It's a talking dog (collar)!

3/20/2019 Amanda Maher

Written by Amanda Maher

 

To clarify one big thing that the internet has been confused about, no, we are not actually reading the dog’s mind in the sense that we are translating her thoughts from dog speak into English. That being said, we are “reading her mind” in that we are detecting the electrical activity from her neurons as she responds to stimuli, so we are reading the activity of her mind.

The idea was inspired by sweet little Dug the Dog from the movie “Up,” and brought together by our team lead, Jessica Austriaco. She would joke with her friend, and eventual team member, Christine Lannon, that they would one day make Christine’s dog, Alma, talk. Jessica decided to make that dream into a reality, and created a project through the Bio-Medical Engineering Society and created Alma the Talking Dog team.

Alma's collar.
Alma's collar.
Before we get into the behind the scenes of how the device functions, let’s talk about brain waves a bit. The neurons in our brains, and dog brains, give off electrical impulses. The frequencies of these signals change when the brain is doing different things. We observe Beta waves of frequencies between 7-12 Hz during peaceful and relaxing activities, and we observe Alpha waves between 12- 30 Hz when the brain is controlling motor functions. In this project, our goal was to monitor these signals, and use them as a cue to understand when Alma was responding to a stimulus—when she saw a treat. We took Alma’s excited electrical signal as an indication that she was excited about the treat in front of her, and had a speaker spit out what we guess she may be thinking, “Treat! Treat! Yes, I want the treat! I do so definitely want the treat! I would be very happy if I would have the treat!”
Credit to Bliss’s Github for the awesome gif (above) that visually explains Fourier transforms!
Credit to Bliss’s Github for the awesome gif (above) that visually explains Fourier transforms!
As you can imagine, there was a ton of work that went into this project. We 3D printed electrodes (a feat which is still amazing to me that we can even do) and performed a couple of iterations on them to make them optimal and strong enough for fit on Alma. We then tested some nickel paint and found that it is electrically conductive, and coated our electrodes in it so the electrical signal of Alma’s brain waves could travel through them. While all of the iterations and tests were taking place with the electrodes, our Hardware Lead Matthew De Venecia and Signal Processing Lead Bliss Chapman were working on toy problems to help us understand EEG data. They performed several iterations of their own on the circuitry and machine learning techniques.

When everything finally came together and we were ready to execute the test, we placed one electrode on Alma’s head, and another on her ear to act as a ground for the circuit. When Alma gave off an electrical signal, it passed through the electrodes and into an analog circuit that will amplified the raw electrical signal, and also performed low and high pass filtering on the signal so we could focus on just the Beta and Alpha range of signal from 7-31 Hz. That information was brought through an Arduino which writes electrical data into a Raspberry Pi. The Pi performed a Fourier transform on the data to decompose it into frequencies. Credit to Bliss’s Github for this awesome gif below that visually explains Fourier transforms! From there, machine learning was performed on the data to classify if the signal we received was Alma seeing the treat or not. If she did see the treat, we programmed the speaker to spit out her excited phrase. If she didn’t see the treat, nothing was played from the speaker.  

The result was pretty awesome, you can see in this clip of our test trial when Matthew walked into the room with the treat in hand, Alma got excited, our system detected that, and the speaker spit out her excited message.

The key to our success is that we had a team of very talented and highly motivated people that each brought something unique to the project. We started with our own resources, a ton of Matthew’s personal circuitry stash, one of my old bandannas from marching band, and my roommate’s sewing kit. It was tremendously exciting as the mechanical lead to see all of the components finally come together and actually function during the final build on the collar. As a sum of our parts, we made something really awesome that we believe has great potential to grow into something bigger.

For that reason, we’re going to continue this project by turning it into a Registered Student Organization. If you’re interested, please fill out this poll!

I’d like to thank everyone on this fantastic team for bringing their unique creative expertise, and Alma for being one of them most well behaved and goodest girls I have had the pleasure to know.

Read more from the (Fluid) Dynamic Trio >>


Share this story

This story was published March 20, 2019.