Resonate – Low-latency, high-resolution spectral analysis

Alexandre Francois built Resonate because the existing spectral analysis tools were too slow, too memory-hungry, and too computationally expensive for real-time audio work. So he wrote his own. The result is an algorithm that evaluates perceptually relevant spectral information from audio signals with low latency, low memory footprint, and minimal computational cost—no buffering required, just a handful of arithmetic operations per sample.

The academic community noticed. Resonate got the best paper award at last June's International Computer Music Conference (ICMC), one of the oldest and most respected venues in computer music research. Francois also presented at the Audio Developer Conference in Bristol last November, and he just returned from this year's ICMC where he extended the original work. That's three consecutive years presenting at major audio research conferences—there's no luck in that, there's only years of solid technical work under the hood.

The model itself builds on the phase vocoder, that classic spectral analysis technique from the 1980s, but Francois reimagined it using an Exponentially Weighted Moving Average (EWMA) approach—a resonator that accumulates signal contribution around its resonant frequency directly in the time domain. Each resonator is a complex number where the amplitude captures the contribution of the input signal component around a specific frequency, and the update happens at every single sample. No FFT windows, no buffering, no waiting for a frame to fill. For the standard audio frequency range of 20-20,000 Hz, that's a big deal.

This is pure research infrastructure—the kind of tool that other audio software, plugins, and instruments get built on top of. If you're building anything that needs real-time spectral analysis, this is the foundation worth checking.