An approach of algorithmic clustering based on string compression to identify bird songs species in xeno-canto database

In this work, we analyze the usefulness of the normalized compression distance (NCD) as a similarity measure to bird species identification through audio samples. As a first approach we review the effect of different compression methods from 7z and CompLearn Toolkit, over subsets of bird audio samples obtained from the xeno-canto database. The performance of each compression method was measured applying hierarchical clustering and projection mapping to the distance matrix, and later on, measuring the quality of both of them. Our results are very promising and show that the identification of a bird species among multiples audio samples is possible through NCD-based-on clustering.