Bird song identification using artificial neural networks and statistical analysis

A system for automatically identifying six bird species by their songs was implemented. Pre-processing of sampled songs extracted temporal measurements of periods of sound and silence within songs. Power spectral densities were used to extract spectral information. Statistical methods were used to reduce data dimensionality and for identification tasks. An artificial neural network was also used for identification. Quadratic discriminant analysis achieved a 93%, and a backpropagation neural network 82% overall accuracy.