Back to directory

Hey Donut

A custom wake-word model for Home Assistant voice assistants, built with microWakeWord, that runs entirely on-device and comes with its full training pipeline so a user can train the same setup on any name of their own.

Hey Donut

Hey Donut is a wake-word model that teaches a Home Assistant voice satellite to respond to the phrase "Hey Donut," running entirely on-device through the microWakeWord framework with no cloud service or account involved. The model ships as two small files, an int8-quantized streaming TensorFlow Lite model and a manifest, and is installed by referencing the manifest in an ESPHome voice-satellite configuration.

A second version of the model was retrained on recordings of the phrase as people actually say it, including fast or glottal-stopped pronunciations that the first version had been trained to reject, substantially improving its detection rate at the same false-accept rate. The repository includes the complete training pipeline under a public-domain license, so it can be used as a starting point to train a wake word for a different name instead of "Donut."

More in Speech

See all 119
  • ELI v2.0

    geteli.tech

    A private AI assistant and local runtime that runs entirely on the user's own hardware: it holds conversations by voice or text, controls the desktop, reads the screen, writes code, and builds a private model of the user over time.

  • VoiceAssistant

    github.com

    A fully local, full-duplex voice assistant for Windows that lets a user interrupt it mid-sentence and answers using visual context captured from the desktop the instant the user speaks.

  • vibe

    github.com

    A cross-platform desktop application for transcribing audio and video offline using OpenAI's Whisper models on the user's own machine, with an optional OpenAI backend when a user supplies an API key.