How Do I Train a Custom Wake Word for Windows?

Train an offline keyword model online and use it in a Windows desktop or local automation application.

Training steps

1. Enter one or more phrases and choose a language.
2. Train, test, and optionally upload recordings you are authorized to use.
3. Export ONNX or TFLite for local inference.
4. Connect keyword labels to your Windows application or automation workflow.

One model, multiple keywords

Voicute supports a multi-keyword model that recognizes several fixed phrases and returns the matching label. This is useful for desktop commands such as next, mute, and stop.

Windows deployment notes

Use ONNX Runtime or TFLite Runtime with the Windows audio stack, and verify microphone permissions, WASAPI devices, sample rate, CPU load, and startup behavior. Thresholds should be measured with the final microphone and room noise.

Supported languages

English, Chinese, Spanish, Japanese, German, French, and Hindi are supported.

Train a custom wake word online →