Optimizing the Detection of Wildlife using Sound and Machine Learning
Justin Brashares, Professor
Environmental Science, Policy and Management
Applications for Fall 2024 are closed for this project.
Monitoring the status of wildlife populations is critical to assessing ecosystem health. Rapid advances in machine learning and engineering have recently allowed researchers to deploy acoustic recorders that autonomously capture and classify population trends for vocalizing species, such as birds. Currently, UC Berkeley is preparing to launching the first ever SoundHub -- an open source data sharing platform for processing animal sound data and trend monitoring in a single workflow. However, while our bird models are proficient at classification, this URAP project aims to test and improve machine learning models for the classification of California's rare and endangered non-avian species (such as wolves, coyotes, frogs, and insects).
Role: We are seeking multiple students interested in biology and wildlife species identification or computer science. The students will identify and verify sound clips for a variety of species including coyotes, dogs, frogs, crickets, and birds. Students will be expected to study and memorize animal calls for accurate identification. Any students with experience in computer science or interest in coding in python are also welcome to participate in training the machine learning models, though this is not required. Sound identification, tagging, and testing can be done on a student's own computer or a campus computer. The student will train with project scientists and then work independently or as part of a team, while attending weekly group check-ins. Group check-ins will be scheduled for a time slot that works with everyone's schedule.
Qualifications: No experience required to apply! If you are interested and dedicated, you can learn to ID wildlife calls. However, previous experience in identifying animal calls is a plus. We can arrange to have additional computational tasks for anyone proficient in python or machine learning, though this skillset is not necessary. All students must have high attention to detail and patience with repetitive tasks. Students can work independently or in teams. Hours per week will be negotiated but at least 6 hours per week is preferred. Remote options available if necessary.
Day-to-day supervisor for this project: Amy Van Scoyoc, Post-Doc
Hours: to be negotiated
Related website: https://nature.berkeley.edu/BrasharesGroup/
Social Sciences Environmental Issues