DeepL Aure Is Live: Voice Translation That Keeps Your Tone

Most real-time translation still sounds like a weather report. The words arrive in the right language, but the person behind them disappears: the pause before bad news, the rush of excitement, the emphasis that turns a routine update into a warning. DeepL's new voice model, Aure, launched November 11, 2026, is built to close exactly that gap. It translates speech in real time while carrying the speaker's tone, emphasis and pace into the other language, in the speaker's own voice.

Aure is not a separate app. It is the model now powering DeepL Voice, the company's real-time speech translation product, and it switches on automatically wherever DeepL Voice runs. There is nothing to install and nothing to configure. The catch, for now, is coverage: Aure starts in eight languages, and the price of using it is measured in translated minutes. This article explains what Aure actually does, what it costs, and where its claims still need independent proof.

What Aure actually does

Until now, DeepL Voice built its translated voice from a static sample: it listened to the opening moments of a conversation, formed a picture of how a person sounded, and then held that picture steady. That captured who was speaking, but not how. A fixed snapshot cannot follow a change in emotion, a shift in emphasis, or delivery that contradicts the words, such as sarcasm. DeepL says so itself in its launch announcement.

Aure replaces that snapshot with continuous analysis. It reads incoming speech moment by moment, using both the translated words and the original audio to shape how each sentence is delivered. If a speaker slows down to make a point, the translation slows down. If they speed up with excitement, the translation follows. DeepL's product page puts it plainly: Aure "learns from your live speech to sound natural in every moment." Its demo lets visitors hear the same sentence delivered with excitement, reassurance, confidence, hesitation or disappointment.

The engineering behind that, as DeepL describes it, separates three things that older systems treated as one: voice identity, prosody (the rhythm and melody of speech) and spoken content. Each is carried into the target language independently, and reinforcement learning refines the result to keep translation accuracy and speech quality intact. That separation is also what lets Aure handle the messy parts of real conversation. It has to isolate the speaker's voice from background noise, echo and audio distortion, and carry a voice across languages without exaggerating accents or distorting how someone sounds.

One claim deserves a flag. On the Aure product page, DeepL says online meetings translated with its tool have "76% fewer critical translation errors than platform-native translation tools." That is DeepL's own figure, and the page publishes no methodology, test set or baseline for it. Treat it as a vendor claim, not a measured result, until an independent evaluation backs it up.

How Aure differs from September's voice preservation

Aure is the second step in a sequence DeepL started two months ago. On September 15, 2026, the company launched voice preservation: new Voice models that keep each speaker's voice recognizable across languages instead of replacing it with a generic synthetic voice. That release also made voice-to-voice translation generally available across online meetings, in-person conversations and the DeepL API, and added a desktop app for Windows and Mac.

The difference is identity versus delivery. Voice preservation answered the question of who is speaking. Aure answers the follow-up: how are they speaking right now? Preservation froze a voice; Aure tracks it as it changes through a conversation. Someone can sound tired one day and enthusiastic the next, or shift from serious to excited while sharing good news, and DeepL says Aure distinguishes those variations and adapts the translated speech to match. As The Decoder noted in its coverage of the launch, that is also why DeepL frames sarcasm as the test case: a flat translation of a sarcastic sentence says the opposite of what the speaker meant.

The September release also set the scale Aure inherits. Voice-to-voice translation is generally available in more than 30 languages, and translated subtitles run across more than 40 languages. Aure itself starts narrower, in eight languages, and DeepL Voice falls back to its previous voice models for the rest. That fallback matters for buyers: Aure is an upgrade inside a product that already covers far more languages than the new model does on day one.

Where you can use it

Aure is available in three places, and all three switch it on automatically. DeepL Voice for Online Meetings runs inside Microsoft Teams, Zoom and Google Meet through the desktop app. DeepL Voice for In-Person Conversations handles face-to-face talk, from a front desk to a factory floor. And the DeepL API for Voice lets developers embed real-time voice translation into their own apps and products. No setup is required for any of them, according to DeepL's FAQ: if the language pair is one of the eight Aure supports, Aure is what runs.

The eight launch languages are English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. More are promised, with no dates attached. For a team whose meetings run in one of those eight, Aure is live today. For everyone else, DeepL Voice still translates, but with the older models, which preserve voice identity without Aure's live delivery tracking.

DeepL also says Aure filters background noise, other speakers and echoes to isolate the primary speaker's voice, which is the feature that decides whether this works in a real office or only in a quiet demo. There is no independent test of that claim yet, and noisy environments are exactly where speech translation usually degrades first. If your meetings happen in open offices or on factory floors, the 14-day trial below is the honest way to find out.

What it costs

DeepL Voice is sold separately from DeepL's text Translator plans, and it is metered in translated minutes per user per month. The figures below are the US storefront prices on DeepL's pricing page, where Translator, API and Voice sit on separate tabs. DeepL shows prices in local currency by region, so the pound and Canadian dollar figures are approximations at November 11, 2026 mid-market rates (1 USD = 0.75 GBP, 1 USD = 1.42 CAD).

Three pieces of fine print matter. First, the annual discount on Voice plans is about 14%: $13 drops to $11, and $38 drops to $32. Second, unused minutes do not roll over to the next month, so a plan sized for your busiest month wastes minutes in quiet ones. Third, teams of 50 or more users are on custom pricing, and Voice Business is the plan that carries the admin features a company actually needs: pooled minutes, central administration and single sign-on.

If you already pay for a DeepL Translator plan, check your allowance before buying Voice separately. Every paid Translator tier includes 30 minutes of DeepL Voice per user per month at no extra charge, which is enough to test Aure in a real meeting or two, but not enough to run a team on it.

What a translated minute actually costs

Dividing the price by the allowance shows what the meter really charges. On monthly billing, Voice Starter works out to about 4.3 cents per translated minute ($13 divided by 300 minutes), and annual billing brings that to about 3.7 cents. Voice Business is about 6.3 cents a minute monthly and 5.3 cents on annual billing.

That produces an odd result worth stating plainly: Business costs more per minute than Starter. The extra $25 a month does not buy cheaper minutes. It buys pooled minutes, so a quiet team member's allowance can cover a heavy user's week, plus the admin controls, the 300-participant meetings and single sign-on. A small team that does not need those features gets more minutes per dollar on Starter. A team that does need them is paying for administration, not for translation.

For scale, a team holding 20 hours of translated meetings a month would use 1,200 minutes. That is four Starter seats at $52 a month on monthly billing, or two Business seats at $76. The Business route costs more, but its minutes are pooled, so the arithmetic only favors Starter if every seat actually uses its full 300 minutes. Most teams should size the plan on their heaviest users, not on headcount.

One comparison puts the per-minute price in context. Mistral's Voxtral 3, a voice agent API rather than a translation product, charges $0.04 a minute. Starter's 4.3 cents sits in the same band, which is a useful sanity check: DeepL is not charging a premium per minute for translation, it is charging for a meeting product wrapped around it.

Privacy: what happens to your voice

Voice cloning raises an obvious question, and DeepL answers it directly on the Aure page. Aure matches your voice dynamically during the live session. It does not create or store a permanent voice profile. Voice data is processed temporarily in memory and deleted when the session ends. Meeting transcripts and translations are not permanently stored either, data is encrypted in transit, and DeepL says none of it is used to train its models.

That is a strong privacy posture for a product that listens to business meetings. It is also entirely DeepL's own account of its behavior. No outside audit is cited on the page. The practical takeaway is narrower: unlike a voice-cloning service that keeps a copy of your voice on file, Aure has no profile to leak, steal or reuse, because it builds the match fresh in each session and discards it. If your company has data-residency or retention rules, the "deleted when the session ends" wording is the line to confirm with DeepL in writing before rolling it out, since the product page does not name retention periods or regions.

The honest gaps

Aure launched today. The honest shape of this launch is a vendor announcement with a live product behind it, not a proven result. Four gaps are worth naming.

First, no independent benchmark exists yet. DeepL's "76% fewer critical translation errors" figure has no published methodology, and no outside lab has tested Aure's tone preservation against a rival. The September release cited a Slator assessment that ranked DeepL Voice first for AI-translated captions, which is a real third-party signal, but it covers captions, not Aure's voice delivery.

Second, eight languages is a narrow launch. The pairs cover the largest European languages plus Japanese and Korean, which fits DeepL's enterprise customer base, but teams working in Hindi, Arabic, Mandarin or Portuguese-adjacent markets outside the list get the older models, not Aure. DeepL promises more languages with no dates.

Third, DeepL publishes no latency figure for Aure. Real-time translation lives or dies on delay. A two-second lag is workable, while a five-second lag breaks a negotiation. The company says Aure adapts "before they have even finished a sentence," which suggests low latency, but that is marketing language, not a number. Test it in a live meeting before committing a team.

Fourth, the sarcasm example is DeepL's framing, not a tested capability. Carrying sarcasm across languages is hard, and no demo or benchmark in the launch material proves Aure does it reliably. The demo on the product page, with emotion buttons for excitement, reassurance, confidence, hesitation and disappointment, is the closest thing to evidence, and a scripted demo is not a test.

Common mistakes when buying DeepL Voice

Buying translated minutes is easy to get wrong in a few predictable ways. The first is paying for voice when captions would do. If your meetings only need people to follow along, the live translated captions in DeepL Voice, or the caption tools already built into Teams, Zoom and Meet, cover the need without burning translated minutes. Voice translation earns its price when people need to speak and be heard, not just read.

The second is sizing by headcount instead of by minutes. A ten-person team that meets in English most of the week might use fewer translated minutes than a three-person team running daily calls with Tokyo. Count the translated hours first, then pick the plan. The 30 free minutes bundled with Translator plans are a no-cost way to measure real usage for a month.

The third is ignoring the language list. Aure's eight languages are the ones that get live tone tracking. If your key language pair is outside that list, you are buying the older voice models, and the tone-preservation pitch does not apply yet. Check the pair before the purchase, not after.

The fourth is assuming the trial covers the team. The 14-day free trial is on Voice Starter only. Business has no trial, so a team that needs pooled minutes and admin controls has to buy blind or test on Starter first and extrapolate.

Frequently asked questions

What is DeepL Aure?

Aure is DeepL's newest voice model, launched November 11, 2026. It powers real-time voice translation inside DeepL Voice, carrying a speaker's tone, emphasis and pace into the target language in their own voice, instead of translating the words alone.

How much does DeepL Voice cost?

Voice Starter costs $13 per user a month on monthly billing, or $11 on annual billing, for 300 translated minutes per user a month. Voice Business costs $38 a month, or $32 on annual billing, for 600 minutes per user a month pooled across the team. Paid Translator plans include 30 Voice minutes per user a month.

Which languages does Aure support at launch?

Eight: English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. More languages are promised with no dates given. For other languages, DeepL Voice automatically uses its previous voice models.

Does Aure store my voice or my meetings?

According to DeepL, no. Aure matches your voice live during the session without creating or storing a voice profile, and voice data is deleted from memory when the session ends. Transcripts and translations are not permanently stored, data is encrypted in transit, and DeepL says it is not used to train its models.

How is Aure different from DeepL's voice preservation?

Voice preservation, launched September 15, 2026, keeps your voice recognizable across languages. Aure goes further: it tracks how your voice changes during a conversation, so shifts in tone, pace and emphasis carry into the translation in real time.

Can I try Aure for free?

Yes, through the 14-day free trial on Voice Starter. Paid Translator plans also include 30 minutes of DeepL Voice per user a month, which is enough to test Aure in a real meeting. Voice Business has no free trial.

Sources

• DeepL — Aure product page
• DeepL via PRNewswire — "DeepL unveils Aure, its most lifelike Voice AI translation model yet," November 11, 2026
• DeepL — Pricing page (Translator, API and Voice plans)
• DeepL — DeepL Voice product page
• DeepL — "DeepL Voice now preserves your voice in real-time multilingual conversations," September 15, 2026
• The Decoder — "DeepL wants real-time translation to finally capture sarcasm and mood shifts," November 11, 2026
• UsagePricing — DeepL pricing tracker (US storefront Voice plan figures, checked November 2, 2026)
• EcomAgentTools — DeepL pricing guide (US storefront Voice plan figures)
• RateMorph — USD to GBP mid-market rate, November 11, 2026 (1 USD = 0.75 GBP)
• Exchange-Rates.org — USD to CAD rate, November 11, 2026 (1 USD = 1.42 CAD)

Related reading on AI Core Updates: PLaMo 3 Translate (free offline translation in 53 languages), Mistral Voxtral 3 (a phone-ready voice agent at $0.04 a minute), and Hume Audio Decision API.