Updated
Updated · The Verge · Aug 26
Google Launches Gemini 3.5 Transcribe in 85+ Languages, Cutting Filler Words and Errors
Updated
Updated · The Verge · Aug 26

Google Launches Gemini 3.5 Transcribe in 85+ Languages, Cutting Filler Words and Errors

3 articles · Updated · The Verge · Aug 26

Summary

  • Gemini 3.5 Transcribe debuted as a new Google audio model that removes spoken fillers like “um” and “uh,” formats text automatically, and supports transcription in more than 85 languages.
  • Google said the model improves on Chirp 3 with better multilingual accuracy and lower wording error rates, while also adapting to custom vocabulary and specialized jargon.
  • The release also adds Gemini 3.5 Live and 3.5 Live Experimental, which aim to handle background noise, mid-sentence interruptions, language recognition and more complex real-time reasoning.
  • Up to 3 speakers can be attributed in pre-recorded audio with word-level timestamps, and the update is rolling out first in English on the macOS Gemini app and Android’s Rambler dictation in select markets.
  • Developers can access the models in public preview through the Gemini API, AI Studio and Antigravity, as Google still has not released the Gemini 3.5 Pro model it had promised for June.

Insights

As Google rolls out advanced audio models, why is the flagship Gemini 3.5 Pro still missing months after its scheduled release?
Can Gemini's new three-speaker limit truly compete with open-source tools like WhisperX in complex, real-world meeting environments?
With Gemini 3.5 automatically removing filler words, are we losing the natural human nuances of everyday conversation?