openWakeWord

0.6.0 verified Fri May 01 auth: no python

An open-source audio wake word (or phrase) detection framework with a focus on performance and simplicity. Current version is 0.6.0. Releases are periodic with enhancements and bug fixes.

pip install openwakeword

Common errors

error ModuleNotFoundError: No module named 'openwakeword' ↓

cause openwakeword is not installed.

fix

Run pip install openwakeword.

error ValueError: Audio frames must be a list of numpy arrays ↓

cause Input to predict() is not a list of numpy arrays.

fix

Wrap your audio chunk in a list: predict([audio_array]).

error FileNotFoundError: [Errno 2] No such file or directory: '.../models/alexa.onnx' ↓

cause Model files are missing (not downloaded). Since v0.6.0, models are no longer bundled.

fix

Run from openwakeword.utils import download_models; download_models() or use the CLI command.

Warnings

breaking In v0.6.0, model files were removed from the package. You must download them separately using `openwakeword.utils.download_models()` or use the CLI. ↓

fix Run `openwakeword.utils.download_models()` or `python -m openwakeword.utils.download_models` before inference.

gotcha The default inference framework on Linux is tflite since v0.5.0. This may not be installed by default; ensure you have the necessary dependencies. ↓

fix Install tensorflow or onnxruntime as needed, and specify framework='onnx' if you prefer ONNX.

gotcha Input audio must be a list of 1D numpy arrays (dtype=float32) normalized to [-1, 1]. Passing raw integer PCM data will produce incorrect results. ↓

fix Convert audio to float32 and normalize before passing to predict.

Imports

OpenWakeWord
wrong
```
import openwakeword
```
correct
```
from openwakeword import OpenWakeWord
```
Direct import of module gives access to submodules, but the main class is OpenWakeWord.
Model
```
from openwakeword.model import Model
```
Model class is used for loading and running wake word models.

Quickstart

Initialize the wake word detector and run prediction on audio chunks.

from openwakeword import OpenWakeWord

oww = OpenWakeWord()
audios = []  # list of audio chunks (numpy arrays)
predictions = oww.predict(audios)