We launched livekit-wakeword, an open-source library that lets you train a custom wake word model from scratch with a single command. It handles synthetic data generation, augmentation, training, and ONNX export all in one shot.

Neat! I want to train a micro wake words for these tiny pocket robots 🐣

livekit-wakeword delivers 100x fewer false positives per hour, 60x lower detection error, and 86% recall vs 69% recall for openWakeWord. It’s also fully compatible so you can drop it into Home Assistant or any existing integration.

Repo:

Blog:

the idea of handling everything from synthetic data generation to onnx export in one shot sounds incredibly streamlined, how does livekit-wakeword handle the potential variability in audio quality during the training process

Awesome

Neat

The performance is on English phrases only or you can train in other languages too?

Synthetic data generation to ONNX export in one command. Wake word training used to require specialized ML expertise - now it's a CLI call.

Custom wake words in one command. The barrier between 'I wish my device responded to my name' and actually doing it just collapsed to near zero.

Very innovative stuff, guys.

Amazing

Boys you are nailing it 🫡✊