About / FAQ

• Why would you make this?

I live near train, train honks a lot.

• What counts as a honk?

Audio is continuously classified as honk or not honk, but detections are limited to once every 4 seconds. So if a single honk lasts 20 seconds, it would be considered 5 honks.

• How does it work?

A neural network was trained on train horn sounds, and non-train horn sounds. A microphone outside listens continuously and classifies what it hears. I used a pre-trained model called yamnet, adding a few extra layers on top and freezing the weights of the lower layers. You can download the detection code on github or read a blog post about it here

• Can you detect where the train is?

At the moment, no, since I don't have a direct view of the tracks. If I could put a camera facing the tracks I could add that feature.

• Can I get real-time alerts to my phone?

For now you can follow @BywaterTrain on twitter, then turn on push notifications for that account. It's super annoying though.

• Can I download historical data?

Not currently, possibly soon. Let me know if the data is interesting to you.

• Where is the microphone?

The mic works best when the train is crossing St. Claude at Press St. It might not detect the horn when the train is further down near the river.