Google unveils Gemini 3.5 Transcribe for cleaner voice input

Google’s new Gemini 3.5 Transcribe marks a step forward in voice input technology, promising significant speed gains and modest accuracy improvements over its predecessor, Chirp 3. The model’s ability to clean up speech by removing verbal stumbles and correcting on the fly positions it as a tool for producing polished text directly from voice, rather than just transcribing exactly what was said. This capability, while convenient for casual use, raises questions about fidelity in contexts where exact wording matters, as the AI technically changes the user’s original phrasing.

The model’s integration into Gboard’s Rambler feature on Pixel 11 and its upcoming availability across Google’s ecosystem suggest a broad rollout. With support for 85 languages and up to three speakers in pre-recorded audio, it aims to handle a wide range of voice input scenarios. As with any AI that interprets intent, users will need to weigh the benefits of faster, cleaner output against the potential for altered transcripts in professional or legal settings.