Paul Repacholi
(1953—2007)
Went to the funeral today. I haven't been to a funeral since my grandfather died. I still find the monotonic group incantation of passages of the Bible to be very, very creepy. Saw a number of the older-guard UCCans.

Prep always had the best stories about the early days of computing. Sometimes you didn't know if he was pulling your leg or was completely serious. One of my favourites was about why filesystems in UNIX were case sensitive. In VMS, filesystems are not case sensitive, a tradition which followed on into MS-DOS, Windows and other operating systems. As Prep told it, the reason that UNIX broke with this trend was that, unlike VMS, UNIX had not yet implemented memory paging and thus did not want to give away precious memory to implementing case normalisation. RIP, Prep. The UCC will poorer without you, though richer for having known you.

More thesis graphs. I've added some more features, gain normalisation and corrected some bugs. After adding some code to reject frames with insufficient RMS signal power (I probably should use zero crossing rate instead) I ran my routines over all of the data I've collected to extract the minimum, maximum, mean and standard deviation for each instrument/feature pair.


Spectral Features: markers are min/max/mean/±1 standard deviation
There are some interesting patterns here, e.g. between the saxophones and between all three woodwinds. I'm hoping to see patterns continue to emerge once I add trumpet and clarinet to the mix.

In my proposal, I have proposed 20 features to use in the classifier, a number of these are perceptual features. This means I need to filter the signal as it would be filtered by our ear and apply logarithmic band conversion. Hopefully I can get onto this in the next couple of days.