bmo-openwakeword / README.md
amrmantawi's picture
Document all three ONNX files
58eb940 verified
|
Raw
History Blame Contribute Delete
1.21 kB
---
license: apache-2.0
tags:
- openWakeWord
- wake-word
- audio
- onnx
library_name: openwakeword
pipeline_tag: audio-classification
---
# BMO openWakeWord models
Custom [openWakeWord](https://github.com/dscripka/openWakeWord) pack for the wake phrase **“Hey BMO”** (trained as `hey beemo`).
## Files
| File | Description |
|------|-------------|
| `hey_beemo.onnx` | Wake-word classifier (“Hey BMO”) |
| `melspectrogram.onnx` | Shared openWakeWord mel feature model |
| `embedding_model.onnx` | Shared openWakeWord speech embedding model |
All three are required at runtime.
Shared feature models come from [openWakeWord-cpp](https://github.com/rhasspy/openWakeWord-cpp) / [openWakeWord](https://github.com/dscripka/openWakeWord).
## Download
```bash
hf download amrmantawi/bmo-openwakeword --local-dir models/ww
```
## local-llm config
```json
"openwakeword": {
"melspectrogram": { "path": "../models/ww/melspectrogram.onnx" },
"embedding": { "path": "../models/ww/embedding_model.onnx" },
"model": { "path": "../models/ww/hey_beemo.onnx" }
}
```
## Training
`hey_beemo.onnx` was trained with the openWakeWord Google Colab notebook on synthetic speech for **hey beemo**.