Published:
My first research project, spanning my M.Sc. at Nile University and an internship at Sony Stuttgart. The goal: automatically separate the direct sound from the diffuse ambience in a stereo recording, to enable surround sound upmixing.
Published:
For my M.Sc. at NUS, I studied what makes song lyrics easy or hard to understand, and built systems to measure it automatically — motivated by a real application: recommending music for language learning.
Published:
My PhD project at Télécom Paris and Deezer, studying how listening context — activity, mood, device, time of day — shapes what music people want to hear, and building systems that learn to predict it automatically.
Short description of portfolio item number 1
Short description of portfolio item number 2
Published:
At Emobot, I led research on automatic speech emotion recognition — pushing accuracy significantly through synthetic data augmentation via emotion conversion, with direct impact on a real-time healthcare application.
Published in The 13th Sound and Music Computing Conference (SMC), 2016
This paper is about separting the primary and ambient sources from a sounds mixture to be used in surround sound upmixing. We propose a PCA-based approach to apply the separation
Published in The 19th International Society for Music Information Retreival Conference ISMIR, 2017
This paper is about estimating the intelligibility of the singing voice in a given song. We propose a set of acoustic features that are relevant for estimating the intelligibility. We also propose an approach for labeling songs with an intelligibility score accroding to human perception
Published in The IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018
This paper is about separting the primary and ambient sources from a sounds mixture to be used in surround sound upmixing. We propose a neural-network-based approach to apply the separation
Published in The 19th International Society for Music Information Retrieval Conference (ISMIR), 2018, 2018
This paper is about estimating the singability of a given song and the factors that make one song more singable than another. We propose a number of acoustic features to automatically estimate the singability of a song.
Published in The IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020
This paper is about auto-tagging music tracks with context-related tags. The paper also presents a dataset of ∼50k tracks labelled with 15 different contexts.
Published in The 2020 International Conference on Multimedia Retrieval, 2020
This paper is about a weighted loss function that accounts for the missing labels in the training set that is easily usable in fine-tuning pre-trained models
Published in 21st International Society for Music Information Retrieval Conference (ISMIR), 2020
This paper proposes a user-aware auto-tagging system for the contextual tags of music tracks
Published in 22nd International Society for Music Information Retrieval Conference (ISMIR), 2022
This paper proposes real-time predictions of the user listening context
Published in IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024, 2024
This paper proposes using synthetic data augmentation via emotion conversion to improve speech emotion recognition models.