blog.google web signal

Google Gemini 3.5 Live Translate Spans 70+ Languages

google voice ai multimodal google gemini translation voice-ai

Key insights

  • Gemini 3.5 Live Translate is publicly available to developers now via the Gemini Live API and Google AI Studio.
  • Google Meet expands from 5 to 70+ languages, covering 2,000+ language-pair combinations for enterprise users entering private preview this month.
  • SynthID watermarking is applied to all translated audio output, embedding authenticity signals into every generated speech segment.

Why this matters

Google Meet's expansion from 5 to 70+ supported languages, covering 2,000+ language pairs, moves real-time multilingual communication from a premium add-on to an expected baseline in enterprise video conferencing. The Gemini Live API public preview gives developers direct access to continuous speech translation infrastructure, removing the need to build and maintain segment-based translation pipelines from scratch. Grab's deployment for driver-rider communication signals the model is already being stress-tested in high-noise, real-world consumer environments, not just controlled enterprise settings.

Summary

Google's Gemini 3.5 Live Translate is now in public developer preview, translating speech in near real-time across 70+ languages while preserving speakers' intonation, pacing, and pitch. The model stays just a few seconds behind the speaker, handles multilingual inputs without manual configuration, and watermarks all audio output with SynthID. Google Meet will grow from 5 to 70+ languages, enabling 2,000+ language-pair combinations for enterprise users entering private preview. Essentially: (Google, Grab, Agora) are deploying this as live multilingual infrastructure for consumer apps, developer platforms, and enterprise calls. - Android users get a new listening mode routing translations through the phone's earpiece, no headphones required. - Grab is piloting it for driver-rider communication; Agora, LiveKit, and Pipecat have already integrated it. The jump from 5 to 70+ languages in Google Meet raises the default expectation for what enterprise conferencing software should support.

Potential risks and opportunities

Risks

  • Developers building production applications on the Gemini Live API during public preview face service disruption or pricing changes when the preview period ends without disclosed terms.
  • Grab's deployment for driver-rider communication routes real-time conversation data through Google's translation infrastructure, creating data-localization exposure in markets with strict privacy regulations.
  • SynthID watermarking in translated audio may introduce compliance complexity for enterprises that need to archive or replay translated Google Meet recordings without disclosing AI involvement to regulators.

Opportunities

  • Agora, LiveKit, and Pipecat, integrated at launch, are positioned to capture developer mindshare as the default real-time communication stack for multilingual voice applications built on Gemini.
  • Enterprises in ride-hailing, logistics, and multilingual customer support can use Grab's pilot as a proof-of-concept to justify early entry into the private Google Meet enterprise preview.
  • Other real-time voice translation vendors now face pressure to match Gemini 3.5's 70-language threshold and intonation-preservation positioning or risk ceding enterprise conferencing and developer RTC markets to Google.

What we don't know yet

  • Exact latency figures beyond 'just a few seconds behind the speaker' are not provided, making it impossible to benchmark against competing real-time translation offerings.
  • Pricing for Gemini Live API access after the public preview period ends is not disclosed, leaving developers unable to model production costs.
  • Translation accuracy across the 70+ supported languages, particularly for lower-resource languages, is not addressed or benchmarked in the announcement.