Google Launches Gemini 3.5 Transcribe for Better Voice Typing

Google has launched Gemini 3.5 Transcribe, a speech-to-text model designed to improve voice typing. It removes filler words like "um" and "uh" and fixes self-corrections, producing cleaner text. The model already powers the Rambler feature in Gboard on Pixel 11 and will expand to more Google products.

Compared to the older Chirp 3 engine, it's faster and more accurate—about 70% faster overall, with a 5.5% error rate in real-time speech, down from 7.32%. It supports 85 languages and can handle audio with up to three speakers.

The model also understands what you mean, cleaning up messy sentences. But it might change your original words, so it's not ideal for exact quotes. You can add custom words for better accuracy with specific terms.