The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← Trends

Supporting case

Google makes offline speech components available for local apps

What happened

Google announced and linked tools for speech recognition and synthesis on November 10. Android developers can call Google-managed system models through ML Kit. Developers building across platforms can download model bundles and run them with Google’s MediaPipe and LiteRT software. The downloadable Parakeet speech-recognition bundle lists a CC BY 4.0 license. Local inference can keep audio off a remote service. Supported devices and performance vary; the performance claims are Google’s.

The system route eases adoption while Google controls its updates. Downloadable bundles give developers more control over the version they ship and when they update it, although dependencies and switching costs remain. This provides implementation context about speech, distinct from local language-model experiments.

November 16 update. Google announced and linked tools for speech recognition and synthesis on November 10. Android developers can call Google-managed system models through ML Kit. Developers building across platforms can download model bundles and run them with Google’s MediaPipe and LiteRT software. The downloadable Parakeet speech-recognition bundle lists a CC BY 4.0 license. Local inference can keep audio off a remote service. Supported devices and performance vary; the performance claims are Google’s.

The system route eases adoption while Google controls its updates. Downloadable bundles give developers more control over the version they ship and when they update it, although dependencies and switching costs remain. This provides implementation context about speech, distinct from local language-model experiments.

Dates and implementation stage

2026-11-10 announcement and linked artifacts; November 16 follow-up evidence

Supporting sources

Related trends