Just as Aquaman can talk to fish, so too can researchers at the University of Washington chat with rats. Or, at least, they can better understand what rats are squeaking about thanks to a deep learning artificial intelligence (A.I.) system with the ingenious name “DeepSqueak.”
Since rats are frequently used in medical experiments, the technology makes it easier for researchers to monitor rat stress levels or other metrics when they are being used in testing. While it’s not exactly a Google Translate for rodent-to-human conversation, it could be useful for decoding the natural patterns of vocalizations made between rats as they communicate — thereby making it easier to understand their responses to stimuli.
“Rats and mice express themselves through ultrasonic vocalization at frequencies that are too high for humans to hear,” John Neumaier, a professor in the Department of Psychiatry and Behavioral Sciences, told Digital Trends. “In the past, researchers have recorded these to gain better insights into the emotional state of an animal during behavior testing. The problem was that manual analysis of these recordings could take 10 times longer to listen to when slowed down to frequencies that humans can hear. This made the workload exhaustive and discouraged researchers from using this natural read out about animals’ emotional states.”
DeepSqueak was developed by Kevin Coffey and Russell Marx, two researchers in Neumaier’s lab. “An undergraduate student isolated and annotated hundreds of calls by hand,” Coffey told us. “Those calls were used to train a rudimentary network that could be used to find thousands of new calls. This larger data set was cleaned by hand, and some calls were randomly augmented, stretched and squished to improve generalizability.”
Previous attempts to carry out this analysis using software was less reliable. Being easily tricked by background noises and generally worse at categorizing sounds. These studies also focused only on simple aspects of vocalizations, such as their frequency range, which can correlate with happy or unhappy emotional states. With DeepSqueak, however, it is possible to look at more subtle features, such as the order of calls and their temporal association with actions and events.
Right now, not enough is known about exactly what all rat squeaks refer to. But tools like this make a very promising platform to build upon. As biologists compile more calls over time, it hopefully won’t be long before a rodent Rosetta Stone becomes a reality.
A paper describing the work was recently published in the journal Neuropsychopharmacology.
