hermes-agent/tools/wakewords/README.md
ethernet 712734436e fix(pm): make bootstrap and bundle ownership explicit
Finish bootstrap uv before PM replaces its store entry. Keep failure
receipts stdlib-only and align the cryptography requirement and override
with the locked version.

Let bundle builders declare launch paths and update ownership. Remove
payload discovery, Store probing, and the unused develop command.
Derive Nix Python from the PM lock and share its provenance stamp.

Document setup, activation, optional dependencies, and distribution
ownership. Targeted Windows tests, relocated runtime launches, Electron
bundling, and bilingual docs builds pass. Native Nix and signed-package
acceptance remain CI gates.
2026-09-08 00:24:51 -04:00

23 lines
1.3 KiB
Markdown

# Bundled wake-word models
`hey_hermes.tflite` — the on-device "Hey Hermes" hotword model. This is the
default detector for the wake word feature (see
`website/docs/user-guide/features/wake-word.md`); no training or setup is
required to say "hey hermes".
- **Engine:** [pyopen-wakeword](https://github.com/rhasspy/pyopen-wakeword)
(rhasspy's maintained fork of openWakeWord; Apache-2.0). Runs TFLite via a
bundled `tensorflowlite_c` library — no onnx, no runtime download.
- **Provenance:** trained with the openWakeWord training pipeline (synthetic
TTS-generated speech), which produces the `.tflite` artifact. Redistribution
is permitted under the openWakeWord license.
- **Label:** the model registers as `hey_hermes` (matches the filename).
- **Runtime:** the `pyopen-wakeword` wheel includes the shared
melspectrogram and embedding models. Starting this engine requires no
model download. Identical model files alone do not establish identical
scores across inference engines or platforms.
To use a different phrase, point `wake_word.openwakeword.model` at an
absolute path to a compatible `.tflite` model. Hermes does not download
models by name, and this engine does not load `.onnx` files. See the
wake-word docs for the training guide and platform limits.