Summary:

Aure updates DeepL Voice across meetings, in-person conversations and the API, with eight-language support and continuous adaptation to a speaker's delivery.

DeepL launched Aure on November 11, 2026, adding a new voice model to its real-time translation products. The change targets delivery as well as wording: DeepL says translated speech can follow changes in a speaker’s tone, emphasis and pace during a conversation.

Where Aure is available

According to DeepL’s product page, the model is applied automatically across DeepL Voice for Online Meetings, DeepL Voice for In-Person Conversations and the Voice API. Supported languages are English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. Other supported DeepL Voice languages continue to use earlier models.

For an existing deployment, that makes the language pair an important first check. A product’s overall language list should not be read as confirmation that every language receives this particular model. Teams evaluating access should use the relevant meeting, conversation or API product route.

What changes in the voice

DeepL describes a shift from setting vocal delivery using an opening sample to continuously using source audio alongside translated words. The intended result is a voice that responds when a speaker becomes more hesitant, excited or emphatic, rather than retaining a fixed delivery throughout the session.

The company says its training separates voice identity, prosody and spoken content, then uses reinforcement learning to refine the output.

A practical multilingual pilot

Evaluate translation accuracy and expression separately. Ask bilingual reviewers whether names, numbers and instructions survive correctly before judging whether a translated sentence sounds natural. Then compare a calm statement, a correction and an urgent request: does the change in emphasis preserve meaning, or accidentally strengthen it?

Test interruptions, background noise and switching speakers using the microphones and meeting setup your team actually uses. Record the language pair and conditions with each result. A polished sample in one language is a useful starting point, not a substitute for checking another team’s workflow.

DeepL’s Aure FAQ says it matches voices live without storing a permanent voice profile, and that meeting transcription and translation data are deleted from memory when the session ends. Organizations should still check their own recording settings and participant-notice requirements before a pilot.

Sources and image credit