Google launches UI dictation app that works without the internet

It is a response to increasingly popular speech-to-text UI solutions such as Wispr Flow, SuperWhisper, and Willow.

Works offline too

The main feature of the application is that it works locally on the device. Once the user downloads speech recognition models (based on Gemma), they can dictate without a connection to the cloud. The application also offers the option to use Gemini cloud models for additional text processing.

Eloquent doesn't transcribe speech verbatim, but also improves it on the fly:

• removes fillers like “um” and “mm”,
• correct the sentence structure,
• creates more fluent and professional text.

Users can further customize the text with commands such as:

• key points,
• formal record,
• shorter or longer version.

The app also lets you import terms and names from Gmail, add your own words, view your dictation history, and search through past recordings. It also displays statistics such as speech rate and word count.

What does this mean for the market?

Dictation UI apps are becoming increasingly popular as speech recognition models advance rapidly. Google is clearly showing the direction of development with this app. It's somehow moving towards more features directly on the device, less reliance on the cloud, and broader integration into operating systems. If the app is successful, we can expect similar features for Android.