By Tyler Pittman
Autonomous acoustic recording units (ARUs) are a popular technology for surveying and monitoring vocal wildlife populations from bats to birds to marine mammals. ARUs are popular because they can collect huge quantities of data across large areas and time spans with very little effort. However, the sheer quantity of data requires advanced computer programs for efficient processing, and current commercial software cannot effectively detect species like wild turkeys that have calls that cannot be easily distinguished from background noises.
In 2017, after manually listening to thousands of hours of wild turkey audio, FWRI began development of a custom program to automate the processing of these audio files by partnering with researchers from Southeastern Universities Research Association (SURA). Commercially available programs are based on the concept of matching the spectrogram of a potential turkey call with that from a known turkey call (i.e., a template). Our approach differs from that by breaking the template into smaller different sized and positioned sections called sub-masks. Additionally, our program weights the sub-masks either positively or negatively toward correct identification and allows the user to assign a set of rules to define which sub-masks take priority over other sub-masks. To date, the preliminary versions of the program have proven to be effective at identifying wild turkey gobbles from unknown audio recording on training datasets. The best-documented performance was 83% correct identifications with only 17% false-positive detections compared to 99% false-positive detection rates from commercially available software. Testing will continue in 2019 to further develop the program in to a useful and efficient tool for monitoring populations of birds and other wildlife.
Note: Cover image shows a recording of a wild turkey gobble represented as the amplitude of sound plotted against frequency and time, also known as a spectrogram.