How would you feel next time you asked your dog, “Who's the good boy?” and were able to understand his response?
Kenny Chu, a professor of computer science and engineering at the University of Texas at Arlington, plans to use machine learning to convert dog barks into audio representations and eventually words. A National Science Foundation (NSF) Undergraduate Research Experience Site Grant will fund his research with a three-year, $483,804 grant.
“Scientists have been trying to decipher the calls of whales, dolphins and chimpanzees for years, but most of the work has been done by biologists and ecologists with little to no computer science background,” Zhu said. “That's changing, and now machine learning is being introduced into data analysis.”
“My research is about natural language processing for humans, but what about looking at how animals communicate?”
To collect the sounds, Chu downloads videos from YouTube and other sources, then filters out all noise in the audio files except for the barking sounds. He uses machine learning to categorize the sounds, breaking them down into syllable-like fragments and assigning each syllable an alphabet-like symbol.
So far, he has transcribed about 10 hours' worth of barks into syllables.
“We look at what's around the dog and what the dog is doing at the time to make a rough guess as to what the dog is trying to communicate and also try to give context to the syllables,” Chu said.
Early indications are that dogs in different parts of the world bark differently: He watched videos of Shiba Inu dogs from the U.S. and Japan and noticed that they didn't bark the same way, with Japanese dogs tending to “speak” faster and in a higher pitch than their American counterparts.
NSF's Undergraduate Research Experiences program encourages active participation of undergraduate students in research. The project involves students from computer science and engineering departments, as well as students studying animal behavior in biology and other fields. The students help collect data by submitting videos of dogs and other animals, and also help run the sounds through a transcription pipeline to create a catalog of different sounds and species.
“Since machine learning plays a role in everything we do in this project, it's a great opportunity for students to learn about big data analytics and explore how it can be applied to areas where machine learning hasn't been applied before,” Zhu said.
– Author: Jeremy Aghor (Faculty of Engineering)
