Frogs are a diverse group of amphibians, with over 6,000 known species. They are found on every continent except Antarctica, and play important roles in the ecosystems in which they live. Despite their importance, many frog species, and more generally amphibian species, are threatened or endangered due to their sensitivity to human activities. There is thus a need for active conservation efforts and biodiversity assessment.
These efforts often include species tracking and identification to evaluate regional differences in fauna or determine the temporal changes in certain populations. Such studies involve large amounts manual labor, including physical field research and the processing of large amounts of visual and audio data.
Fraught with the potential of human error and the urgent need to improve efficiency of these efforts, modern day conservation must to incorporate the use of technology. Thus, one of the target applications would be to use machine learning to process the large amounts of field recordings taken by biologists and classify any recognizable sounds. However, this can have its difficulties with the large amounts of background noise in the wild. Nevertheless, in this project I attempt to use linear algebra techniques to improve the effectiveness of machine learning in classifying frog calls from raw field research audio data.