Meet Cochl’s Team: Donmoon Lee, Co-founder & Researcher
What does Cochl’s research team do?
We are building an AI model for Cochl’s API and SDK that tags a sound input. We select candidates for the new sounds to be added for the next product update, manage the sound data collection process, which sometimes involves collecting by ourselves, and optimize the algorithm to be brought to the real world.
What are you responsible for at Cochl’s research team?
I do a lot of experimental trials to improve our algorithms. A machine learning algorithm could easily get outdated when following a standardized structure. So I constantly search for new ML algorithms that could be applied to enhance our solutions. I refer to a lot of computer vision algorithms since computer vision is the largest field in machine learning.
What has been your favorite project so far?
My projects usually take three to six months to complete. I recently made a Voice Activity Detection(VAD) and it worked as I had expected. VAD is the detection of the presence or absence of human speech, used in speech processing. Speech processing is very interesting and yet could a very challenging field since it involves distinguishing factors like age and gender and so far is far from perfection.
How does music and audio differ from computer vision from a researcher’s perspective?
First of all, dealing with music and audio data is different. It is not easy to get started with a proper audio dataset, as well as to define an audio object — some sounds have length of 1 second, and some have 5 seconds. For computer vision, there are abundant datasets freely available such as MNIST, CIFAR and ImageNet. Unlike audio datasets, image datasets are normally pre-processed to 256x256 pixels. It’s an old problem this field has been facing that the audio data lack quality and consistency. There has been recent efforts for example by Google’s AudioSet which is made up of 10-second sound clips.
What made you decide to study at Music and Audio Research group at Seoul National University?
I initially wanted to study Acoustics at Seoul National University. However, the Acoustics division was having problem finding a successor to then-retiring professor. Luckily my friend who graduated from Acoustics recommended me to study Music information retrieval (MIR), and that’s how my journey at Music and Audio Research group began. I was thinking to leave SNU after obtaining M.S. degree, but there was a turning point. I won the first prize at Big Data Challenge, and that gave me confidence that I can excel in this field — so I ended up pursuing Ph.D.
As a postgraduate student, how do you balance your study and work?
Frankly, it’s a tough job. I set my timeline to submit two papers within this year and graduate next August. Juggling work and study wasn’t easy however, and my paper submission has been delayed.
What kind of competency do you need to be a better researcher?
I think what’s required of a good researcher at a company is different from that at school. What I believe a good researcher at a company, especially in this field is, 1) should have great interest in one’s field, 2) should have persistence not to lose one’s direction, and 3) should be able to implement ideas with programming languages.
What was the most memorable event during your time at Cochl?
Winning DCASE challenge in 2017. In March 2017, we started building Cochl as a company. Four out of six co-founders participated in DCASE 2017. Four of us together scored two 1st places and two 2nd places in four tasks at the competition. It was so touching that one of the co-founders shed “tears of joy”.
What is your favorite part of Cochl’s culture and benefits?
For me, it’s the flexible work hours. We are very fortunate given the fact that very few companies in Korea offer this kind of discretion to use time freely during work hours. It has really helped me to focus on school work when I needed.
Do you have any goals you want to achieve right now?
Completing my Ph.D study is on top of my list. To submit my papers and graduate soon.
What’s something most people don’t know about you?
The fact that I love active sports. I boxed for a year and been doing Brazilian jiu-jitsu (BJJ) for two years. I was getting ready for jiu jitsu competition until the outbreak of COVID-19 situation.
Aside from work, what keeps you busy and what do you enjoy doing?
I love staying home and I spend a lot of time planning and making a quality meal.
What’s your favorite dish when you cook at home?
I recently made a stir-fry tofu noodle mixed with oyster sauce, bacon, and lettuce. It was quite nice!
What is your favorite sound?
I like the sound of Hammond B3 organ, especially its watery sound effect played on a rotating Leslie speaker. It’s one of the classics of the electronic organs, and you can manually add various sound effects to it!
Please give a piece of advice to people who are interested in joining Cochl’s research team.
If you are someone who can see the bigger picture, rather than indulge in a specific topic — for data scientist, this could refer to someone who is self-conscious and balanced, rather than being too much into certain algorithms or recent trends — we’d love to hear from you!
