How Google Assistant's architecture powers voice features in your apps
summary
Overview of how Google Assistant maps voice to app actions: speech is streamed to backend with speech biasing to improve transcription; NLU extracts intents and slots, supports multi‑turn dialogue and followups; Android apps plug in via App Actions with capability tags and shortcuts.xml (static shortcuts, inline inventory) so Assistant emits Android intents. Platform supports 66 Built‑in Intents, brandless queries, app install suggestions, and expands triggers for Auto and WearOS.
AI-written summary · about
chapters (9)
- 0:00Introduction to Assistant
- 1:47Vision and Capabilities
- 3:42Technical Architecture
- 5:25Speech and Understanding
- 7:47Context and Dialogue
- 10:47App Actions Framework
- 13:21Implementation Details
- 16:38New Platform Features
- 18:16Recap and Resources
related talks
20mBuilding Voice Apps for the Google Assistant
15mDeveloping for Assistant across devices
27mSession 1: Connect your Android app to Google Assistant
16mHow to voicify your Android app
13mKeynote Interview with Google's Rebecca Nathenson
27mAndroid & Assistant: a match made in heaven
24mDemystifying the Google Assistant for Android developers
27mGoogle Assistant for Android developers
