I need an Android library that can sit quietly in the background, keep the microphone open, and call back into the host app the moment a predefined keyword is spoken. The library will be embedded in an internal enterprise application, so footprint, privacy, and battery impact come first; basic accuracy is more than enough for our use-case. I am fine with any implementation approach. The API should feel familiar to Android Studio users, expose simple start/stop methods, and deliver a clean listener callback for the wake event. Deliverables • Kotlin/Java API: init(context, "keyword"), start(), stop(), setListener(…) • Background service that survives activity/configuration changes and Doze • Sample app showing permission handling and wake-word callback • Gradle project with full source and build script • README covering integration steps and any model-training instructions Acceptance criteria - Detects the keyword within ~700 ms when spoken at normal volume - CPU use while idle stays under 5 % on a mid-range device (e.g., Pixel 4a) - No network access after initialisation - Operates on Android 8.0 (API 26) and above If you can meet these points while keeping the library lean and enterprise-ready, I’d love to review your approach and timeline.