neurafly
Real audio in, real neurons firing, real time. A terminal visualizer wired to the actual FlyWire whole-brain connectome.
138,333 leaky integrate-and-fire neurons and 2,052,622 weighted synapses rendered as a living disk in braille.
Neurons
138,333
Synapses
2,052,622
Simulation tick
100 Hz
Connectome data
18 MB
Problem
Audio visualizers synthesize shapes from frequency data. The motion looks organic but carries no structure: bars, particles, and waveforms driven by the same FFT numbers. neurafly instead couples live system audio to a real biological wiring diagram, so activity propagates, reverberates, and decays through genuine fly-brain topology.
Key decisions
Real connectome, not a model
Wiring comes from FlyWire release v783: 138,333 neurons and 2,052,622 weighted synapses. Inhibitory sign follows presynaptic neurotransmitter probabilities.
cava-style band processing
A 2048-point FFT folds into 24 logarithmic bands, each judged against its own rolling spectral floor, with equal-loudness EQ, auto-gain, and a peakiness gate.
Fixed-tick pipeline
Four modules (audio, sim, figure, canvas) share a fixed 100 Hz simulation tick, so each band injects current into its own neuron subset deterministically.
Terminal-native rendering
A float phosphor buffer at 2x4 sub-cell resolution folds into braille, and colors come exclusively from ANSI 30-37 so the disk adopts the user theme.
Tradeoffs
- Linux only (PipeWire/PulseAudio) and a terminal with Unicode braille support
- The 18 MB connectome is excluded from the published crate, which falls back to a synthetic network
- Regenerating the network needs the Zenodo feather file and the Python preprocessing tooling