A recent breakthrough in bioacoustic deep learning technologies – a method for automated detection of animal sounds – has been made at Cornell K. Lisa Yang Center for Conservation Bioacoustics. Lisa Yang Conservation Center. Dr. Shyam Madhusudhana, a postdoctoral researcher in the Lab of Ornithology, has built a toolkit that enables bioacoustics to create complex voice recognition models with just a few lines of code.
tool kit, kojoin a recent study that outperformed marine analysts in detecting blue whale calls by D.
Blue whale D-calls are calls of varying frequency produced by male and female whales, unlike the well-known whale song produced only by males. While whale songs are often predictable and easily identifiable, D calls are erratic and are produced less frequently.
However, while blue D whale calls are difficult to identify, monitoring their presence allows for a better understanding of their migration patterns and vocal behaviors.
Acoustic monitoring has long been followed as a viable method for recording rare species that lack sufficient visual data. In recent years, machine learning algorithms have shown promise Results in the analysis of acoustic monitoring data. In the marine biome, where visual surveys are difficult to perform, this method becomes even more important in efforts to track the movement and habits of different aquatic species.
This is where Koogu comes in.
“As long as someone has their own annotated data set [of acoustic monitoring]They can take Koogu and build their own model,” Madhusudhana said.
This methodology was adopted by a team of researchers in the Australian Antarctic Department led by Brian Miller. The researchers used Koogu to build an automated detection model for their study of blue whale calls.
they study, co-authored by Madhusudhana, titled “Deep Learning Algorithm Outperforms Experienced Human Monitor at Detecting Blue Whale Calls: Dual Monitor Analysis.” It found that human experts detected 70 percent of D calls while the model accurately detected 90 percent of whale calls. The model’s detection rate was also much faster than marine analysts, lacking the fatigue factor associated with human analysis.
The study is only the first case in which Koogu has been used effectively. However, according to Madhusudhana, Koogu is far from being limited to only detecting marine audio data.
“Koogu isn’t just a whale calling toolkit – [it is] It’s just a convenient way to build machine learning solutions – anything from whales to birds as well as insects,” Madhusudhana said.
Koogu has the potential to be an influential tool in the field of bioacoustics. While there has been significant development in the field of machine learning, most of the development in the field of phonetics has to do with human speech. Madhusudhana said that Koogu bridges the gap between the two.
“If you’re looking at a visual representation of sound – like a spectrogram You can treat it as an image and apply image classification techniques to it,” Madhusudhana said.
Koogu converts audio data into a form that machine learning models can use for visual classification. Madhusudhana emphasized that most of the model was configurable. Any dynamic audio expert can change the parameters and then modify how the audio is converted into images. Next, the images are categorized using the image classification model.
“If I tried to develop a solution based on a dynamic sound neural network, there would probably be a few hundred lines of code required. What I did was [enabled you to] “Call three or four jobs and you’re done,” Madhusudhana said.
The goal was for bioacoustics and other researchers to be able to use their data and knowledge of the field and combine it with Koogu’s functions to efficiently analyze sounds. Koogu’s unique importance lies in the process of converting audio to image.
As Madhusudhana explains, each sound is converted into a color map to easily distinguish one sound signal from another. When this is compressed into an image for image classification, significant data loss occurs. Koogu avoids this data loss, which greatly increases accuracy.
This feature is especially evident in audio recordings of low or medium intensity. Recordings like this make blue whale calls difficult to detect—especially in the case of human experts.
The open source universal voice recognition toolkit has greatly simplified the automated voice recognition process.
But acoustic whale monitoring is only one part of the equation, according to Madhusudhana. “Our goal is to preserve biodiversity across species – and that was the goal of Koogu – [to] You have a very generic thing that anyone around the world can use.”