Google Tests the Waters with a New Offline-First Dictation App

Google has introduced a new, experimental dictation application for iOS named "Google AI Edge Eloquent." The release was notably understated, appearing without fanfare. The app enters a field with established players like Wispr Flow and SuperWhisper, but with a distinct focus: it's designed to function primarily without an internet connection.

Once users download the app and its compact Gemma-based speech recognition models, they can dictate directly on their device. A live transcript appears as they speak. Upon pausing, the software automatically removes verbal hesitations such as "um" and "ah," delivering polished text. Users can then apply several transformations to the copy, including summarization into key points or adjustments to its length and formality.

A toggle allows for local-only processing, though a cloud mode employing Gemini models is available for enhanced text refinement. The app can personalize its vocabulary by pulling terms from a connected Gmail account and also accepts custom word entries. It maintains a searchable history of dictation sessions, tracking metrics like words per minute.

Google's App Store description positions Eloquent as a tool to convert natural speech into professional text. While currently an iOS-only test, the listing previously hinted at an Android version with system-wide keyboard integration, similar to features offered by competitors. That reference has since been removed, with a note stating an iOS keyboard is forthcoming. The company's quiet launch suggests this is a trial run. If the approach resonates, its underlying technology could inform future transcription capabilities within Google's broader ecosystem, particularly on Android.

Source: TechCrunch