The word 'fine' can end a conversation or start an argument. Anyone who's tried to untangle a misunderstanding by email knows how much of a message lives outside the words. Live voice translation has that extra job, too.
On November 11, DeepL introduced Aure, a model it says follows changes in a speaker's tone, emphasis, emotion and pace as it translates. The aim is to carry the delivery into another language while keeping the speaker's recognizable voice. For multilingual meetings, that's an appealing promise: you might hear the hesitation along with the answer.
This builds on the voice preservation DeepL introduced in September. The company describes an interesting limitation of its earlier approach: an initial voice sample could establish who was talking, but couldn't keep following changes in how they spoke. Someone who starts calmly doesn't necessarily stay calm for the whole conversation. Aure continuously uses incoming speech and the translated words to shape the delivery, according to DeepL.
I would want to hear what happens to a polite reservation. In an illustrative supplier conversation, 'we could probably manage Friday' leaves room for uncertainty. A translation that sounds confidently cheerful could leave the listener expecting a firm delivery date. The words, the tone and the resulting understanding all deserve a place in the trial.
DeepL says Aure arrives automatically in its Voice products for online meetings, in-person conversations and the Voice API. It initially covers English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. Other supported DeepL Voice languages continue using existing models. Check the actual language pair you use before getting excited about the demonstration.
For a business working across languages, I'd start with a short internal conversation and a bilingual colleague who can compare both sides. Use familiar material: a change to an order, a tentative date, a disappointed response. Ask the listener what they think was agreed. A natural voice is pleasant; shared understanding is what makes the experiment useful.
Tell participants you're trying live translation, and keep the first exchange free of sensitive customer information. DeepL's Aure documentation says it doesn't retain a permanent voice profile and deletes temporarily processed voice and meeting data when the session ends. That's useful information for evaluating the service. The announcement alone doesn't establish that it fits every conversation your business handles.
These are DeepL's descriptions of its new model, rather than results from a test we've conducted. I'd be glad to hear a colleague's personality survive the language change. I'd still put the agreed delivery date in writing.
Sources and further reading
