Isolating biological acoustic transient signals

The authors investigate two methods for screening data to locate specific biological acoustic transient signals. The first method is based on a continuous hidden Markov model (HMM) and the second method is based on a matched filter. An experiment is presented which compares the performances of the two methods in isolating a single note from the song of a bowhead whale in the Arctic. Initial results indicate that the HMM screener performs better than the matched filter screener in the task of locating bowhead notes and rejecting white Gaussian noise. However, both methods identify a high proportion of interferers as possible bowhead notes. These interferers are in the frequency band of the notes, and sometimes resemble a portion of a note.<>