Disentangled Representation Learning of Music for Configurable Attribute-Based Retrieval

• Grant number: 25/02767-4
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: May 10, 2025
• End date: August 09, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Diego Furtado Silva
• Grantee: Lucas de Souza Brandão
• Supervisor: Dmitry Bogdanov Kondratiev
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Universitat Pompeu Fabra (UPF), Spain
• Associated to the scholarship: 24/13907-9 - Disentangled Music Representation Learning for Configurable Search by Features, BP.IC

Abstract

Music plays a fundamental role in human daily life. With the evolution of production and distribution technologies, especially streaming platforms, the music industry has grown significantly. To enhance user experience on these platforms, it is essential to develop tools that leverage both the characteristics of the music and the preferences of consumers. Recommendation systems are crucial in this context but still have limitations. Traditionally, they suggest songs similar to those the user already listens to but do not allow direct interaction, such as specifying which attributes to modify based on a reference track. This project aims to develop an interactive recommendation mechanism, enabling users to specify particular characteristics to discover new songs based on a reference track. To achieve this, we use disentangled representation learning, which maps a musical recording into a vector space where different subspaces correspond to distinct properties, such as danceability, aggressiveness, or specific emotions. This approach will allow users to adjust specific attributes of a query to find new music, offering a more personalized experience.