AI Can Translate Speech Now. Interpreting It Is a Different Job

Everyone Shipped Live Translation This Year. None of Them Shipped Interpretation

In the space of a few months, OpenAI, Google, Zoom, and Alibaba each shipped a version of the same feature: live AI speech translation. OpenAI followed its real-time translation model with GPT-Live, a conversational voice model built into ChatGPT. Google expanded Gemini into live speech translation across its product line. Zoom rolled out its own in-house version, in beta, with acknowledged limits. Alibaba’s Qwen team tripled its speech translation coverage from 18 languages to 60. 

Four major platforms, four launches, one quarter. Translation is becoming a default feature, not a specialized service. That’s worth sitting with, because it changes what “language solutions” means as a category — and it changes what’s actually left to compete on. 

Translation and interpretation are not the same job 

Translation converts words. Interpretation manages meaning in motion: a negotiation with three parties talking over each other, a technical briefing where the wrong register undermines the speaker’s authority, a legal deposition where a mistranslated qualifier changes the record. None of that is a speed problem. It’s a judgment problem. 

Slator’s analyst desk ran a piece this month cataloguing where current AI interpreting systems fall short — the kind of thing that shows up under real meeting conditions rather than clean demo audio: overlapping speech, mid-sentence self-correction, register shifts between a formal opening and an informal aside, tracking who said what across a multiparty exchange. These aren’t edge cases in enterprise and government settings. They’re the baseline. 

That’s the gap the current wave of live-translation launches doesn’t close. Fast, accurate word conversion is now table stakes, built into whatever conferencing tool a team already has open. What determines whether a high-stakes multilingual meeting actually goes well is still the interpretation layer on top of it. 

Institutions are already living this tension 

NATO posted three Brussels roles this month to lead what it’s calling an AI-enabled transition of its linguistic services — a Deputy Head of Linguistic Services, a Translation Project Manager, a Computational Linguist. All three postings also carry a line disqualifying any application prepared using AI writing or translation tools. 

That’s not a contradiction so much as a preview of how serious institutions are actually approaching this: AI is welcome to do the job it’s ready for, but the moment stakes rise, accountability defaults back to a qualified person. Not blended together — one or the other, deployed deliberately depending on what’s on the line. 

What this means in practice 

The right question for enterprise and government buyers isn’t “AI or human, permanently.” It’s “which one does this specific meeting need.” A weekly internal sync across five markets doesn’t carry the same risk as a regulatory hearing or a cross-border negotiation — so it shouldn’t require the same mode of interpretation. 

That’s the choice KUDO builds for: human interpreters or AI interpretation, selected per meeting, not merged into a single system that tries to be both at once. Same platform, same 200+ languages, and the buyer decides which mode fits the occasion — high-touch human interpretation when the outcome depends on judgment and nuance, AI interpretation when the need is recurrence or scale, and the stakes allow for it. 

Speech translation getting commoditized is a good thing. It means the conversation with enterprise and government buyers can finally move past “can AI translate this” and into the more useful one: which meetings need a person in the loop, and which don’t — and building a setup flexible enough to answer that question meeting by meeting, rather than locking in one mode for everything. 

How is your organization navigating the shift toward hybrid language strategies?

Explore how KUDO gives enterprise and government leaders the flexibility to deploy elite human interpreters or enterprise-grade AI interpretation—meeting by meeting, on demand.