ElevenLabs tests live voice translation in more than 90 languages

ElevenLabs launched Eleven Realtime Translation, a live voice translation model, in alpha (a test step) for enterprise customers of its API (a software connection). Each person speaks in their language and hears the other person in theirs, in a natural voice, during the conversation. The model works in more than 90 languages with automatic language detection.

How it works

ElevenLabs names three uses:

• Telecom operators put it in the voice layer of their network.
• Contact centers put it between the agent and the caller.
• At live events, each attendee hears the speaker in their language.

A user can select one of more than 17,500 voices or a voice clone (a copy of a voice). The voice stays the same for all of the session. Teams set how fast translation starts: higher quality for complex conversations, or the lowest wait for fast replies.

The proof

ElevenLabs gave no test results and no numbers for speed or correct translation. Deutsche Telekom makes new voice features with ElevenLabs.

What is not known

ElevenLabs gives no price and no date for general release. The company does not give a number for how correctly the model translates.

Insight: ElevenLabs aims to sit inside the call path

ElevenLabs sells translation to telecom operators for the voice layer, and Deutsche Telekom makes voice features with ElevenLabs. On 20 October it also launched voice agent models. Mistral says that its Voxtral 3 models operate in the EU or on your servers. This looks like a move to control the conversation path, while Mistral offers control of where audio goes. Leaders must find where audio goes before they select it.