Google has quietly launched AI Edge Eloquent, a new offline-first dictation app for iOS designed to convert speech into structured, polished text using on-device artificial intelligence.
The app leverages Gemma-based automatic speech recognition models, allowing users to transcribe speech locally without requiring a constant internet connection. Once models are downloaded, users can dictate in real time, with the system automatically removing filler words and refining sentences into clearer written output.
Eloquent includes tools to transform transcripts into different formats, such as summaries, formal text, or shorter versions. It also supports keyword customization, enabling users to import names and terminology or define custom vocabulary for improved accuracy.
While the app can operate entirely on-device, users have the option to enable cloud processing, which integrates Gemini models for enhanced text refinement. Additional features include searchable transcription history, speech metrics such as words-per-minute, and session tracking.
The launch positions Google within a growing segment of AI-powered productivity tools focused on speech-to-text workflows. Competitors in this space are leveraging advances in large language models to improve transcription quality and usability.
