The International Research Fellowship Program enables U.S. scientists and engineers to conduct nine to twenty-four months of research abroad. The program's awards provide opportunities for joint research, and the use of unique or complementary facilities, expertise and experimental conditions abroad.
This award will support a twenty-two-month research fellowship by Dr. Mark D. Skowronski to work with Dr. M. Brock Fenton at the University of Western Ontario in London, Ontario, Canada.
This research will significantly advance the state of the artin bioacoustic signal processing by leveraging methods developed during the past several decades for automatic speech recognition (ASR). The acoustic domain is a significant medium for many species of animals (e.g., bats, birds, marine mammals, frogs, and insects) and has been used extensively to observe and monitor various animal activities. Acoustic analysis is typically performed using statistical tests on signal characteristics determined important by an expert observer. Such expert-based systems are pervasive in bioacoustic analysis, including early ASR research. However, ASR research shifted during the 1980s from expert-driven methods to data-driven methods, and today's state of the art for ASR systems incorporates the machine learning paradigm primarily because machine learning better accounts for the natural variability of human speech as well as that of the recording environment. The current study will rigorously explore the capabilities of the new paradigm of analysis methods in several ways under the direction of one of the worlds foremost authorities on bats. The machine learning methods from ASR will be extended to include several species and more acoustic features that are robust to noise and overlapping signals. Machine learning methods will also be used to model sources of variation (e.g., habitat differences) and different levels of classification (e.g., guilds, genera, and individuals). Furthermore, systematic optimization techniques will be applied to the methods to optimize detection and classification performance for both off-line analysis and real-time field monitoring.
