2017년 하반기 이후로 이 페이지 업데이트를 멈춥니다.
논문 관련 최신 정보는 제 구글 스칼라를 참고하세요.
Queen Mary University of London, C4DM (Centre for Digital Music), Oct 2014 - Mar 2018
- Transfer Learning for Music Classification and Regression tasks
Keunwoo Choi, George Fazekas, Mark Sandler, Kyunghyun Cho
18th International Society of Music Information Retrieval (ISMIR) Conference, 2017
arXiv | pdf | github
제가 음악 태깅용으로 학습한 네트워크를 특징값 추출기로 사용하고 이걸로 6가지 데이터셋에 적용해 음악 장르 분류, 음악/음성 분류, 음악 감정 분류 등 다양한 작업을 수행해봤습니다. 기존 오디오 피쳐보다 좋은 성능을 보여주네요.
- Convolution Recurrent Neural Networks for Music Classification,
Keunwoo Choi,George Fazekas, Mark Sandler, Kyunghyun Cho
IEEE Conference on Acoustics, Speech, and Signal Processing (ICASSP), New Orleans, USA, 2017
arXiv | pdf
Convnet을 특징 추출로 쓰고, 그렇게 추출된 특징을 RNN으로 학습하는 구조를 CRNN (Convolutional recurrent neural network)이라고 합니다. 이 구조를 이용해 음악 태깅을 했습니다. 그리고 아주 면밀한 비교 실험으로 여러가지 구조를 벤치마크했습니다.
- Towards Music Captioning: Generating Music Playlist Descriptions
Keunwoo Choi, George Fazekas, Mark Sandler, Brian McFee, Kyunghyun Cho
Late-Breaking Demo Session of 17th International Society of Music Information Retrieval (ISMIR) Conference, New York, USA, 2016
arXiv | pdf | github
자동으로 음악 플레이리스트를 생성한다면 그에 맞는 '설명'도 자동으로 생성되어야하지 않을까요? 관련 연구 초록입니다. 2+1페이지 분량의 간단한 설명입니다.
- Explaining Deep Convolutional Neural Networks on Music Classification
Keunwoo Choi, George Fazekas, Mark Sandler
arXiv:1607.02444
arXiv | pdf | blog post (w/ reviews) | demo | press article
컨브넷을 음악에 적용하면 어떤 특징값이 어떻게 학습되는지 분석한 논문입니다. 6+1페이지 분량.
- Towards Playlist Generation Algorithms using RNNs Trained on Within-Track Transitions
Keunwoo Choi, George Fazekas, Mark Sandler,
SOAP Workshop (Workshop on Surprise, Opposition, and Obstruction in Adaptive and Personalized Systems), Halifax, NS, Canada, 2016
arXiv | pdf | blog post
데이터 기반의 플레이리스트 자동 생성, 참 좋은데 문제는 쓸만한 데이터가 많이 없습니다. 노래 내부의 '흐름'을 레퍼런스로 플레이리스트를 만들어주면 어떨까요? 4페이지 분량의 워크샵 논문입니다.
- Automatic Tagging using Deep Convolutional Neural Networks
Keunwoo Choi, George Fazekas, Mark Sandler
17th International Society for Music Information Retrieval Conference, New York, USA, 2016
arXiv | pdf | blog post | github (pre-trained weights)
컨브넷을 써서 자동 태깅 문제를 푸는 논문입니다. 7페이지 분량. 간단한 모델을 깃헙에 공개했으니 참고하세요.
- Text-based LSTM Networks for Automatic Music Composition
Keunwoo Choi, George Fazekas, Mark Sandler
1st Conference on Computer Simulation of Musical Creativity, Huddersfield, UK, 2016
arXiv | pdf
LSTM으로 드럼과 재즈 코드를 자동으로 생성해주는 논문입니다. (Single-column) 8페이지 분량.
- Auralisation of Deep Convolutional Neural Networks: Listening to Learned Features
Keunwoo Choi, Jeonghee Kim, George Fazekas, Mark Sandler
Late-Breaking Demo Session of 16th International Society of Music Information Retrieval (ISMIR) Conference, Malaga, Spain, 2015
paper | poster | demo | code
위의 "Explaining Deep Convolutional.., 2016"의 초안이 된 논문입니다.
- Understanding Music Playlists
Keunwoo Choi, George Fazekas, Mark Sandler
ML for Music Discovery Workshop at 32nd International Conference on Machine Learning, Lille, France, 2015
paper pdf | slides | arXiv link
음악 플레이리스트 데이터셋을 간단히 분석해보았습니다.
ETRI, during Mar 2011- June 2015
- A System for Korean Traditional Music Notation System : A Definition of XML File Format and An Editor Application
Keunwoo Choi, Tae-Jin Park, Yongju Lee, and Kyeongok Kang, Oct-2013, 135th AES Convention, Audio Engineering Society
pdf - A Texture Window Approach for Harmonic-Percussive Source Separation of Jazz Music
Keunwoo Choi, Sep-2013, 26th Joint IEEK Conference on Signal Processing, Institute of Electronics Engineers of Korea
pdf - Multichannel-to-Wave Field Synthesis Upmixing Technique Based on Sound Source Separation
Keunwoo Choi, Tae-Jin Park, Joengil Seo, and Kyeongok Kang, Sep-2013, 52nd AES Conference, Audio Engineering Society
pdf - A Mono-To-Stereo Upmixing Algorithm Based on the Harmonic-Percussive Separation
Keunwoo Choi, Jun-2013, KSOBE Conference, Korean Society Of Broadcast Engineers
pdf - Harmonic and Percussive Separation Based on NMF and Tonality Mask
Keunwoo Choi, Sang Bae Chon, and Kyeong-Ok Kang, Dec-2012, Journal of ETRI Letter
pdf | ETRI Journal link
- Automatic Volume Control for Mobile Devices
Keunwoo Choi, Seokjin Lee, Mingu Lee, Sang Bae Chon, Hoon Heo, and Koeng-Mo Sung, Nov-2011, Acoustic Society of Korea Conference
pdf
- Subjective Assessment of Localization Quality in a WFS Reproduction system using Front Loudspeaker array
Hyunjoo Chung, Keunwoo Choi, and Koeng-Mo Sung, May-2011, Acoustic Society of Korea Conference
pdf (in Korean)
Master's degree course in SNU, during Mar. 2009- Feb. 2011
- A NMF-Based Mono-To-Stereo Blind Upmix
Keunwoo Choi, Feb-2011, Master's Thesis at Seoul National University
pdf
- Analysis and Source Separation of Left, Right, and Center Channel for Upmix
Keunwoo Choi et al, Nov-2010, Acoustic Society of Korea Conference
pdf
- Mono-To-Stereo Blind Upmix Using Non-Negative Matrix Factorization and Decorrelator
Keunwoo Choi, Sang Bae Chon, Seokjin Lee, and Koeng-Mo Sung, July-2010, Journal of Acoustic Society of Korea
pdf (in Korean)
- Trajectory Sampling for computationally Efficient Reproduction of Moving sound Sources
Nara Hahn, Keunwoo Choi, Hyunjoo Chung, and Koeng-Mo Sung, May-2010, 128th AES Convention, Audio Engineering Society
pdf | AES link
