Technology
Google introduces Gemini 3.5 Transcribe speech recognition model
The new AI audio model removes filler words, formats spoken text, and is rolling out across Google's mobile and desktop platforms.
The short version
- Google launched Gemini 3.5 Transcribe, a new AI audio model designed to turn spoken speech into formatted text while automatically filtering out disfluencies like filler words.
- The model automatically detects over 85 languages, recognizes custom jargon and alphanumeric strings, and provides speaker attribution for up to three people in pre-recorded audio.
- Gemini 3.5 Transcribe currently powers Gboard Rambler on Android devices and the Gemini app on macOS, with public previews available for developers and enterprise customers.
- Google plans to bring the speech-to-text functionality to web fields in the Chrome browser, as well as to Search Live, Gemini Live, Docs, Keep, and Gmail.
Key facts
- Google launched Gemini 3.5 Transcribe alongside updates to its Gemini Audio lineup, which includes Gemini 3.5 Live and 3.5 Live Experimental.[The Verge]
- The model automatically detects and transcribes over 85 languages and handles self-corrections, background noise, and speech interruptions.[9to5Google · Engadget · The Verge]
- Gemini 3.5 Transcribe achieved an average Word Error Rate of 4.0% for streaming and 2.6% for non-streaming use-cases in evaluations by Artificial Analysis.[9to5Google]
- The model features speaker attribution for up to three speakers with word-level timestamps on pre-recorded audio, with support for more than three speakers listed as experimental.[9to5Google · Engadget]
- Gemini 3.5 Transcribe powers Gboard Rambler on Android devices, the Gemini app for macOS, and Google Antigravity.[9to5Google · Engadget]
- Developers can access the model in public preview through Google AI Studio and Google Antigravity, while enterprise access is offered via the Gemini Enterprise Agent Platform.[9to5Google]
What remains uncertain
- The specific release timeline for bringing Gemini 3.5 Transcribe to Chrome web fields, Search Live, Docs, Keep, and Gmail has not been detailed.[9to5Google · Engadget]