The Line Spectrum Mode: The Axis Is Logarithmic Because the Bins Are

The catalogue describes Line Spectrum as A smooth log-frequency line with stable peak behavior. Read the drawing code on its own and that sentence looks wrong, which makes this the most interesting post in the set: the claim is true, and none of the evidence for it is in the file that draws the line.

What the renderer actually does

The draw code places sample number i at i divided by the last index, times the width. That is a strictly even spacing. It does not bend the axis, and at the default frequency scale the resampler that feeds it walks the band array evenly, interpolating linearly between neighbouring bands. If you audit the renderer alone, you will conclude the axis is linear in band number, and you will be right about the renderer.

Until October 2026 there was a second logarithmic curve, in the motion kit's element-band helper, which the bar, ray and ring engines use to give element number i its own slice of the spectrum. Bending an array that was already log spaced sent most of Radial's rays to the same few bass bins, so the helper now walks the array evenly as well. This mode never used it: it draws a polyline over an array, so it has no element index to curve.

Where the curve actually is

Two hops upstream. The per-frame audio snapshot fills its band array from the analyser's spectrum-band accessor, and that accessor does not hand back raw FFT bins. It builds 128 bands from 40 Hz to 14 kHz, each the average of the FFT bins whose centres fall inside it. A 2048-point FFT has a bin about every 23 Hz, so up to roughly 790 Hz each band is a single bin, the finest detail the analysis has; above that the bands are spaced logarithmically.

So the array this mode walks is already spaced by frequency: one bin per band through the bass and low mids, then by octave. Walking it at even x intervals therefore produces a logarithmic frequency axis across almost all of its width, which is what the catalogue claims. The axis is log because the BANDS are log, not because the drawing bends anything, and if you ever move this mode onto a raw linear FFT source the description stops being true without one character of the draw code changing. The Frequency scale control works upstream too: mel and linear respread the samples over the bands before the renderer sees them.

Resolution, range and level

Bins sets how many samples the line is built from, shipped here at 128 and capped at 256 inside the branch. Spectrum range then sets how much of the 40 Hz to 14 kHz axis the line spans, from the bass up. It used to default to 0.86, because the top of the old 20 Hz to 20 kHz array was near-silent air that barely moved. The axis now stops at 14 kHz and Auto-level lifts every band toward its own recent range, so the default is 1 and the right-hand end moves with the hi-hats instead of lying flat.

Smoothing lengthens the one smoothing stage the spectrum passes through, a filter across frames rather than across bins. It smooths the line in TIME: a high value makes the curve glide and a low one makes it chatter. It does not blur one frame's shape sideways.

Peak hold, made legible

This mode ships a peak hold of 420 milliseconds, and that value is worth a second look. The draw code reads it and the markers honour it, but the Spectrum engine declares no peak-hold control at all: the number arrives through this mode's own authored defaults and there is no slider for it in either editor. The hold you get here is the hold the mode was written with.

Each held peak is drawn as a small mark a few pixels wide above the curve, and the marks FADE across their hold rather than vanishing at the end of it. That is what makes the duration legible from a single frame: with a hard expiry, a short hold and a long one produce identical pictures and you could only tell them apart by staring at one marker until it disappeared.

Height, anchor and headroom

At the default bottom anchor the baseline sits at 94% of the frame height, which is nearly flush. Sample height is a fixed amplitude plus contributions from level, beat pulse and onset burst, and the whole thing is soft-limited against the distance from the baseline up to the top of the frame box. Soft-limited, again, rather than rescaled: dividing by the worst case would shrink ordinary levels to leave room for peaks that mostly never come, so the spectrum would read as permanently quiet.

When to reach for it

Anything that wants an honest readout rather than an effect: mix demos, plugin walkthroughs, mastering content, and the analytical half of a music channel. It is one row and one stroke, so it is also the cheapest thing in the engine to render at high resolution. For the same data with history behind it, read The Frequency Ridge Mode: Only the Front Row Is a Ridge; for the underlying transform, How FFT Turns Audio Into Visualizer Bands.

Open Line Spectrum in the gallery or go to the editor.