Google’s Gemini 3.5 Transcribe Promises Cleaner Voice Input—With a Catch

Google’s Gemini 3.5 Transcribe Promises Cleaner Voice Input—With a Catch
Google is betting that voice input becomes far more useful once software stops faithfully reproducing every hesitation. But Gemini 3.5 Transcribe’s polish comes with a trade-off: the model may alter not just the clutter in a sentence, but its wording.

The rollout began with Google positioning Gemini 3.5 Transcribe as a successor to its Chirp 3 speech-to-text engine. The company says the model takes voice from speech to final text about 70% faster, while cutting its live-speech error rate from 7.32% to 5.5%. It is designed to remove “ums” and corrections, recognize custom vocabulary, work across 85 languages and distinguish as many as three speakers in recorded audio.

Google’s pitch is not simply transcription but interpretation. On X, CEO Sundar Pichai said developers can build apps that understand a user’s “speech / intent,” including in multi-speaker settings, with language detection and jargon adaptation available out of the box.

That ambition explains both the appeal and the risk. For quick dictation, cleaning verbal stumbles could spare users tedious typo fixes. Yet the model “does technically change the wording of what you said,” an uncomfortable prospect for sensitive notes, quotes or records where precision matters as much as readability.

Availability is arriving in stages. It already powers the Rambler feature in Gboard on Pixel 11 devices, while macOS users of the Gemini app are receiving it in English. Developers can access it through the Gemini API, AI Studio and Antigravity; Chrome integration is promised “soon.”

A separate report clarified the boundaries of the announcement after Google initially supplied information suggesting Gemini 3.5 Live and 3.5 Live Experimental would launch alongside Transcribe. Google later said only Transcribe was being announced and offered no new date for the other models. For now, the message is straightforward: cleaner speech-to-text is here; the broader Gemini Audio upgrade is not.

Continue reading https://foxvector.com/stories/01a03fc3-b53d-39f3-73f5-35276edd233b