It’s gonna be the future soon, and I can’t wait!
It was only a matter of time before someone leveraged machine learning and algorithmic AI to parse cat vocalizations, and thanks to Javier Sanchez, translating your cat’s meows — and trills, huffs and chirps — is now a reality.
Sanchez was a member of Amazon’s machine learning team contributing to the development of Alexa, the now-ubiquitous virtual assistant operated by voice commands.
“I got to see how the sausage was made, how they train their models and work with all the data science platforms,” Sanchez said. “So I was fresh off the heels of that and I was thinking, ‘Well, we could do something similar with cats and it could be an app.’”
Sanchez’s new employers at the tech firm Akvelon saw promise in the idea and gave him the green light.
There are two layers to the concept: The first one involves nine or 10 “intents” common to all or most cats. They include vocalizations for “Feed me,” “Hey human!”, “Let me out,” and “Pay attention to me,” among others.
Sanchez didn’t guess or intuit the meanings — they’re based on research by a team at the University of Milan, who built a data set of cat vocalizations by attaching tiny microphones to cats and recording everything the fluffsters say. Each feline utterance was analyzed and catalogued by frequency, rhythmic quality and context, among other traits.
The second layer is where it starts to get really interesting: By using MeowTalk like Shazam, the app will start recording your cat’s particular trills, chirps and meows and — with your help — eventually piece together what they mean.
As with dictation software and machine learning in general, the more data the app gets, the better its translations become.
This is important because, while cats share many sounds, each cat develops its own unique vocalizations:
With MeowTalk, you can create a profile for your cat and start using its auto-recognition to translate your cat’s meows and start mapping its language. While some translations are built-in and inherent to the app, translations specific to your cat require you to train the app to recognize your cat’s specific vocabulary and intentions. Translations you deem to be incorrect can be corrected via the app. MeowTalk is not static; instead it learns and evolves with each translation that you confirm, adding to its corpus, just as we would add new words into our own memory banks or language processing programs.
At the same time thousands of other cat owners are also using the app, feeding the algorithm more data, which the app uses to improve itself. Development is ongoing, with future changes reflective of user (and cat) feedback.
“A tool like this can help certain people bond even more with their cats, especially if they can’t be in contact with other people on a regular basis,” Sanchez said. “So this could be a real game changer for a key demographic that have cats.”
Applying what they’re learning via the app, Sanchez and his team are also working toward their next goal: Giving your cat a human voice. They’re developing a small device that clips on a cat’s collar and translates meows into human speech in real time.
That tech has the potential to give me nightmares. Imagine Buddy having a human voice and saying “Gimme snacks now, servant!” “Open the door, butler!” “You’re 23 seconds late with dinner!”
Maybe I’ll pass on the collar device. In the meantime I plan to download the iOS version of MeowTalk and give it a spin. I’ll report back in a week or two after giving it some time to adjust to the Budster. If any of our readers give it a shot, we’d like to hear your impressions as well.