Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarElevenLabs Blog→ original

ElevenLabs Launches Live Voice Translation for Phone Calls in 90+ Languages for Business

ElevenLabs has opened an alpha of Eleven Realtime Translation, a voice translator for live calls. A customer speaks Spanish, an agent speaks German, and each hears the other in their own language. It runs on the Eleven v4 Turbo model, which knows more than 90 languages and detects on its own which language the person is speaking. A single agent can serve customers in many languages without an interpreter on the line. For now, access is limited to enterprise clients of the ElevenAPI.

AI-processed from ElevenLabs Blog; edited by Hamidun News
ElevenLabs Launches Live Voice Translation for Phone Calls in 90+ Languages for Business
Source: ElevenLabs Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceElevenLabs Blog ↗
Editor-in-Chief
Key metric #1Deutsche Telekom is building features on it for its Magenta AI Call Assistant
Key metric #2Voices are chosen from more than 17,500 options or cloned from your own; the voice does not change during a session
Translation timing is configurablemore accurate for billing disputes, faster for casual conversation

Picture a support call: a customer in Madrid speaks Spanish, and the only available agent speaks only German. Normally, the person waits on hold until someone with the right language is free, or struggles to explain the problem in a language that isn't theirs. ElevenLabs wants to remove that constraint. The company has opened an alpha of Eleven Realtime Translation, a live voice translation model in which each participant speaks their own language and hears the other in theirs. It covers more than 90 languages, and so far only enterprise clients of the ElevenAPI, ElevenLabs' developer platform, have access.

How it works in practice

The setup is simple. Translation happens while the conversation is ongoing and is delivered in a natural, expressive voice. There's no need to record a phrase, wait for processing, and then play back the result separately.

ElevenLabs describes three scenarios:

  • Telecom operators embed translation into the call layer itself. Any call on their network can then take place between people who speak different languages.
  • Contact centers place translation between the agent and the caller. One employee can serve customers in far more languages, without keeping an interpreter on the line.
  • Live events: each listener hears the speaker in their own language.

The first public partner named is Deutsche Telekom. It is using Realtime Translation to add new features to its Magenta AI Call Assistant together with ElevenLabs. Lena Drubel, Vice President of Consumer AI, called this a "completely new era" for voice and a way to remove barriers between people. These are the partner's words, not an independent assessment. The announcement contains no data on how the service performs under real-world load.

Voice, languages and settings

Underneath is Eleven v4 Turbo, a low-latency speech synthesis model that ElevenLabs has optimized for real-time tasks. It supports more than 90 languages and detects on its own which language the person is speaking. There is no need to select the language in advance.

The company says the same voice sounds consistent across different languages. For a brand, this matters: a support agent in Germany, Brazil and India will sound like the same "person." Voices can be picked from a library of more than 17,500 options or created as a custom clone, either Instant or Professional. The voice does not change during a session.

Another setting controls speed. The team decides how soon after the start of a phrase translation begins. For complex conversations such as billing disputes or insurance claims, it is better to wait and get a more accurate translation. For quick exchanges where the pace of replies matters, the minimum delay is chosen. This is an honest trade-off: the sooner the system starts speaking, the greater the chance it gets the meaning of a phrase wrong.

What to keep in mind

The announcement was written by the marketing team. It contains no latency figures, no translation quality metrics, no pricing, and no comparison with other solutions. The "alpha" status means the product may change significantly and access is restricted. The authors don't describe how the model handles accents, interruptions, line noise, or specialized terminology. Contact centers encounter all of these constantly.

There are also questions businesses will have to answer for themselves. Should a customer be told that the other party is speaking with a synthesized voice? How should records of translated conversations be stored? Who is responsible for a translation error if it causes a customer to misunderstand contract terms? For banks, insurers and healthcare services, these questions lead to legal risks, not just technical ones.

For developers and product managers, the takeaway is this. If your product depends on voice support or international calls, it's worth requesting alpha access and testing the service on your own call recordings. Pay particular attention to difficult scenarios: numbers, names, and plan names. The choice between speed and translation quality is best made separately for each call type, not once for the entire system.

Story

Key Questions Answered

How it works in practice?

The setup is simple. Translation happens while the conversation is ongoing and is delivered in a natural, expressive voice. There's no need to record a phrase, wait for processing, and then play back the result separately. ElevenLabs describes three scenarios: Telecom operators embed translation into the call layer itself. Any call on their network can then take place between people who speak different languages. Contact centers place translation between the agent and the caller.

What are the key details on Voice, languages and settings?

Underneath is Eleven v4 Turbo, a low-latency speech synthesis model that ElevenLabs has optimized for real-time tasks. It supports more than 90 languages and detects on its own which language the person is speaking. There is no need to select the language in advance. The company says the same voice sounds consistent across different languages.

What to keep in mind?

The announcement was written by the marketing team. It contains no latency figures, no translation quality metrics, no pricing, and no comparison with other solutions. The "alpha" status means the product may change significantly and access is restricted. The authors don't describe how the model handles accents, interruptions, line noise, or specialized terminology. Contact centers encounter all of these constantly. There are also questions businesses will have to answer for themselves.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…