Disentangled Music Representation Learning for Configurable Search by Features

• Grant number: 24/13907-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: October 01, 2024
• End date: January 02, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Diego Furtado Silva
• Grantee: Lucas de Souza Brandão
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated scholarship(s): 25/02767-4 - Disentangled Representation Learning of Music for Configurable Attribute-Based Retrieval, BE.EP.IC

Abstract

Music plays a fundamental role in human life. With the evolution of production and dissemination technologies, especially streaming platforms, the music market has grown significantly. To improve user experience on these platforms, it is essential to create tools that rely on both the characteristics of the music and its consumers. Recommender systems are crucial to this experience, but they still show limitations. Traditionally, they find music similar to what the user is used to listening to, but they do not allow the user to interact with the system, for example, specifying which characteristics they want to modify based on a reference. This project aims to develop an interactive recommendation mechanism, allowing users to indicate specific characteristics to find new music from a reference song. To do this, we will use disentangled representation learning, which represents a musical recording in a vector space where different subspaces correspond to different properties, such as danceability, aggressiveness or specific emotions. This will allow users to modify specific characteristics of a query to find new music, providing a more personalized experience.