DeepL’s Aure carries tone into live translation in eight languages

DeepL introduced Aure, a model for live voice translation, on November 11. The company says the model adjusts translated speech as the speaker changes their tone, emphasis or pace. Korean is among the first eight languages. In a supported conversation, the listener can hear the translation in their chosen language along with changes in the way the speaker delivers it.

DeepL says it applies Aure automatically to supported languages across online meetings, in-person conversations and its Voice API. The first eight are Korean, English, German, Spanish, French, Italian, Japanese and Portuguese. Other languages continue to use earlier models. DeepL lists transcription, subtitles and spoken output separately in its Voice language guide, which already lists Chinese among its spoken outputs.

Translation follows changes during the conversation

In September, DeepL introduced voice preservation so listeners could distinguish speakers by sound in a multilingual meeting. That release established the speaker’s vocal identity from a sample at the beginning of the conversation, according to the company. DeepL says Aure uses the original audio alongside translated words as the conversation continues. If a speaker speeds up later or stresses a particular word, the change can shape how the translated sentence sounds.

DeepL calls the setting that makes translated audio resemble the speaker Speaker Match. Its help center says the service takes a temporary sample from the current conversation and discards it at the end of the session. Speakers can adjust how their translated voice is heard. The setting requires no prerecorded sample or voice profile carried over into another meeting.

A voice license for each online listener

DeepL Voice works with Microsoft Teams, Zoom and Google Meet. When a meeting is connected, a bot joins to stream audio for translation and posts a link in the meeting chat. DeepL lists access for outside participants and an enabled chat among the meeting requirements. Participants select their preferred language in the translation window. DeepL supports up to 300 people in a meeting. Breakout rooms fall outside its translation coverage.

Each person who wants to hear the spoken translation needs the DeepL Voice desktop app and an eligible Voice license. Captions have a different licensing rule. The person connecting the meeting needs a license, and the other participants can follow the captions from the link without a DeepL account or subscription. DeepL says spoken translation minutes are charged to the allowance of the person who connects the meeting, while each listener still needs an eligible license.

Meeting text and voice samples have separate retention rules. DeepL says it keeps the latest messages temporarily in memory and removes them at the end of the call. Transcripts or translations participants have accessed can remain on their devices for up to 30 days, with the period set by an administrator.

An in-person conversation can run on a device shared by two people. DeepL describes a one-to-one mode in which speakers take turns and the app shows and reads out the translation. Its iOS and Android apps also offer a split view so people facing each other can each read their side of the screen.

DeepL also offers a group mode for in-person conversations. A licensed host starts the session and shares a QR code or link. Participants join on their own devices without needing their own plan or app. DeepL instructs participants to connect a headset when speaking.

DeepL’s Voice API lets developers add live transcription and translation to an application. The developer reference allows translated text in up to five languages in a session. Applications that also play translated speech request audio in one language when they open the session. If the request omits that option, the API returns text.

DeepL says Aure applies automatically for supported languages in existing Voice API integrations that request audio. Developers still choose the language and output format. The company plans to add more languages and gave no dates for those additions.