An Automated Pop Song Mashup System

PhD Thesis Proposal Defence


Title: "An Automated Pop Song Mashup System"

by

Mr. Xinyang WU


Abstract:

Music mashups merge elements from different songs, transforming familiar 
tracks into new and captivating musical creations. Automated systems for 
generating mashups have emerged to simplify this creative process, yet 
replicating the nuance, creativity, and appeal of handcrafted mashups remains 
a significant challenge. Existing approaches typically address mashup creation 
as a monolithic problem, relying on rule-based harmonic similarity metrics or 
binary neural network classifiers that fail to capture the multi-dimensional 
nature of mashup quality.

This proposal presents a systematic approach to automated music mashup 
generation by decomposing the problem into three complementary stages. First, 
we develop precise temporal alignment methods, including main beat pattern 
extraction and local beat warping, to achieve downbeat-accurate 
synchronization between source tracks. Second, we investigate musical 
compatibility through large-scale listening tests across multiple 
stem-swapping strategies, develop a perceptual ranking model for predicting 
mashup quality, and explore adaptive instrumental rearrangement to improve 
compatibility between stems. Third, we address mix engineering challenges 
using diffusion models for frequency equalization, volume balancing, and 
spectral blending. Together, these contributions advance the state of 
automated mashup generation toward professionally polished results.


Date:                   Wednesday, 6 May 2026

Time:                   10:00am - 12:00noon

Venue:                  Room 2612A
                        Lift 31/32

Committee Members:      Prof. Andrew Horner (Supervisor)
                        Prof. Gary Chan (Chairperson)
                        Dr. Arpit Narechania