Hidden Markov Model Based Animal Acoustic Censusing: Learning from Speech Processing Technology

Individually distinct acoustic features have been observed in a wide range of vocally active animal species and have been used to study animals for decades. Only a few studies, however, have attempted to examine the use of acoustic identification of individuals to assess population, either for evaluating the population structure, population abundance and density, or for assessing animal seasonal distribution and trends. This dissertation presents an improved method to acoustically assess animal population. The integrated framework combines the advantages of supervised classification (repertoire recognition and individual animal identification), unsupervised classification (repertoire clustering and individual clustering) and the mark-recapture approach of abundance estimation, either for population structure assessment or population abundance estimate. The underlying algorithm is based on clustering of Hidden Markov Models (HMMs), commonly used in the signal processing and automatic speech recognition community for speaker identification, also referred to as voiceprinting. A comparative study of wild and captive beluga, Delphinapterus leucas, repertoires shows the reliability of the approach to assess the acoustic characteristics (similarity, dissimilarity) of the established social groups. The results demonstrate the feasibility of the method to assess, to track, and to monitor the beluga whale population for potential conservation use. For the censusing task, the method is able to estimate animal population using three possible scenarios. Scenario 1, assuming availability of training data from a specific