Gboard is learning to turn your nonsense into better text

Voice-to-text transcription aims to save Android users the hassle of manually typing on their tiny virtual keyboards. A single filler word or moment of brain fog can interrupt dictation on your Android keyboard, forcing you to start over. Google last month began publicly exploring ways to improve speech recognition and dictation with Google AI Edge Eloquent, an iOS-only app that uses on-device AI models to clean up and polish written text. It seems Android users’ patience has paid off, as Gboard is getting similar speech-to-text improvements with “Rambler,” a new feature debuting as part of the Gemini Intelligence suite.

Google announced Rambler under Gemini Intelligence during the Android Show I/O edition, and it’s designed to understand your meaningless “babbling.” Since it’s integrated with the Gboard keyboard, it’ll work anywhere on Android. If you want to use Rambler to convert speech to text in a Google Messages conversation, Notion document, or Slack thread, the tool will be there. It will be available in every app that supports Gboard, making it more convenient to use than Google AI Edge Eloquent on iPhone.

Rambler tries to figure out what you’re trying to say with text instead of typing out every word exactly. It understands context, so filler words like “um” or “like” are left out of the transcription for clear reading. It also knows when you have repeated yourself or tried to improve yourself. This means that if you start dictation in Gboard and say something like, “No, actually, I’d do that instead,” Rambler will know to ignore what was written previously and focus on your correction instead.

Gboard’s Rambler feature aims to add polish to speech-to-text output. Often, achieving quality transcription requires multiple attempts or manual editing, making you wish you had typed manually first. Using an advanced multilingual Gemini model, Rambler transcribes and cleans text output in real-time, saving you the step of doing manual editing.

This Gboard Dictation mode is optional, and uses only audio to transcribe your speech to text. Then, the text is processed to add polish and correct any errors or remove unnecessary filler words. The approach ensures that your audio is never stored or saved – it’s only used for real-time transcription.

Gemini’s multilingual model helps Rambler understand and transcribe speech using multiple languages ​​in a single message. “Whether you’re mixing English with Hindi or some other combination, Rambler understands context and nuance, ensuring that your message sounds exactly like you meant it – only more sophisticated,” Google explained in a blog post.