Google Translate now translates your phone calls live in 70+ languages

You pick up your phone and call a supplier in Osaka. They speak Japanese, you speak English. Gemini Live Translate starts converting each sentence in real time, preserving your tone of voice and speaking pace. No interpreter on hold, no language switch required. The feature is free and already in the Google Translate app on your phone.

How does Gemini Live Translate work?

It translates while the speaker is still talking. Most translation apps wait until you stop speaking, then process the sentence. Gemini 3.5 Live Translate runs a few seconds behind the speaker, similar to a simultaneous interpreter in a conference booth, except this one fits in your pocket, handles 70 languages, and never needs a break.

Think of it as a live captioner for speech, except the output is audio in a different language. The model listens to both voices, identifies which language each speaker is using, and routes translations in both directions at once. You don't need to set a language pair before the call. Speak English and hear Japanese. Your counterpart speaks Japanese and hears English. The system figures out the rest.

Two things stand out compared to earlier translation tools. First, intonation is preserved: if you speak slowly and deliberately, the translation sounds deliberate too. Speak fast and confidently, and the AI matches that register. Second, every piece of translated audio carries an inaudible SynthID watermark, which makes it verifiable that the speech was AI-generated. That matters in regulated sectors where recorded calls require disclosure of AI involvement.

Automatic language detection covers more than 2,000 language pair combinations. You don't pre-select English-to-Japanese or Spanish-to-Mandarin. The model detects both sides and translates accordingly.

How do you set it up in the Google Translate app?

Open the Google Translate app on Android or iOS and tap "Live translate" in the bottom-left corner. Connect your earbuds and start talking. Translations play through your earbuds automatically.

On Android, there's an extra option worth knowing about. The listening mode routes translated audio through your phone's earpiece, exactly the way a regular phone call works. Hold your phone to your ear and you hear the translated voice directly. No earbuds, no speakerphone. For a call with a supplier in a noisy facility, this is the option you want.

iOS requires earbuds for now. That's a meaningful gap for spontaneous calls, and Google hasn't announced when it might change.

What does it cost?

For consumers: nothing. The feature is built into the free Google Translate app, with no Google One subscription or Workspace license required.

Developers integrating live translation into their own apps can access the Gemini Live API, currently in public preview at approximately $0.023 per minute. An hour-long meeting costs around $1.38 in API fees. For context, professional conference interpreters in the EU typically charge between $85 and $160 per hour depending on the language pair and specialization.

The global language services market reached an estimated $79 billion in 2025, with interpretation services accounting for roughly 32% of that total, according to industry analysts at Coherent Market Insights. Real-time AI translation doesn't displace all of it. High-stakes interpretation in legal, medical, and diplomatic settings still requires human professionals. For the everyday layer of international business communication, the cost equation has shifted considerably.

Which languages does it support, and how well?

Google supports 70+ languages across most major global language families, including combinations like English-Japanese, Spanish-Mandarin, and Arabic-Portuguese, without using English as an intermediate pivot language.

According to TheAIDaily's AI Workforce Statistics 2026, more than 60% of knowledge workers at multinational companies regularly communicate across at least two languages for day-to-day work. Real-time voice translation addresses a persistent friction point that video conferencing has only partially solved.

Google hasn't published quality benchmarks for specific language pairs. The comparison point is their own previous version: Google Meet previously offered live captions in only five languages, all routed via English. The 3.5 model brings direct translation between 70+ languages across more than 2,000 combinations, with no English pivot required.

Worth noting: systems that generate synthetic speech in the EU fall under the EU AI Act's transparency obligations (Article 50). AI-generated audio must be disclosed as such. The SynthID watermark embedded in Gemini Live Translate satisfies this requirement. If your organization records calls for compliance purposes, update your data governance documentation to reflect that translated audio segments are AI-generated.

When is it coming to Google Meet?

Google started a closed preview for Google Workspace customers in June 2026. A broader rollout is planned for later this year.

In Meet, a control in the meeting toolbar activates speech translation for all participants, regardless of which language they're speaking. The previous Meet offering was limited to captions in five languages, all English-centric. The upgrade to 70+ languages with more than 2,000 pair combinations is a meaningful change for multinational teams on Google Workspace.

For reference, Microsoft Teams has offered live translation via Microsoft Translator for several years, with support for 100+ languages. The technical difference is the translation approach: Teams uses turn-based translation (it waits for the speaker to stop before translating), while Gemini translates continuously. In fast-paced conversations with interruptions, that distinction is noticeable. In structured presentations, it probably isn't.

What are the limitations?

A few-second lag is real. In a fast back-and-forth, both sides are working around a delay that doesn't exist in a direct conversation. Natural rhythm changes when one side is always half a sentence behind.

The feature works best with clear speech in a quiet environment. Background noise, strong regional accents, and speakers talking over each other reduce accuracy. Google lists noise robustness as a design goal but hasn't published performance figures for specific real-world conditions.

There is no offline mode. You need a reliable internet connection. On a remote job site abroad with patchy connectivity, that rules out the feature.

For contract negotiations, legal discussions, or medical consultations where a mistranslated phrase carries real consequences, a professional interpreter remains the safer option. A specialist catches deliberate ambiguity, culturally specific phrasing, and what isn't said. The AI translates words, not subtext.

How does it compare to the alternatives?

Google, Microsoft, and Apple each offer free live speech translation. The differences come down to translation model, language coverage, and platform integration.

Google's advantages are continuous translation and intonation preservation. Microsoft Translator supports more languages and has a working Teams integration today. Apple Translate covers only 20 languages and has no meeting-room option. DeepL, strong in text translation, doesn't offer live speech features.

For enterprise buyers choosing between Google Workspace and Microsoft 365, the translation gap is narrowing. Teams has the head start on meeting integration; Meet has the more natural-sounding output. Both will evolve over the rest of 2026.

When does it still make sense to use a human interpreter?

Contract negotiations, legal hearings, and medical consultations where a mistranslated word carries consequences are still the right cases for a professional interpreter. An experienced specialist catches deliberate vagueness, culturally specific signals, and the meaning behind what isn't said. The AI translates utterances. It doesn't interpret intent.

For the majority of daily international business communication, that level of precision isn't the requirement. You're coordinating a delivery with a logistics partner in Vietnam. You're following up with a prospect in Brazil. You're troubleshooting a hardware issue with a manufacturer in Taiwan. For those calls, Gemini Live Translate is available, free, already on your phone.

Grab, the ride-hailing platform across Southeast Asia, is already piloting the feature for the more than 10 million calls drivers and passengers make each month. Two people with no shared language, connecting in real time through a translation layer neither had to configure. Earlier this month, Google also launched Gemini's ability to control your screen. The pace at which Google is embedding its AI models into existing products is striking.

The feature is live in the Google Translate app now. Test it with a colleague who speaks a different language, or call a foreign supplier and hold your phone to your ear. Within a few seconds, you'll know whether it fits your workflow.