Proof of concept. Capture sound from a microphone, build a reference profile of noise and/or sounds and then search for anomalies (difference from the reference).
The first phase after launch is to build a profile:
After collecting the statistics, an acoustic profile (model) is built.
In anomaly tracking mode, CPU utilization is about 1% (Intel i5).
When an anomaly is detected, a message is displayed and an mp3 file and a service mfcc file (not currently used) are written to disk.
An example of what the recordings look like in the audio editor.
