DeepL said in its 11 November release that it had launched Aure, a model for its real-time translation product DeepL Voice that adjusts translated speech as a speaker’s tone, emotion and pace change. Earlier DeepL Voice technology used the opening moments of a conversation to set how a translated voice would sound. DeepL says Aure instead keeps analysing the original audio alongside the translated words, so later changes in delivery can affect the output. The company says it trains the model to handle voice identity, rhythm and emphasis, and spoken content separately, then uses reinforcement learning to refine the result.
Discussing a purchase across languages could, if Aure works as DeepL describes, carry a change in urgency into the translated voice rather than keeping the tone set at the start of the conversation.
DeepL says Aure is available in DeepL Voice for Online Meetings, DeepL Voice for In-Person Conversations and the DeepL API for Voice, initially supporting English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. DeepL says it clears transcripts and translated text from memory at the close of a session instead of retaining either indefinitely.
