Whilst I am working on re-designing the PCB, to use the correct esp32-d0wd footprint, I am also going to investigate making use of TensorFlow via Python, to recognise different acoustic events in a house.
To do this I will record various sounds in a house using a microphone and a laptop, I will then attempt to classify these events using TensorFlow. I was planning on looking into using a convolutional neural network to achieve this.
Some of the acoustic events I will attempt to detect, include:
- Door opening
- Tap running
- Light / mains switch being turned on/off
- Smoke alarm
- Gas hobs on oven being turned on
- Phone ringing
- Microwave oven / standard oven beep
- Kettle finished boiling
- Fridge turning on
- Toaster popping
I am also curious if an end user could train the system to detect new events by allowing the classifier to be updated.
I've just started evaluating the classifier from https://github.com/drscotthawley/panotti/ using sound files for smoke alarms etc. from - https://freesound.org/browse/ . It makes use of TensorFlow for creating the model.