Google has introduced Gemini 3.5 Transcribe, a new speech recognition model that goes beyond simply transcribing audio to text and can also structure chaotic spoken language. The model supports over 85 languages, recognizes specific terms, and can remove spam.
Google has introduced Gemini 3.5 Transcribe, a new speech recognition model that goes beyond simply transcribing audio to text and can also structure chaotic spoken language. The model supports over 85 languages, recognizes specific terms, and can remove spam.
One of the next uses for Gemini 3.5 Transcribe will be voice input in the Chrome browser. Google plans to allow users to dictate text directly into any text field on a website — for example, to write messages, posts, or give voice commands to the AI.
Gemini 3.5 Transcribe is designed to handle imperfect speech. The model can recognize specific terminology and unusual spellings, as well as correctly handle numbers and combinations of letters and numbers.
When processing already recorded audio, the model is able to distinguish up to three speakers and determine a timestamp for each spoken word.
Separately, Google emphasizes the model’s ability to structure speech. That is, the user does not have to formulate thoughts perfectly while dictating — Gemini can remove parasitic words and turn the spoken stream into more organized text.
According to Google, the technology is already being used in the Rambler feature on Pixel 11 smartphones and in Gemini for macOS. The company also plans to integrate it into Search Live, Gemini Live, Google Docs, Google Keep, and Gmail. Gemini 3.5 Transcribe will be available to developers via an API.
