The Data
Parrot Data
Initially, the goal was to classify parrots based on their calls. This analysis proved unsuccesssful because of the non-stationary nature of parrot calls throughout their lifetimes. The data used was recordings of calls and squawks recorded by Liz Bradley in Irene Pepperburg's lab.
Nips4b
Next, the nips4b data set found in Automatic large-scale classification of bird sounds is strongly improved by unsupervised feature learning by Dan Stowell was used for machine learning analysis. This data set came from the Sabiod Neural Information Processing Scaled for Bioacoustics challenge, and contained recordings with several species present. This didn't align well with the goals of the project, so it is no longer being used.
xeno-canto
Currently, recordings from xeno-canto.org are being used. This allows large numbers of recordings of a single species to be imported at once.
What's next?