The ADC 2025 conference wrapped up earlier this week, and I wanted to share my thoughts. As always, The conference was a highly informative and engaging event that brought together experts in the field of audio development. Like last year, it featured a range of sessions, workshops, and networking opportunities that provided valuable insights into the latest trends and technologies in audio development (see my previous trip report for more details.

Day 1 — GPU Audio Workshop

I started the conference attending the GPU Audio workshop. The presenters demonstrated how to build an audio source separator inspired by the HS-TASNET work from L-Acoustics, using their SDK to run the model on the GPU. Unfortunately, the speakers were unable to attend in person and had to present over Zoom.

The session also felt a bit overly ambitious: while the technology is impressive, I think a simple “Hello World”–style introduction to their framework would have made for a more accessible and impactful workshop.

Afternoon — ADCx Sessions

The afternoon was devoted to the ADCx talks—rapid-fire 18-minute sessions meant to provide quick introductions to a variety of topics. A few standouts:

Josh Coalson (FLAC) gave an excellent history of the FLAC format. One line that stuck with me:

“If you have an idea, likely many people have the same idea. What helps you succeed is the way you execute it.”

David Su’s “50 Ways of Making a Sine” was a fun survey of generating a sine tone in different programming languages. It had a very cute ending—definitely worth watching once the recordings are released.

Christian Luther’s “Mind the Spike” was a great talk analyzing how performance issues can emerge, using Zipf’s law to understand performance statistics and identify where problems may hide.

Scheduling, Lock-Free Structures, and Memory Models

Rachel Susser gave a very clear and well-diagrammed talk on efficient task scheduling in a multithreaded audio engine. She compared several scheduling strategies—including work stealing—with helpful explanations of where each approach can struggle, especially when many threads wake simultaneously.

Dave Roland’s talk on lock-free queues explored multiple implementation techniques and the different performance characteristics they exhibit. Very practical and insightful.

Timur’s session on std::memory_order was probably the first time the C++ memory model really clicked for me. Understanding that memory orders describe an abstraction of 3 different aspects of memory complexity (the compiler reordering, the CPU microcode reordering, and the caching behavior) was a eye-opening. A really great talk.

Strongly Typed Units for Audio

Roth Michaels presented a compelling exploration of using strongly typed units in audio software. The inclusion of non-linear units like dB was especially interesting—I’m curious to see how that design direction evolves. The vocabulary-style approach to physical units could lead to safer and more expressive code, and it was great to see all examples rooted in audio contexts.

Keynotes

Both Keynotes were engaging, challenging, and memorable. Julian Storer’s ADC 2015 to 2035 was both a walk down memory lane and a chance to peek a little into the future. And Lu Wilson’s How Many Heads Do You Need was unlike anything I was expecting. I think this keynote challenged my understanding of how to be creative when coding, and I felt like I was watching a completely abstract way of making music with programming that I hadn’t even imagined existed.

Wrapping up

As always an excellent conference, a great venue, awesome technical ideas to learn from, and a wide variety of developers to talk with! I hope that I get a chance to meet you at this conference next year!