๐ November 11, 2026 โข DeepL press release
DeepL today unveiled Aure, a new AI voice model that powers DeepL Voice real-time translation. The announcement came in a company release from Cologne dated November 11, 2026. Aure continuously adapts translated speech to changes in tone, emotion, emphasis and pace, using both the translated words and the original audio. The previous version of DeepL Voice generated translated speech from a static voice sample captured at the start of a conversation. That approach captured who was speaking but not how, so sarcasm and mood shifts were lost in translation.
During training, the model separates voice identity, prosody and spoken content. Prosody is the melody of speech, and because the components are separated, each of them carries into the target language independently. Reinforcement learning keeps the output free of accent. Aure builds on the voice preservation technology DeepL launched in September, extending it from preserving a speaker's identity to preserving the speaker's delivery.
The model works automatically with no extra configuration in DeepL Voice for Online Meetings, DeepL Voice for In-Person Conversations and the DeepL API for Voice. At launch Aure supports 8 languages: English, German, Spanish, French, Italian, Japanese, Portuguese and Korean, with more to follow. DeepL Voice continues to serve other languages through its existing models.
For meetings and live conversations the difference shows up in delivery. Pauses, urgency and irony survive the crossing into another language instead of flattening into a synthetic monotone, so participants can read far more of a discussion from the translated audio. Teams working across languages get this without changing their workflow, since Aure applies to the existing DeepL Voice products automatically, and developers can add the same capability to their own products through the DeepL API for Voice.
DeepL also addresses the privacy question directly: the company says the technology does not permanently store meeting transcripts or translations and deletes session data from memory when a session ends.
Aure is available now in the three DeepL Voice products, with language coverage set to expand beyond the initial 8. For multilingual teams the model moves real-time translation a step closer to conveying not only what was said but how it was said. It also marks a direction for voice translation overall, from intelligible output toward output that preserves the speaker's delivery, which carries a large share of the meaning in live conversation.
Sources: prnewswire.com/news-releases/deepl-unveils-aure-its-most-lifelike-voice-ai-translation-model-yet-302935694.html - the-decoder.com/with-its-new-voice-model-deepl-voice-now-translates-intonation-and-speaking-speed-into-other-languages-as-well
Continue the conversation: https://t.me/AIAutomationArchitecture ยท Bluesky
Top comments (0)