Open Set Acoustical Bird Species Recognition Using HFCC

Robert Wielgat , Tomasz Potempa, Daniel Król, Agnieszka Lisowska-Lis

Acoustical identification of bird species is often used for the detection of bird species in the monitored area. Such detection would be money- saving, and especially helping in work- and time- consuming actions that are undertaken in hardly- accessible areas and need intensive work of specialist educated and skilled to recognise birds by only vocalisation. In the paper, bird voices recognition problem from open set is considered. Polish bird voices recorded in a highly noised municipal environment were recognized. Human Factor Cepstral Coefficients and Dynamic Time Warping were fundamental procedures used in recognition process. Bird voices were recorded digitally simultaneously by cardioidal and hipercardioidal unidirectional microphones. For recordings, relatively high 96 kHz sampling frequency has been chosen in order not to attenuate important information in bird voices laying in the higher frequency range. Digital Filtering was applied to the original recordings in order to enhance recognition accuracy. At the feature extraction stage of experiments effectivity of Human Factor Cepstral Coefficients and Mel-Frequency Cepstral Coefficients were compared indicating superiority of HFCC based approach. An optimal parameters of HFCC signal processing were selected. At the classification stage Dynamic Time Warping was chosen.

Robert Wielgat, Tomasz Potempa, Daniel Król, Agnieszka Lisowska-Lis
Department of Technology,
Higher State Vocational School in Tarnów,
Tarnów,
Poland
rwielgat@poczta.onet.pl