Resonant: speech models made of oscillators, not attention.
Most speech models are a stack of matrix multiplications wrapped around a spectrogram. Resonant asks a different question: what if the middle of the network were a physical system — a field of coupled oscillators, driven by the audio itself, whose synchronization does the computing? It is a hypothesis built to fail cleanly, with parameter-matched controls and verdicts written before the runs.
The idea
Speech is oscillation — vocal folds cycling, resonances ringing, syllables landing at 4–8 Hz. So the core here is a grid of Kuramoto oscillators on a torus. Audio enters as a driving force, the oscillators pull each other toward agreement, and the readout is simply where every oscillator sits on its circle and how fast it is turning.
Because the coupling is a circular convolution, one step costs a single FFT, and the phase field doubles as a fixed-size streaming memory — no cache that grows with the length of what you said. The physics of the configuration in the guide is roughly 2,000 numbers.
Oscillator based models could usher in a new paradigm for AI and computing, helping us better understand intelligence and the brain. Project Resonant, is an early experiment, but is showing promise. It is a step towards that future.
Learn more
How a machine hears a number
The interactive guide: pressure waves through Fourier, mel spectrograms, vocoders, and the oscillator core — with the whole pipeline running live in your browser on real spoken digits.
Read the guideThe repository
The code, the architecture, every experiment log, and the verdicts — including the ones that went against the hypothesis. Going public once it has had a clean-up pass.
Coming soon