Gemini 3.5 Transcribe

Gemini 3.5 Transcribe is a Google DeepMind AI model that turns spoken audio into written text. Think of a stenographer sitting in on a meeting, listening to a recording and typing it out. Except this stenographer automatically strips out filler words like "um" and "uh," recognizes over 85 languages on its own, and when up to three voices are mixed in a recording, it labels who spoke when.

A recently revealed screen previews a feature that lets you hand this stenographer a glossary in advance. Even a great listener can easily mishear an unfamiliar company name or industry jargon and write down a more common word instead. So a custom vocabulary feature is being prepared that lets users pre-register unfamiliar words — like company or product names — so the model prioritizes recognizing them correctly.

However, since Google itself stated that the screen it shared is "a recreated example, not an actual screen," it's not yet confirmed exactly when this feature will roll out to all users, or which devices and regions will get it first.

The name shows up in coverage like "Google DeepMind shared a new screen for Gemini 3.5 Transcribe on X," when reporting on previews of upcoming features. It's worth noting, though, that the shared screenshot isn't an actual app screen but a recreated example made by Google, and this announcement alone doesn't confirm when or on which devices the feature will actually be available.

Try transcribing a meeting or interview recording in an app that supports this model, and see how frequently used company names or technical terms initially get mistranscribed. Once the custom vocabulary feature actually launches, register those words in the list and run the same recording again to compare how much the accuracy improves — that's a good way to feel the feature's real impact.