one target language. That is the right shape for a presentation: one speaker, many listeners, each reading in their own language. It is the wrong shape for a negotiation, an interview or a support call, where two people speak two languages and both need to read the other. You can approximate it by having each side switch their own captions, but the transcript that survives will be single-language, and in Teams the interpreter's output is not in the recording at all.

The third is what happens afterwards. Google's Speech Translation is unavailable during recordings. The Teams Interpreter agent's translated transcripts are live only. If you need the translated text as a document, a subtitle file or something searchable next week, the built-in features are not designed to give it to you.

How live translation actually works, and where it stops

A simultaneous system cannot wait for the sentence to end. It has to commit to words while the speaker is still producing the clause that will decide what those words mean, and every design decision in the field is a trade between that wait and the quality of the guess.

Google Research describes the trade directly in its write-up of real-time speech-to-speech translation (19 November 2025, checked 22 September 2026). The parameter is called lookahead, and the finding is blunt: "While a longer lookahead improves translation quality by providing more context, it negatively impacts the real-time communication experience." Their system settles on roughly a two-second delay as workable for most languages. That is Google's number for Google's system, not a universal constant, but the shape of the curve is.

German makes the problem visible. Take "Ich habe den Vertrag gestern mit dem Lieferanten besprochen." The verb that fixes the meaning, besprochen, arrives last. A system that starts translating at "Ich habe den Vertrag" has already committed to "I have the contract", which is wrong, and has to walk it back when the verb lands. English does the same thing with long relative clauses: "the supplier we signed with last year, who has since been acquired, is asking for..." leaves the translation of the subject unresolved for a dozen words. This is why translated captions visibly rewrite themselves. It is not a bug in a particular product, it is the cost of going first.

Researchers treat revision as a design axis rather than an embarrassment. The evaluation framework in End-to-End Evaluation for Low-Latency Simultaneous Speech Translation (Huber et al., submitted August 2023, last revised July 2025) explicitly compares systems that revise their output against systems with fixed output, and measures quality and latency together, because either number on its own can be made to look good.

Two things follow for the person running the meeting.

One-way or two-way is a workflow decision, not a feature comparison. Decide before the meeting who needs to read what. One speaker to a mixed audience is a one-way problem, and everyone should be reading the same target language. Two people in two languages is a two-way problem, and each of them needs the other's words rendered into their own. Choosing the wrong one produces a session where half the room is reading a translation of a translation.

Machine translation is not interpreting. Where the stakes are legal, medical or a works-council matter, a qualified human interpreter is still the answer, and in many of those settings it is a requirement rather than a preference. A system that commits to a verb before hearing it is fine for following a product roadmap discussion and unsuitable for a disciplinary hearing. Use the software for comprehension and the human for consequence.

Where Kalima fits

Kalima's live translation runs on your own device rather than inside the meeting platform. It captures what your microphone hears and what your speakers play, transcribes it and translates it as people speak. Nothing joins the call, so nobody appears in the participant list, and there is no licence to buy on the organiser's side. That is the difference that matters for the guest-in-someone-else's-tenant case above: it works the same whether the meeting is in Zoom, Teams, Meet, Webex, a Slack huddle or a room with a laptop on the table.

You choose the mode when you set up the session, before you press record:

• One-way renders everything spoken into a single target language. This is the presentation case.
• Two-way handles two languages in both directions at once, so each person reads the other's words in their own language, side by side with the original. This is the conversation case, and it is the one the built-in features do not cover.

Both modes cover more than 60 languages, and translation is included on every plan rather than sold as a tier. Kalima detects the spoken language automatically and labels each part of the transcript, which matters in the very common German meeting where two English sentences appear in the middle of a German one: the segment carries a language badge and the translation follows the switch instead of breaking on it.

Afterwards the translation is a real artefact, not a live overlay. Viewers you share a session link with follow the translated transcript as it updates, and they never cost a seat. When you export subtitles, a track selector lets you choose the Original or the Translation, so you can hand a video editor an SRT or VTT file in the audience's language and keep the original separately. The full record, with both texts and word-level timing, comes out as JSON or as a printable PDF.

What it does not do: it produces translated text, not a synthetic translated voice in the call. If you specifically need spoken output, the Teams Interpreter agent and Google Meet Speech Translation do that and Kalima does not.

Before you record other people, tell them. Say at the start that you are transcribing and translating the conversation, and give people the chance to object. Meeting terms and local law both apply: in Germany, § 201 StGB protects the confidentiality of the spoken word, and the safe path is the consent of everyone in the conversation. The fact that Kalima does not connect to the meeting platform is a technical description of how it captures audio, not permission to record without saying so.

If you want to price it against what a translated-captions add-on costs per seat, the plans are on pricing.

Frequently asked questions

Can I translate a Zoom recording afterwards?

Yes, by uploading the file rather than replaying it. Kalima extracts the audio in your browser, so for an MP4 the video itself never leaves your device, and translation runs in the same pass as transcription at no extra time cost. Two limits to know: you pick one target language per upload, and file size and duration depend on your plan (30 minutes per file on Free, 180 on Plus, 300 on Pro and Team, with a five-hour ceiling on every plan). The longer walkthrough is in you have the recording but no transcript.

Do I have to be the host to translate a meeting live?

For the built-in features, effectively yes. Zoom translated captions depend on the host's account and their pre-session language settings, and Teams translated captions depend on the organiser's Teams Premium or Copilot licence. A tool that captures audio on your own device does not have that dependency, because it never asks the platform for anything.

Does Teams translated captions work if only I have Teams Premium?

Partly. Microsoft's support page states that if the organiser holds Teams Premium or Copilot, all participants can use translated captions without their own licence, and that if only individual participants hold one, only those people see the Translate to option. So your own licence buys captions for you alone, which is often enough, but it does not enable the feature for the room.

Can two people read each other in their own language at the same time?

Not with the built-in captions, which give each reader one target language. Two-way translation between a language pair, shown side by side with the original, is the mode to look for. In Kalima you pick Language A and Language B in the Translation step of session setup.

Is live translation good enough for a contract negotiation?

For following the discussion, usually. For the words that end up binding someone, no. A simultaneous system commits to a translation before the clause is finished and corrects itself afterwards, which is acceptable when you are keeping up and unacceptable when a specific phrase is the point. Use it alongside a qualified interpreter in those settings, not instead of one.

What is the difference between translated captions and an interpreter agent?

Captions are text you read; an interpreter agent produces audio you hear. Microsoft's Interpreter agent can speak the translation in a simulated version of your own voice, and Google Meet's Speech Translation does something similar. The trade is cost and reach: both are gated on more expensive licences, both carry usage ceilings (20 hours per person per month for Interpreter, 90 minutes per meeting for Speech Translation), and neither one leaves you with a translated document afterwards.

Related

• Live translation: one-way and two-way modes, the language list and how the translated transcript is laid out.
• Live transcription: what the underlying transcript does before anything is translated.
• Multilingual meeting transcription: the sibling problem, when people switch languages inside one sentence.
• Kalima for remote teams: the wider picture for distributed teams that do not share a first language.
• Kalima for events and conferences: translated captions for an audience, and share links for people who are not in the room.