Custom wake-word detector runs at 5.3% CPU on a $15 Raspberry Pi Zero 2 W
A developer has built a lightweight, always-on wake-word detector that runs at just 5.3% CPU usage on a single core of a Raspberry Pi Zero 2 W, a $15 quad-core ARM device. The system sits between a basic Voice Activity Detector and a full Automatic Speech Recognition engine, requiring only a small model under 100 KB while delivering sub-100ms response times. Unlike existing options such as Picovoice, which recently ended its free tier, or openWakeWord, which demands significant training effort, this solution aims to be simple to install and free to use with a default wake phrase. Benchmarks were recorded over a 60-second sustained live-microphone test, with performance on other Pi models estimated by scaling for clock speed and architecture. The project targets hobbyists who want a vendor-independent, offline-capable voice trigger without per-activation costs or cloud dependency.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in