Google launches Gemini 3.5 Transcribe for more accurate, context-aware speech-to-text

# What Gemini 3.5 Transcribe is and where you'll see it

Google introduced Gemini 3.5 Transcribe as its latest speech-to-text model. It replaces the previous transcription engine, Chirp 3, and is designed to deliver cleaner, more accurate transcripts across live and pre-recorded audio. Google says the model is rolling out broadly: Search Live, Gemini Live, Docs, Keep, Gmail, the Gemini app, and Gboard will get the update. Developers will access it through the Gemini API in Google AI Studio and Antigravity, and enterprise customers via Gemini Enterprise offerings.

# Practical improvements you can expect

Gemini 3.5 Transcribe focuses on cleaner output and better context handling. Concrete features announced by Google include:

• Automatic removal of filler words like "um" and "ah" so transcripts read more smoothly.
• Auto-formatting and voice-driven editing so spoken corrections and edits can produce natural text with less manual cleanup.
• Automatic language detection for more than 85 languages, plus improved handling of regional accents and dialects.
• Custom vocabulary support to recognize specialized jargon and unconventional spellings you supply.

• Function calling that lets the transcription model hand off tasks (for example, image generation) to other Gemini models.

# Accuracy and performance details

Google reported a 4% word-error-rate (WER) on streaming audio and 2.6% WER on pre-recorded files across diverse real-world conditions that include background noise and conversational AI interactions. Those values indicate lower error rates than the previous model according to Google's testing and are meant to reduce the time users spend correcting voice-to-text output.

# How Google is integrating the model

Some products already show the model in action. The Rambler feature in Gboard uses this transcription capability in select countries and languages. Google also plans Chrome integration so you can "talk to type" in any web field. Beyond consumer apps, the model will be made available to developers through Google AI Studio and Antigravity APIs, and to enterprise customers via Gemini Enterprise Agent Platform and Gemini Enterprise for Customer Experience.

# What this means for different users

• Casual users: Expect cleaner voice typing in Gmail, Docs, Keep, and the Gemini app with fewer filler words to edit out.
• Professionals and teams: Custom vocabulary and multi-speaker timestamps aim to make meeting notes and interviews more accurate and ready to share.

• Enterprises: The model will be included in Gemini Enterprise products intended for customer experience and agent workflows.

# Short summary

Gemini 3.5 Transcribe targets faster, more accurate and more usable transcripts by removing filler words, recognizing specialized vocabulary, handling many languages and accents, and attributing speakers in recordings. Google is rolling it out across consumer apps, developer tools, and enterprise products.