Abstract:
The focus of this project is sound identification of birds, more specifically
loons, through their calls or songs. There are many methods of sound identification
that exist today, but most of them are constrained by the fact that they either
require individuals to have similar calltypes or do not take into account the
change in vocal characteristics over time. Also, there is some loss in accuracy
due to background noise or varying degrees of recording quality. There was some
interesting research published that applied human-speech recognition methods
for classification of birds to some success, but even this was limited by its
lack of temporal stability. Another interesting characteristic of the loons is
that their call (or song) changes with a change in location. The objective is
to determine a scheme of sound identification for loon individuals with feature
extraction that is stable over time and location, which would greatly increase
the ease with which these birds are studied.
Full report (pdf)