Spectral and Cepstral Analysis of Colombian Birdsongs using Multidimensional Scaling

Birds population monitoring is a widely used approach for assessing the state of biodiversity. Since a birdsong contains features specific to distinct species, acoustic signals recording and processing are used for addressing the birds population characterization. This work deals with the analysis of information contained in recordings of birds from the Colombian ecosystem. For this purpose, the signals are processed using the Fourier transform and a novel cepstral representation based on the wavelet transform. The spectral and cepstral metrics are analysed using the multidimensional scaling (MDS) method, that yields a two- or three-dimensional visualization of the data. The interpretation of the results is based on the patterns of points in the MDS chart and the metadata of the birdsongs recordings. The results evince that the spectral and the proposed wavelet cepstral representations characterize different aspects of the birdsongs dataset. Emerging patterns are related with the bird species, the taxonomic order and the altitude above the sea level. Indeed, the MDS approach is an attractive option for unveiling characteristics embedded in the birdsongs.