I've been documenting an effort to automate the detection of
first/second crack (or any other audible event) using realtime audio
analysis of an ongoing roast in a thread on HB. Check it out here for
all the gory details:
I'm posting now to the list as I actually have something I think to be
worth sharing with the larger Artisan community:
This video is a recording of a full roast with a couple test versions
of my audio classification engine running in the background. The
green and purple lines in the graph indicate the classifiers'
confidence that they're "hearing" the sound of first crack. The
YouTube link above should take you straight to the beginning of that
process. The KNN classifier (in green) is pretty spot-on, both have
about 3-4 seconds of latency, and there are very few "blips" of false
positives before the actual sound of FC (and none of those positives
are above 50% confidence).
I still have a lot of work to make this thing usable for the general
public, but I do have it nailed down to the point where it's fully
operational and well integrated into Artisan, launching everything via
Alarms and sending the data back into Artisan via the External Program
feature. So far, so good!