Sonavi is a sound detection system that alerts Deaf and Hard of Hearing users to environmental sounds through customizable smartwatch vibrations. We harness the power of on-device machine learning to detect and classify sounds captured by a Wear OS smartwatch, process them on an Android mobile device, and deliver haptic feedback—all without requiring an internet connection. We implemented the YAMNet audio classification model using LiteRT (formerly TensorFlow Lite) for real-time sound detection. The system allows users to detect pre-trained sounds as well as create custom sound profiles by recording or uploading their own audio samples, making it highly personalized and adaptable to individual needs. ⭐ Features 🎧 Real-time Sound Detection: Captures audio from your Wear OS smartwatch and processes it instantly on your Android phone 🤖 On-Device Machine Learning: Uses YAMNet model via LiteRT (TensorFlow Lite) for accurate sound classification 📳 Customizable Vibration Patterns: Set unique vibration alerts for different sound types 🎵 Custom Sound Training: Create personalized sound profiles by recording or uploading 3+ samples of specific sounds you want to detect 🔒 Privacy-First: All processing happens on-device; no data leaves your phone ⚡ Low Latency: Optimized communication between watch and phone for quick notifications fonte; https://github.com/xyugen/sonavi