Winning the Barkopedia Competition at IJCAI 2025

My team WhatDoesTheDogSay won the Barkopedia Competition at IJCAI 2025 in Montreal, Canada!

What is Barkopedia?

Barkopedia is a large-scale competition for canine vocalization analysis, hosted at IJCAI 2025. The challenge was designed to push the boundaries of machine learning applied to animal communication, providing a diverse benchmark dataset of real-world dog barks sourced from platforms like YouTube and Reddit.

The competition featured eight tasks spanning two broad categories:

Demographic & Identity Classification:

• Dog Age Group Classification
• Dog Sex Classification
• Dog Breed Classification
• Individual Dog Recognition

Contextual & Audio Processing:

• Dog Activity & Environment Classification
• Dog Emotion Classification
• Dog Vocal Separation (DVS)
• Dog Vocal Detection (DVD)

Our Approach

I competed across the demographic and identity classification tasks — age, breed, and individual recognition. The approach adapted modern deep learning techniques originally developed for human speech and audio processing to the domain of canine vocalizations. Working with the noisy, variable nature of real-world bark recordings was a fun and unique challenge compared to the typical time-series and signal processing problems I work on.

Being invited to present the winning solution at IJCAI in my home city of Montreal was a great experience!