Frameless processing of bird songs using set membership identification

This paper describes a novel approach to feature extraction from bird songs using a set‐membership identification (SMI) algorithm. The low computational complexity of the SMI algorithm allows frameless pointwise feature estimation and real‐time processing. Both energy‐based end‐point detection and set‐to‐point classification methods are incorporated in this SMI processing for enhanced labeling performance. The described algorithm serves as a front end to a fully automated bird identification system in which training data collection is automated by scheduled computer generated phone calls to a cellular monitoring station. RASTA processing of feature vectors compensates for the telephone channel effects.