Spider conversations deciphered using machine learning and contact microphones

Arachnids are natural dancers. After millions of years of evolution, many species rely on fancy footwork to communicate everything from courtship rituals to territory fights to hunting strategies. Researchers typically observe these movements in the laboratory using so-called laser vibrometers. After the tool's light beam is directed at the target, the vibrometer measures the tiny vibration frequency and amplitude caused by the Doppler shift effect. Unfortunately, the cost and sensitivity of such systems often limit their use in the field.

To find a solution to this long-standing problem, graduate students at the University of Nebraska-Lincoln recently combined a series of small, inexpensive contact microphones with a machine learning program for audio processing. After packing his bags, he headed into the woods of north Mississippi to test his new system.

His Noori Choi research results, recently published in the journal Communications Biology, demonstrate an unprecedented approach to detecting spider movements across forest substrates, which are extremely difficult to detect. Masu. Choi spent two humid summer months setting up 25 microphones and traps in a 1,000-square-foot section of the forest floor, waiting for local wildlife to vibrate. Ultimately, Choi left the Magnolia State with his 39,000 hours of data, including more than 17,000 vibration series.

Of course, not all of these tweets were the wolf spiders Choi was hoping for. Forests are noisy places filled with active insects, chatty birds, rustling branches, and even the unpleasant sounds of human life, such as aircraft engines. These sound waves are also absorbed by the ground as vibrations, so they must be excluded from the arachnids of interest to scientists.

"Vibroscape is a more crowded signal space than expected because it includes both air and substrate vibrations," Choi said in his recent university profile.

Previously, this analysis process was a frustratingly tedious manual process that could severely limit the scope of your studies and datasets. But instead of overwhelming the roughly 1,625-day record, Choi turned to machine learning, which can filter out unwanted noise while separating the vibrations of three different species of wolf spiders (Schizocosa stridulans, S. uetzi, and S. duplex). Developed the program.