Google launches Eloquent iOS app for offline AI dictation

Google released Eloquent for iOS on Monday, an offline-first dictation app that transcribes speech on the device using Gemma-based automatic speech recognition models.

The app is free to download. Users must download the Gemma ASR models to start dictating. During a session the app shows live transcription, and when a user pauses Eloquent filters out filler words and stumbles and polishes the text.

Below each transcript users can apply transformation options labeled ‘Key points,’ ‘Formal,’ ‘Short,’ and ‘Long’ to reshape the output for different uses.

Eloquent provides a local-only processing mode that disables cloud services and keeps speech recognition on the phone. When cloud mode is enabled, Google’s Gemini models run in the cloud to perform additional text cleanup.

Users can import names, keywords and specialized terms from their Gmail account if they grant permission. The app also accepts custom words to improve recognition of jargon and niche vocabulary.

Eloquent saves a history of transcription sessions, includes a search function across past entries and displays session metrics such as total words and words per minute.

The App Store description states that the app edits out ‘ums,’ ‘uhs,’ and mid-sentence self-corrections to produce cleaner prose.

The App Store listing references an Android version and outlines Android-specific features, including the option to set Eloquent as the default keyboard and a floating button for quick transcription. Google lists the app as experimental and has not provided a timeline for an Android release.

Eloquent joins other AI-powered transcription apps, including Wispr Flow, SuperWhisper and Willow. The app’s offline processing keeps recognition local and can reduce the amount of data sent to cloud services.