Industry

Bridging Research and Real-World Applications in MIR

Abstract

This keynote delves into the path from Music Information Retrieval (MIR) research to practical applications, highlighting source separation as a pivotal technology. With a background in machine learning and sound engineering, I will discuss the journey of developing and deploying state-of-the-art source separation models, addressing the unique challenges of making these models accurate, efficient, and accessible to a broad user base.

A core focus will be on optimizing source separation models for real-time, low-latency edge deployment. I’ll explore the technical intricacies of ensuring these models perform reliably on mobile and constrained devices, where resource limitations challenge both speed and fidelity. Topics will include the trade-offs and design considerations in model compression, latency reduction, and maintaining separation quality. This deep dive will illustrate how we bridge the gap between research and real-world use, making advanced audio separation tools available for musicians, producers, and listeners on-the-go.

Attendees will gain practical insights into the evolving landscape of MIR technology, as well as strategies for overcoming the complexities of real-time deployment, positioning source separation as a powerful, accessible tool in modern music technology.

Overview Program