Audio Reactive LED Strips Are Diabolically Hard

April 8, 2026
Close-up of hands repairing a circuit board with soldering tools at a workspace.
Photo by www.kaboompics.com on Pexels

A ten-year rabbit hole

It has been reported that a project begun in 2016 to make LED strips react to music in real time has turned into something much bigger than the author expected — ten years, 2.8k GitHub stars, Hackaday write-ups, and it has allegedly been wired into nightclubs, Alexa setups, and beginner electronics benches. The creator says he’s still not satisfied. Why? Because what looks simple — blink lights to beat — is really a squeeze job: you’re trying to compress the nuance of sound into a tiny line of pixels. Fun, maddening, and oddly human.

From volume tweaks to addressable pixels

Early work was humble: non-addressable strips, three color channels, time-domain tricks. Read 10–50 ms of audio, low-pass it, map intensity to brightness. One color for fast transients, another for slow decay; exponential smoothing for adaptive gain. That gets you something in an afternoon. It also gets boring fast. So he moved to WS2812 addressable LEDs to buy more spatial resolution — more pixels, more problems.

The FFT that didn’t solve it

The next obvious idea was frequency-domain processing: take a short chunk of audio, run an FFT, map bins to pixels. He had 144 LEDs and thought, why not 144 bins? It kind of worked. Mostly it didn’t. Energy clusters into a handful of bins, leaving most LEDs dark. Cropping and hacks helped a little, but the naive FFT looks great on a monitor and lousy on a strip. The lesson: more pixels don’t automatically mean more information unless you choose the right features and perceptual mapping.

Pixel poverty — the real enemy

He names the problem plainly: Pixel Poverty (aka Feature Famine or Compression Curse). On a 144-pixel strip you can’t afford to waste space; each pixel must carry perceptually meaningful information. That insight — more signal design than brute-force math — is the emotional crux. It’s also why the project keeps evolving: different filters, mel-scale tricks, smarter mappings. Ten years in and it’s still a work in progress. Will it ever feel “done”? Probably not. And that’s part of the charm.

Sources: scottlawsonbc.com, Hacker News