Gboard can learn from your keyboard and dictation use to help improve Gboard for everyone. Gboard can learn through techniques known as federated learning, ephemeral learning, and conventional learning.
Learn about Gboard’s learning modelsFederated learning
A technology called federated learning helps Gboard learn new words and phrases. Federated learning doesn’t send the text you speak or type to Google, but will send what it learns to Google, where it will be combined with learnings from other users to create better speech and typing models. Gboard only uses federated learning while your phone charges, is connected to Wi-Fi, and isn't in use. Learn how federated learning works.
When you talk to Gboard, Gboard may use ephemeral learning to improve speech models and make speech recognition a better experience for everyone. As your audio is translated into text, a training algorithm may learn from your audio in real time. The audio data is never stored on a server and is deleted from short-term memory within minutes. Ephemeral learning happens automatically, and can’t be turned off. Learn how ephemeral learning protects your privacy.
When you talk to Gboard, you can also choose to donate snippets of your speech to improve speech models and make speech recognition a better experience for everyone through conventional learning. Snippets include a maximum of 15 seconds of your speech. If there’s silence or unrecognized audio, snippets may be up to 25 seconds. Google won’t keep these snippets beyond 18 months. Learn how conventional learning works.
Change federated learning settings
- On your Android device, tap Settings .
- Tap System Language & input.
- Tap On-screen keyboard Gboard Privacy.
- Choose your preferences:
- Personalize for you: Improves voice and typing recognition just for you. Audio recordings and transcripts of what you say and type will be saved on this device. You can delete them at any time on your keyboard when you tap Settings Privacy Delete learned words & data.
- Improve for everyone: Improves voice and typing recognition for everyone. Audio recordings and transcripts of what you say will be saved on this device.
- Delete learned words and data: Clears all on-device data that Gboard has saved.
Donate audio snippets to improve dictation through conventional learning
Important: Data donation is voluntary. If you use dictation, you might be prompted to share your audio snippets to help improve dictation. If you agree, snippets of your audio input on Gboard are sent to and stored by Google to improve speech recognition for everyone through conventional learning. Even if you do not donate audio snippets, Google may still learn from your audio in real-time through ephemeral learning or from your on-device data through federated learning.
Stop sharing audio snippets
You can stop sending your audio data anytime.
- On your Android device, open any app you can type with.
- At the top of your keyboard, tap Settings Privacy.
- Under “Voice,” turn Audio donations on or off.
Shared audio snippets aren’t associated with you
Since your audio snippets aren't associated with you or your account, you won’t have access to them after they’re sent to Google.
We only use audio snippets to improve speech technology