Publications / Talks

Exciting Trends in Music Technology with Deep Learning

NUS Hackers Talk 2024

Summarizes recent applications of deep learning in the music technology industry, uncovering use cases such as music generation, source separation, voice conversion, cover song detection, etc.
NUS Hackers is a student-run open-source community club in NUS, which hosts the biggest student hackathon (Hack&Roll) annually in Singapore.

Parameter Inference of Music Synthesizers using Deep Learning

Audio Developer Conference (ADC) 2022

An introductory talk that covers how synthesizer parameter inference facilitates the sound design process, and gives an overview of the recent works that use deep learning to perform parameter inference, as well as the challenges that were faced in solving this task.

Semi-supervised music emotion recognition using noisy student training and harmonic pitch class profiles

MediaEval 2021, Emotions and Themes in Music Challenge

An experiment on leveraging semi-supervised learning on music emotion recognition with noisy student training. Ranked 6th on the challenge.

Music FaderNets: Controllable Music Generation Based On High-Level Features via Low-Level Feature Modelling

Hao Hao Tan, Dorien Herremans.
International Society for Music Information Retrieval (ISMIR) Conference 2020.

Using “faders” (latent regularization & disentanglement) to control low-level musical features and “presets” (GM-VAE) to capture the relationship between the “faders” and the abstract, high-level feature for controllable music generation.

Generative Modelling for Controllable Audio Synthesis of Expressive Piano Performance

Hao Hao Tan, Yin-Jyun Luo, Dorien Herremans.
Machine Learning for Media Discovery (ML4MD) Workshop, ICML 2020.

Introducing GM-VAEs on temporal piano performance modelling, hence allowing fine-grained controllability on performance style features (articulation & dynamics) for audio synthesis.

ChordAL: A Chord-Based Approach for Music Generation using Bi-LSTMs

Hao Hao Tan
Creative Submission Extended Abstract, International Conference of Computational Creativity (ICCC) 2019.

A two-stage music generation pipeline with a chord generator and a chord-to-note generator as a seq2seq task. Chord embeddings unveil the pattern of the Circle-of-Fifths.