More about HKUST
An Automated Pop Song Mashup System
The Hong Kong University of Science and Technology
Department of Computer Science and Engineering
PhD Thesis Defence
Title: "An Automated Pop Song Mashup System"
By
Mr. Xinyang WU
Abstract:
Music mashups transform familiar recordings into new works by recombining
elements of existing songs. The practice demands precise rhythmic alignment,
musically compatible material, and a polished final mix. Automated systems
have sought to replicate this craft, but they typically treat mashup creation
as a single monolithic problem. Timing is aligned only globally, so small
tempo errors accumulate over the course of a song. Compatibility between songs
is judged by fixed similarity rules or binary classifiers that do not reflect
the graded nature of listener preference. The final mixing stage receives the
least attention of all, although stems from different productions must be made
to sound as one.
This thesis argues that automated mashup generation becomes tractable when the
problem is decomposed into three stages and when quality is measured as graded
listener preference rather than a binary label. For temporal alignment, the
system locates the rhythmic backbone of each song and corrects timing at the
level of individual downbeats, so that the combined material stays locked to a
shared beat. For musical compatibility, large-scale listening tests establish
which combinations of separated stems listeners prefer, and a family of
vocal-conditioned models learns the correspondence between vocals and
accompaniment. For mix engineering, a restoration model corrects the tone and
level of the instrumental so that it sits naturally beneath the vocal.
Together, these contributions form a modular, interpretable pipeline that
edits and recombines real recordings rather than synthesizing new audio,
preserving the identity that defines a mashup and keeping the result enjoyable
to listen to.
Date: Tuesday, 25 August 2026
Time: 1:00pm - 3:00pm
Venue: Room 2132A
Lift 22
Chairman:
Committee Members: Prof. Andrew HORNER (Supervisor)
Dr. Arpit NARECHANIA
Prof. Gary CHAN
Prof. Ross MURCH (ECE)
Dr. Jiafeng LIU (Central Conservatory)