Google rolls out AI transcription that auto-removes filler words and handles 85+ languages

Google has expanded its Gemini Audio capabilities with three new AI models designed to improve how the platform handles voice input. The updates introduce Gemini 3.5 Transcribe, a dedicated transcription service, alongside enhanced versions of its live conversation tools.

What's New#

The standout feature is Gemini 3.5 Transcribe, which processes audio with a focus on cleaning up natural speech patterns. The model automatically strips out filler words such as "um" and "uh," formats text intelligently, and can recognize specialized terminology across more than 85 languages. It also marks timestamps at the word level and can attribute speech to up to three different speakers in pre-recorded files.

Users gain voice-controlled editing capabilities—Google says you can refine transcriptions using only spoken commands. The system accepts custom vocabulary lists, allowing it to learn domain-specific jargon and unusual spellings without requiring manual fixes afterward.

Alongside Live Updates#

Gemini 3.5 Live and Gemini 3.5 Live Experimental round out the rollout, building on Google's existing voice chat mode. The former handles mid-sentence interruptions more smoothly and improved language detection, while the experimental variant narrates its reasoning process in real time as it tackles complex problems.

Rollout and Availability#

These features are launching today in English for macOS Gemini app users and on Android via the Rambler dictation feature in select regions. Developers can access them now through public preview on the Gemini API, AI Studio, and Antigravity. Chrome support is coming soon.

The update arrives as Google continues to delay the long-promised Gemini 3.5 Pro, which was originally scheduled for June but remains unreleased.