Intel / article

Realtime Translation

Source: elevenlabs.io
Author: ElevenLabs editorial sitemap
Date: Published Oct 21, 2026

Context

ElevenLabs has launched Eleven Realtime Translation, a model that translates a spoken conversation while it is happening. Each person talks in their own language and hears the other side in theirs, rendered in a synthetic voice. The company pitches it at places where language is a staffing problem: a telecom operator can build it into the calling layer so calls on its network work across languages, a contact center can put it between and caller so one agent covers more languages without an interpreter joining the call, and a live event can give each attendee the speaker in their own language. Deutsche Telekom is named as an early user, building calling features for its Magenta AI Call Assistant.

The translation runs on Eleven v4 Turbo, which ElevenLabs describes as its low-latency speech model for realtime use. The practical point of that choice is voice consistency: the same voice performs across more than 90 languages with automatic language detection, and it stays fixed for the whole session. Customers can pick from more than 17,500 library voices or use an Instant or Professional Voice Clone, so a brand or an individual agent can sound the same in every market. Deployments also choose how quickly translation starts, trading a slower, higher-quality setting for complex calls such as billing or claims against the lowest latency for quick back and forth.

What the post does not supply is any measurement of that tradeoff. It gives no latency figure, no translation-quality result and no pricing, and the product is available only in alpha to enterprise customers of ElevenAPI. The announcement establishes the design and the intended deployments, not how well a translated call holds up in practice. It enters a category that already has a large-lab entrant: Google earlier shipped Gemini 3.5 Live Translate, also a speech-to-speech translation model.

Key takeaways · AI-distilled

Eleven Realtime Translation lets each person on a call speak their own language and hear the other in theirs as the conversation happens. It is in alpha for ElevenAPI enterprise customers, and Deutsche Telekom is building it into its Magenta AI Call Assistant.
It runs on the low-latency Eleven v4 Turbo model by default, covers more than 90 languages with automatic language detection, and keeps one chosen voice, from a library of more than 17,500 voices or a custom clone, consistent for the whole session.
Deployments can trade quality against latency by setting how quickly translation starts: higher-quality translation for complex calls such as billing or claims, or the lowest latency for fast back-and-forth.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.

Read the source elevenlabs.io