Skip to content

Latest commit

 

History

History
31 lines (25 loc) · 1.18 KB

soundscape_sensors.md

File metadata and controls

31 lines (25 loc) · 1.18 KB

Soundscape sensors

We might be interested in recording sound from a audio sensor located at a single place continuously, ie. 24/7. This might be useful eg. for gathering data for some exploratory analysis, building a training dataset for machine learning, and performing an actual ML task.

Examples of usage

  • car traffic activity
  • presence and activity of bird species
  • monitoring weather - wind, rain, ...
  • human activity in cities

Consider a small Raspberry Pi mini-computer act as the sensor.

What technical problems we would need to solve?

  • how and where to store the data?
    • raw mono PCM sampled at 44kHz take ~3.5 GB per day
  • how to deliver the data to the storage?
    • connectivity
  • how to compress the data?
    • we might assume a typical soundscape might be quite boring (ie. sparse) and could be compressed well
    • see dynamic compression
  • how to deal with outages?
    • power outage, upgrade, ...
  • how to explore the data?
    • how to find typical patterns?
    • how to find interesting anomalies?
  • how to precisely timestamp the data?
    • NTP?
  • how to combine recordings from multiple sensors?
  • how to combine the data with other modalities?