How researchers use machine learning to re-create chirps and trills produced by forest insects when Dinosaurs lived on Earth

You must have heard the ancient insect sounds and songs in Steven Spielberg’s dinosaur movies wherein the trills and high-pitched chirps fill the primeval forests as characters move through large leaves and trees. While those sounds have been whipped up by Hollywood sound engineers, scientists are taking the help of technology for a different approach. They are using machine learning to reconstruct them straight from 165-million-year-old fossilised insect wings.According to a study published in the Proceedings of the National Academy of Sciences, an international team of researchers blended exceptionally preserved fossil remains, biomechanical tests on living species, digital simulations and machine learning models to resurrect a partial acoustic portrait of the Middle Jurassic period. If verified, one of the discovered signals stands as the earliest recorded evidence of ultrasonic animal communication which predates the rise of echolocating bats by tens of millions of years.Commenting on the paper, University of Denver behavioral ecologist Robin Tinghitella, who researches cricket evolution but was not part of the study, noted that listening to a hypothetical prehistoric soundscape generated directly from data is a compelling scientific milestone.How researchers decoded ancient sound from wing designAcoustic waves cannot turn to stone, but the specialised bone-and-cuticle structures that generate them do leave a fossil trail. Crickets and katydids (bush crickets) produce noise via stridulation. One forewing carries a serrated vein known as a “file,” studded with minuscule, toothlike ridges. Rubbing an opposing wing's scraper across this ridged file sets the wing membrane into rapid vibration. By measuring the exact physical spacing and length of these ridges alongside the wing's overall surface area, researchers can reliably determine an insect's acoustic frequency and pitch.The investigation centered on 20 distinct fossils spanning nine separate species discovered within the same geological horizon in modern-day Inner Mongolia. This shared environment gave scientists the opportunity to model a community of creatures that inhabited the very same prehistoric forest canopy.Scientists are validating models with lasers and machine learningTo translate flattened fossils into living audio, researchers developed biomechanical models and tested them against modern crickets and bush crickets. Using precision laser systems, they mapped how today's insect wings flex and resonate during song.Having checked the mechanical equations on existing species, the scientists then tested those parameters on fossilized wings. Machine learning algorithms were brought in to process the repetition rates, estimating how fast these extinct species rattled out their core sound pulses.The resulting audio profiles paint an unexpected picture of the ancient night. For example, most Jurassic species trilled in lower tonal registers similar to contemporary crickets. However, an extinct katydid relative called Sigmaboilus peregrinus emitted frequencies between 20 and 22 kilohertz, essentially crossing the boundary into ultrasound just outside normal human perception.Interestingly, rather than buzzing or rasping like some modern insects, many of these species produced narrowband pure tones that resemble sharp electronic beeps or clear, high-pitched whistles.Study co-lead Fernando Montealegre-Zapata, a sensory biologist at the University of Lincoln, explained that these concentrated musical frequencies make it exceptionally difficult for listeners to pinpoint where the noise originated—a vital defense mechanism against predators hunting in the dark.Montealegre-Zapata's team suggests that as early mammalian predators developed sharper hearing to track down calling insects, evolutionary pressures favoured cleaner, higher-frequency tones that shielded their positions.Get the latest technology news and updates. Download the TOI App.