DeepL launches DeepL Voice for real-time text translations

Anyone who has ever sat through a multilingual meeting knows the pain of translation lag. Two seconds of dead air is all it takes to completely derail a live conversation.

Now, DeepL is finally taking on speech. The German AI translation unicorn just launched DeepL Voice, a real-time tool that listens to spoken audio and instantly turns it into translated text captions.

Subtitles first, clones later

The product is built around two distinct everyday setups: face-to-face chats and virtual meetings.

For in-person talks, there is a mobile mirror mode. You put your phone flat in the middle of a table, speak naturally, and the screen shows translated text facing both speakers simultaneously.

For virtual calls, DeepL Voice generates live translated subtitles. Right now, it exclusively integrates with Microsoft Teams, which DeepL says covers the bulk of its corporate base. Support for Zoom or Google Meet is still up in the air.

On the language side, DeepL can currently listen to 13 languages, including English, German, Japanese, Korean, French, and Spanish. It can output subtitles in all 33 languages already supported by its standard translator.

Notably, DeepL Voice is purely text-based. There are no synthesized voiceovers or voice cloning features yet, and no public API at launch.

According to CEO Jarek Kutylowski, that deliberate choice comes down to speed. Generating audio on the fly still adds too much latency for live, natural dialogue, whereas text can be processed virtually instantaneously.

To soothe enterprise privacy jitters, DeepL noted that while voice data must travel to its servers for processing, none of it is stored or used to train its AI models.

Why it matters

For corporate teams living inside Microsoft Teams, this is an immediate quality-of-life upgrade. DeepL has long had a reputation for beating Google and Microsoft on nuance, and having that engine deliver instant meeting subtitles removes a massive language barrier without requiring clumsy third-party bots.

For the broader AI ecosystem, this marks an interesting philosophical contrast. While startups like ElevenLabs and Panjaya are racing toward expressive voice cloning and lip-synced dubbing, DeepL is prioritizing raw speed and translation accuracy. Interestingly, Kutylowski pointed out that ElevenLabs actually uses DeepL under the hood for its own translations.

We think skipping synthetic voice output for now is a smart move. In high-stakes business meetings, nobody cares if an AI sounds like you if it makes you wait three seconds to respond. By doubling down on text subtitles and low latency, DeepL is playing directly to its core enterprise strength.