What did AI actually solve?

The new option is listening to a translated meeting with speaker-matched delivery on supported desktop computers. Confirm language support and listener licenses first, then judge it on a conversation where mistakes are easy to correct.

01

What changed

On September 15, 2026, DeepL announced new speech models that carry characteristics of a speaker’s voice into live translation. The company also made voice translation generally available through a desktop app spanning Teams, Zoom and Google Meet. This is a change in the available meeting experience, not a claim that machine translation is now error-free.

DeepL’s current help pages describe the feature as Speaker match: synthesized speech resembles the person talking instead of using a standard preset. It works only with DeepL’s own speech-generation models; languages served by another provider use a preset voice. The feature matches vocal characteristics, not someone’s identity.

The listening path matters. A bot receives the meeting audio and the desktop app plays its translation while silencing the original meeting sound. The app does not take over your microphone or deliver your remarks back into the call. Each listener chooses a language on their own device.

02

What this could change for you

For someone who works across languages, translated audio offers a way to follow a discussion without reading every line of captions. Voice matching adds a familiar-sounding delivery to that option. Whether it makes a particular conversation easier still depends on the speakers, language and translation quality; no measured benefit is established here.

For the documented desktop route, DeepL lists Voice Starter, Voice Business and Voice for Meetings Business as eligible plans, and every person listening to translated audio needs their own license. Translation minutes come from the meeting translation manager’s allowance. Caption-only participants have different rules and can read a connected meeting’s translations without their own license.

Install the desktop app on supported Windows or Apple-silicon Macs and sign in with the appropriate account. The published requirements specify Windows 10 or later on x64, or macOS 14.2 or later on M1 or newer hardware; Intel Macs are excluded. Check the current plan price and your organization’s permissions before arranging a group trial.

Try it yourself

Check access before a multilingual meeting

For the documented desktop route: supported Windows or Apple-silicon Mac, a DeepL account and eligible Voice license for each audio listener; confirm language support, pricing and organizational permissions. Browser-beta access is not established here.

  1. Check your computer, listener licenses and language support against the official documentation. Confirm that external participants and meeting chat are permitted.
  2. Install and sign in to the desktop app, then connect an ordinary Teams, Zoom or Google Meet meeting.
  3. Choose your listening language and start voice translation; verify an access code if prompted. Compare unclear phrases with the speaker.

What to watch for: This is a suggested check based on current instructions, not a firsthand test. Speaker match availability differs by language; captions and audio have different licensing rules.

Read DeepL’s meeting translation instructions ↗

Suggested activity. Official guidance checked .

03

What it does not prove

The evidence establishes an announced release and its documented controls and access rules. We have not tested the app firsthand or verified a gain in comprehension, meeting speed or translation accuracy. A familiar-sounding voice can still carry an incorrect translation.

Voice matching is not universal across languages. DeepL’s current help documents the generally available desktop route and says voice translation is unavailable in its web app. The September release separately mentions a browser beta for external participants; its access conditions are not established here. The requirements above cover the desktop route.

Use an ordinary, low-stakes conversation to assess the experience. Ask speakers to confirm important details when a translation is unclear; this product update does not establish suitability for medical, legal or other consequential interpretation.

Primary sources

DeepL: Voice preservation release

DeepL product release · September 15, 2026

View the announcement ↗