Intelligent Music Post-Production

ABSTRACT

This presentation introduces ongoing published research in Intelligent Music Post-Production at Sony, with a focus on applying large language models and agentic pipeline to music mixing. Music mixing and mastering is a highly subjective and technically complex process that requires expert decisions about audio effects, processing order, and parameter settings. We first explore how well LLMs understands and can interact with audio effects, providing a flexible and interpretable framework for controlling music production tools. Building on this, we present our ongoing direction toward agentic music post-production, where specialized agents analyze audio, plan mixing decisions, execute audio-processing tools, and iteratively refine the resulting mix. The presentation highlights both the potential of LLM-driven production workflows and remaining challenges, particularly fine-grained audio understanding and reliable parameter tuning.

BIOGRAPHY

Junghyun (Tony) Koo is a Research Scientist at Sony, working on artificial intelligence for music and audio. His research focuses on intelligent music production, controllable generative models, and protective AI, with particular interests in music mixing and post-production, audio effects modeling, and training data attribution. He received his Ph.D. in Intelligence and Information from Seoul National University, where his research focused on deep learning approaches for music production and audio effects.