Due to their many genetic traits similar to humans, rats have been ideal subjects for medical research for many decades. However, do not forget that they are social creatures and their behavior is also interesting from the point of view of science.
Scientists at the University of Washington have developed an unusual instrument called DeepSqueak for studying ultrasound signals from rats. It is a highly accurate neural network capable of analyzing and classifying animal sounds.
Learning the "language" of rats is a time-consuming task. Most of these sounds are above the human threshold of 20 kHz. As a result, their analysis takes 20 times longer than the sound itself. For example, the analysis of one hour of "rat" audio requires 20 hours of algorithms.
This is where DeepSqueak comes in handy, which is a deep learning algorithm that can extract and sort rat voices from raw audio recordings. The core of the program is software used by unmanned vehicles to prevent collisions with obstacles. By improving it, scientists were able to group sounds according to similar characteristics and even learned how to identify specific patterns. Their presence may be a sign of primitive syntax. This, in turn, indicates that rats may have their own language.
The researchers have made their algorithm publicly available on GitHub in order to attract as many enthusiasts as possible to work on the study of the rat language.