openWakeWord

openWakeWord is a wake word detection library for developers building voice apps that listen for a chosen word or phrase. It lets an application detect voice triggers on its own hardware.

• English voice triggers: Included models recognize phrases such as "hey jarvis", "alexa", weather requests and timer commands, with some tolerance for accents, speaking speed and phrasing.
• Custom phrases: Training tools use synthetic speech to create detectors for your own wake words, reducing the need to record spoken examples.
• Speaker filters: Optional verifier models can restrict activation to known voices when false triggers are a problem, but they also make the app less likely to respond to unfamiliar speakers.

Only English models are supported. Custom training has a Google Colab option that runs in the cloud; the resulting detector can run locally. The simpler training route may produce models that perform poorly in some deployment settings. The library's code is open source under Apache 2.0.

The v0.6.0 release provides the package and pretrained wake-word models. The code uses Apache 2.0; the included pretrained wake-word models use CC BY-NC-SA 4.0, so their commercial use is restricted. See the project licensing notice.