Matching human impressions to animal sound recordings

[Yan Wong <>]

Hello,

I’m (very) new to this field, so apologies if this question has a well-known 
answer. I’m interested in trying to match human impressions of animal noises to 
recordings of the animals themselves, in an automated way. This seems to fall 
halfway between normal bioacoustic matching algorithms and human speech 
recognition. For example, I imagine that most birdsong algorithms take account 
of absolute pitch, whereas a human impression will clearly be of a radically 
different pitch.

Visual inspection of spectrograms of animal sounds and my impressions of them 
implies that this should be doable (I’ve put up some pictures at 
http://yanwong.me/?p=873). I was wondering whether to take a generic approach, 
using some sort of measure of joint entropy of the human-impression spectrogram 
and the real-animal spectrogram. Does anyone have any pointers as to software 
libraries that might be useful to me? Most of the pointers to online software 
seem to be to interactive programs, whereas I’m looking for libraries of 
statistical sound-analysis algorithms that I can modify, for example written 
for R, Matlab, or whatever.

Thanks

Yan Wong
Oxford