Back To Schedule

From Paper to Plugin

LLM’s and the Future of Audio Engineering Work

LLM's turn out to be surprisingly good at transforming academic papers, or even high-level descriptions and manuals, and producing working code. Transforming academic DSP papers into solid, efficient, runnable code AND a usable and inspiring product requires a set of intricate cross-domain skills. Many people in the ADC audience overlap with some of these - music making and product design, systems programming, DSP and signal processing - but few of us can fully cover all of these domains.

LLM's are useful to fill in some of these gaps - in this talk, I'll demonstrate practical strategies for implementing DSP algorithms from research papers using agentic coding. I'll cover prompting, testing, sound design and aesthetic refinement, and rapid prototyping. While LLM's cannot currently replace careful filter design or hand-optimized C++ code, they CAN help us quickly go from paper/sketch to usable prototype in a drastically shortened timeframe, and with a quality and rigor that can be difficult otherwise. We'll cover one hands-on example where we go from academic paper to tested and correct audio plugin over the course of the talk.

As LLM's improve, they will get closer to being able to cover the research-to-product pipeline more fully. I'll finish the talk by speculating about how we might see our roles as audio developers in a future where some of our specific and uncommon skills are accessible to anyone with a Claude account. I'll propose several perspectives to imagine the role and value we might have in a landscape where expertise and specialized skills are changing rapidly.