An over engineered doorbell detector based on machine learning
Install docker:
curl -fsSL https://get.docker.com -o get-docker.sh
sudo get-docker.shInstall git, git lfs and docker compose
sudo apt install git git-lfs docker-compose
This repository is structured as follows:
raw_data: folder to keep track of original, uncut recordings for reference. It's always a good idea to keep the original data around. split_data: folder to put the manually extracted bell sounds in and the automatically extracted non-bell and non-silence parts. bell: bell sounds of varying lengths noise: ambient noise sounds of varying lengths
Quickstart on your own data:
Extract the bell sounds and put each audio file under split_data/bell
Throw the ambient sounds file in raw_data as ambient.wav, it will be split as part of the notebook
Run the notebook to generate a model file
Add a `creds.py` file under `deploy` and add an `app_token` and `client_token` variable of your pushover api keys
Copy the deploy folder (also containing the trained model file now) to the raspberry pi
You can use `screen` to run the python script until it is deployed with docker in the next version
TODO: add retraining code for corrected windows
https://github.com/veirs/sounddevice/blob/master/Dockerfile
TODO:
- [*] Get model server running
- Get dashboard running
- Get retraining pipeline running
- [*] Update model server features
- Implement model hotswap in model server