Targeting Input Data for Acoustic Bird Species Recognition Using Data Mining and HMMs

In this paper we propose the integration of Data Mining with Hidden Markov Models when applied to the problem of acoustic bird species recognition. We first show how each of them is applied on an individual manner, contrast their results and devise a model to combine them for targeted classifications. Previous work has shown that large collections of spectral attributes are needed in order to represent the structure of bird songs, therefore elevating the computational requirements when applied to distributed sensor networks. Data Mining is used to reduce the dimensionality of the spectral attributes and for classification. Hidden Markov models represent a traditional approach and require strong song preprocessing. Our results show that Data Mining can yield efficient results with low requirements and could serve to target HMMs input parameters.

• Type: Conference
• Year: 2007
• Where: ICDM
• Authors: Erika Vilches, Ivan A. Escobar, Edgar E. Vallejo, Charles E. Taylor
• Keywords: Data Mining, Hidden Markov Models, ICDM 2007, Spectral Attributes