First let it be known that I know absolutely nothing about processing audio files - I’m looking for a software solution for a specific task & am checking the Audacity community for suggestions. If this question’s been answered - sorry - I did look but haven’t found it.
Audio channels on digital security camera files … the cameras run & run & create audio on one channel per camera as well as video. In my case it’s specific audio events like specific car horns that will alert to nefarious activity. What my attorney is telling me is that he needs to be able to scan audio files for specific frequency signatures & append a log file with times & dates - then manually check tagged times & dates for the expected activity associated with the known audio signature.
I’ve talked to digital security vendors & they’re telling me that the technology just isn’t there yet - which seems unlikely in a world where known miscreants don’t dare make a voice call on any phone.
So my question - what do audiophiles think? Is this do-able? We are looking for a more sophisticated trigger than a mere decibel spike that could be anything unrelated to the activity of interest.
MATLAB (or one of the free MATLAB cones) can do [u]FFT[/u] analysis (technically DFT with digital audio) and then you’re free to do whatever you want with the “numbers”.
But… One car horn is going to sound different (and have a different signature) from another car horn, the distance will make a difference, and the sound bouncing off a wall or building will make a difference, and I just don’t know if you’d get anything useful.
And… You could probably accomplish a lot “manually” in the time it would take to design & program such a system.
The objective is to reduce potentially endless hours of review & monitoring.
So in the car horn example - a known sampling of a particular horn that pulls up & goes off in a particular location among buildings that don’t get up & wander around - that horn won’t produce a signature unique enough to search for & find in past & future files. If that can’t be done then unique voices certainly can’t be identified as signatures to search for & locate in files that accumulate to 24 hours in a day, 168 hours in a week, 730 hours in a month & so on per camera.
I’m not even sure I care if the scan confuses some fool & his ball bat beating on a steel door with barking dogs. More precision would be nice but dogs usually bark for a reason - that’s worth looking at. Surely something out there can note the difference between that & the neighbor’s spooky-quiet Chevy Volt.
Will I really have to turn this into somebody’s McJob?
“Sound Finder” can pick out “loud” (high amplitude) sounds from a quiet background Audacity Manual
You can limit the frequency range by filtering the audio before running Sound Finder. For example, if you are only interested in sounds that are in to 1000 to 2000 Hz range, then you can make detection more accurate by filtering out frequencies below 1000 Hz (High Pass filter) and above 2000 Hz (Low Pass filter) before scanning with Sound Finder.
If you need to pick out specific low level sounds that are buried within a loud noisy background, then that is very much more difficult and would require complex pattern matching that is beyond the scope of any off the shelf software.