Google shipped a new dictation app for iOS that transcribes speech entirely on your phone - no internet connection required. The app uses Gemma, Google's family of AI models built to run on consumer hardware like phones and laptops rather than remote servers. That means audio never gets sent anywhere.
Most AI voice-typing tools, including Wispr Flow, route your audio to cloud servers. Cloud transcription generally produces better accuracy, but every word you dictate gets uploaded. For people dictating client calls, legal notes, medical observations, or financial conversations, local processing matters.
On-Device vs. Cloud Accuracy
Gemma models are small enough to run without a data center behind them, and transcription is a lighter task than generating full documents, so the models can handle it on a phone without obvious lag. The tradeoff is performance at the edges: accents, technical vocabulary, background noise, and fast speech tend to trip up on-device models more than cloud alternatives.
Whether this app matches Wispr Flow or Apple's built-in dictation in daily use is unproven. Google has distribution advantages - it's already on your phone, likely free, and integrates naturally with iOS apps. But distribution alone doesn't win productivity users who have already committed to a workflow.
The Missing Editing Layer
Wispr Flow built its following on more than raw transcription. It edits out filler words and false starts, cleans up sentence fragments, and works across most Mac apps through system-level integration. If Google's app is transcription-only without an AI cleanup layer, the two tools are solving different problems.
The offline capability is real. For anyone who has avoided AI dictation because audio gets sent to a server, this removes that concern. For productivity users who rely on AI-cleaned output, the question is whether Google expands the feature set.