Keunwoo Choi

PhD ยท AI Researcher / Engineer ยท New York City ยท last.first@gene.com

I lead a team of MLEs and data engineers to train internal large language models (LLMs) in Roche / Genentech.

Previously, I applied machine learning and deep learning techniques to solve music and audio AI -- problems such as music tagging, source separation, transcription as well as audio understanding and synthesis.


Experience

Senior Principal ML Scientist

Leading development of large Language Models.

March 2023 - Present

Adjunct Professor

DM-GY 9103/Q Deep Learning for Media

Spring 2024

AI Advisor (Former AI Scientist/Director)

Gaudio Lab, Seoul, Korea

Leading an AI music/audio team covering music source separation, audio generation, and 3D audio.

March 2022 - Present

Senior Research Scientist

ByteDance AI Lab, Mountain View, USA

Large-scale deep learning systems for music and video content creation, social media, and music streaming service. Music research roadmapping for scalable and sustainable growth.

August 2020 - March 2022

Research Scientist

Spotify, New York, USA

Music content understanding and discovery for 300M+ users of the global leader in music industry

July 2018 - August 2020

Researcher

ETRI, Daejeon, South Korea

3D audio reproduction systems for immersive acoustic experience

Feburary 2010 - Jun 2014

Education

Queen Mary University of London

PhD in Computer Science, Centre for Digital Music, London, UK
Supervisor: Mark Sandler and George Fazekas
Thesis title: Deep Neural Networks for Music Tagging
October 2014 - March 2018

New York University

Visiting Scholar, Center for Data Science, New York, USA
Supervisor: Kyunghyun Cho
June 2016 - Janurary 2017

Seoul National University

Master in Electric Engineering and Computer Science, Seoul, South Korea
Supervisor: Koeng-Mo Sung
Thesis title: A NMF-Based Mono-To-Stereo Blind Upmix
March 2009 - Feburary 2011

Seoul National University

Bachelor in Electric Engineering and Computer Science, Seoul, South Korea
Full tuition fee waiver scholarship (2005 - 2009)
March 2005 - Feburary 2009

Activities

๐Ÿ† I received Best Paper Award at ISMIR 2017, the most prestigous music information retrieval research conference.
๐ŸŽ–๏ธ I was also nominated for Best Paper Award at ISMIR 2023.

๐Ÿ–ฅ I am the creator and maintainer of Kapre, a Python package for Tensorflow (Keras) Audio Preprocessing layers.
โŒจ๏ธ I am a contributor of librosa, Keras, and torchaudio.

๐Ÿ“„ I first-authored 10+ peer-reviewed papers at venues including ISMIR, ICASSP, EUSIPCO, ICML Workshop, etc.
๐Ÿ“– I review papers at NeurIPS, ICML, ICLR, ISMIR, ICASSP, AAAI, etc.
๐Ÿ’ก I created and have co-organized an AI challenge for sound synthesis in 2023 and 2024.
๐Ÿ’ก I am one of the general chairs of ISMIR 2025, with Prof. Juhan Nam and Prof. Dasaem Jeong.

๐Ÿ“• I co-authored a web book about music classification. We used it in our tutorial at ISMIR 2021
๐Ÿ“• I also wrote another web book, 'LLMs <3 MIR', 2023 May. It is a tutorial on large language models for music information retrieval.

๐ŸŽผ I make music as keunwoo.OOO, as a composer, lyricist, singer, and producer.
โœ‰๏ธ I try to convert my friends into JazzBees through my tiny, personal, and non-intimidating newsletter called JazzBuzz.