About This Project

This project provides a custom Hey Yossi wake-word model designed for the microWakeWord library. It is an int8-quantized TFLite model weighing 61 KB, specifically engineered for low-latency, on-device audio classification. The model enables voice-activated triggers for applications such as the Ava Android satellite app and ESPHome-based voice assistants.

The model was trained using a MixedNet architecture on a dataset of real-world recordings. It supports integration with Home Assistant voice pipelines, providing a local, private alternative for keyword spotting. This solution is intended for developers and hobbyists looking to implement custom voice activation on edge devices with limited processing power and memory.

Project Details

Tags

microwakewordtflitewake-wordkeyword-spottingesphomeaudio-classificationdataset:danielrosehill/wake-wordslicense:cc-by-nc-4.0region:us

Explore More Projects