DeepL launches Aure to preserve emotion in real-time voice translation

Aure adds changing expression to voice translation

Aure powers DeepL Voice, DeepL’s real-time AI voice translation product. According to DeepL, the model continuously analyzes incoming speech and uses both the translated words and the source audio to shape how each sentence is delivered in another language.

The company says this approach allows translated speech to reflect how someone speaks, rather than only what they say. A speaker’s pauses, emphasis, urgency, enthusiasm and changes in delivery can therefore be carried into the translated conversation. DeepL also says Aure can convey delivery that contradicts the words, such as sarcasm.

Jarek Kutylowski, CEO and Founder of DeepL, said the company wants to adapt translation to the way people naturally speak instead of asking them to change how they communicate.

From voice identity to voice behavior

Aure builds on DeepL’s recent voice preservation technology, which lets people communicate across languages in their own recognizable voice rather than a generic synthetic one. That earlier capability preserved the identity of a speaker’s voice. Aure is intended to capture how that voice changes throughout a conversation.

Previously, DeepL Voice could generate an expressive translated voice from a static sample. The opening moments of a conversation helped determine how that voice should sound, but a fixed sample could not follow changes in emotion, emphasis or pace. Aure instead adapts continuously, including before a speaker has finished a sentence.

DeepL says the model must separate voice identity, prosody and spoken content while filtering background noise, echo and audio distortions. Its training approach carries those elements into a new language independently. Reinforcement learning is used to refine translation accuracy and speech quality while producing natural sounding speech without an accent, according to the company.

DeepL says it does not permanently store meeting transcripts or translations. The data is deleted from memory when a session ends, leaving Aure to adapt to each speaker in real time. The model is designed to account for differences in how someone may sound from one conversation to another, such as appearing tired one day and enthusiastic the next.

Available across DeepL Voice

Aure’s capabilities are applied automatically, with no additional configuration required, across DeepL Voice for Online Meetings, DeepL Voice for In-Person Conversations and the DeepL API for Voice.

The model initially supports English, German, Spanish, French, Italian, Japanese, Portuguese and Korean, with more languages to follow. DeepL Voice will continue to support other languages through its existing voice models.

DeepL positions the technology for conversations where delivery can affect trust, authority and collaboration, including customer relationships, negotiations and international teamwork. The company says leaders at Pioneer Corporation, Aramark, Avendra International and Inetum are already using DeepL Voice for multilingual communication and collaboration.

For users, the update means that real-time translation can carry more of the speaker’s expression across languages, rather than presenting a translation based mainly on words and a fixed voice profile. DeepL Voice was also recognized in Fast Company’s 2026 Next Big Things in Tech list in the Applied AI category.