Why We Hired a PhD Physicist for an ML Project
While developing DART, Certus’s Machine Learning (ML) based TAK plugin for gunshot detection, identification, and localization, the team reached an inflection point. They had expertise in software, machine learning, and signal processing to an extent. However, they lacked a team member who deeply understood signals, math, and ideally audio.
Ryan Quinn, Director of Software Development at Certus Innovations, recalled his thought process on the matter, “You need someone who really understands signals and math and ideally audio. You know, maybe someone who has a doctorate in physics and a background in music.”
Shortly after Certus was selected for a TACFI, Peter Scialdo, President of Certus Innovations, challenged Ryan to address a critical gap in the TEAM-ML project. As Ryan recalls, Peter put it simply: “You know, you got this, like, kinda skill hole in your project here where you need someone who wants to do a lot of math and physics and, and maybe some machine learning. What are you gonna do about that?”
Ryan responded, “You know, Pete, you're not gonna believe this ... but I might actually, know the guy for that."
Eric Dohner, a PhD in Physics, has a unique background for Certus’s TEAM-ML projects: seven years of hands-on experience with Fourier transforms, built through implementing them by hand in Fortran, a scientific programming language, during his doctoral work. His expertise also extends to the mathematical foundations of music, giving him a unique perspective on the intersection of math, physics, and sound.
That background turned out to translate directly: within his first two hours on the job, he identified that an impulsive sound like a gunshot can't be modeled as a sum of sines, a distinction the team had spent roughly two years working toward independently.
Because of TACFI, Certus Innovations had the funding to bring on a dedicated math and physics specialist for the TEAM-ML project, and the search started with a direct conversation between team leadership about a skill gap that needed filling. Eric now leads feature extraction research for the project, working on problems like frequency-adaptive windowing that the team had identified but hadn't had the bandwidth to pursue.
That combination of math, physics, and audio expertise is now driving Certus’s work to push audio ML forward. In a recent conversation, Ryan, Sinclair, and Eric explored the challenges and opportunities ahead, from gunshot and UAS detection to building realistic audio datasets and developing more efficient approaches to signal processing. Watch the full conversation on YouTube.
Continue the conversation: Watch the full discussion on YouTube
