Google’s AI transcription tool eliminates filler words like ‘ums’ and ‘ahs’ for cleaner audio.

Google has enhanced Gemini Audio with new transcription features that can automatically recognize specialized terminology and support over 85 languages. The introduction of Gemini 3.5 Transcribe follows the earlier release of 3.5 Live Translate, while anticipation builds for the upcoming Gemini 3.5 Pro model, which was expected in June.

According to Google, 3.5 Transcribe marks a significant leap forward from the previous transcription model, Chirp 3, particularly in terms of multilingual capabilities and reducing wording errors. This new model allows users to “edit naturally with just your voice” and can automatically format text, eliminating filler words like “um” and “uh.”

Users can also input a custom vocabulary, enabling 3.5 Transcribe to adjust its transcriptions to accommodate unique spellings and specialized jargon, minimizing the need for manual edits. Additionally, it can identify speech from up to three speakers in pre-recorded audio and provide word-level timestamps.

Starting today, Gemini 3.5 Transcribe is being rolled out in English for all macOS users of the Gemini app, as well as the Rambler dictation feature on Android in select regions and languages. Developers can access it in public preview through the Gemini API via AI Studio and Antigravity, with Chrome support expected to arrive soon.

Correction, August 26th: Google initially mentioned two other Gemini Audio models, 3.5 Live and 3.5 Experimental, in information shared before publication, but has now clarified that only 3.5 Transcribe is being announced today.