We all ramble, backtrack, and stumble over our words when we speak out loud. Google's new Gemini 3.5 Transcribe is built keeping our imperfect speech in mind. Announced as the company's most precise speech-to-text model yet, it takes your messy, unpolished speech and turns it into clean, formatted text without expecting you to sound like a news anchor first.
The model automatically strips out filler words like "um" and "ah," so your transcribed text doesn't carry every verbal hiccup. It catches self-corrections in real time - if you say "let's meet Tuesday, no wait, Wednesday," it understands you meant Wednesday and adjusts. You can also edit your text using nothing but your voice. The model can delegate tasks like image generation or file analysis to other Gemini models through function calling, a capability currently live in the Gemini app on macOS.
Google reports a word error rate of just 4% for real-time streaming and 2.6% for pre-recorded audio, with support for 85+ languages, regional accents, and specialized jargon. It can even attribute speech to up to three speakers with timestamps. The model already powers Rambler, the dictation feature on Android's Gboard, along with the Gemini app on macOS. Chrome support is coming soon, letting you dictate directly into any web field. Developers can access it now in Google AI Studio and Antigravity. Enterprise users access it through the Gemini Enterprise Agent Platform. It's also expanding into Search Live, Gemini Live, Docs, Keep, and Gmail.
For ordinary users, Gemini 3.5 Transcribe means voice input no longer punishes imperfection. The most annoying part of speech-to-text was seeing filler words preserved verbatim. Now AI understands what you actually meant. The next time you're rambling into your phone, Gemini might just understand you better than you understand yourself.