The Drop · August 16, 2026

Audio spectrum visualizer: bars, rings and waves that follow the vocal

Search for an audio spectrum visualizer and two different products come back. One draws the frequencies as moving geometry. The other burns a waveform and captions onto a podcast clip. This is the first kind, running as an overlay on a video that already cuts to the drums.

Photo via Unsplash

The control lives in the Overlays group and is called Vocal visualizer. It ships off, which is why plenty of people never find it. Four looks sit under it: spectrum bars along the bottom, a spectrum ring in the center, a waveform ribbon, and a glow pulse. One of them is not reading the spectrum at all.

What the overlay is actually reading

The label names its own band, 260 Hz to 3.4 kHz, and that is a deliberate slice out of the middle of the spectrum rather than a rounded description of the whole thing.

The analyser runs at an FFT size of 512. A Web Audio analyser hands back half that many frequency bins, so 256 numbers per frame, spread linearly from zero up to half the sample rate. On a 44.1 kHz file that puts roughly 86 Hz in each bin, and the overlay reads bins 3 through 40: just above 250 Hz to around 3.4 kHz.

That band is where a voice sits, along with the body of a snare and the top of a synth lead. What it skips matters more. Bins 1 and 2 are the sub and the kick, and if they were included they would tower over everything else on every frame, so the shape you saw would just be the drum pattern again. Cutting the low end out is the entire reason the overlay looks like it is tracking the vocal rather than the beat.

Each bin comes back as a whole number from 0 to 255, a decibel reading scaled into a byte rather than a volume in any absolute sense. A loud master fills the frame, a quiet one leaves the bars short. Nothing normalises that for you.

Spectrum bars and the spectrum ring

These two are the same 36 readings drawn two ways.

Bars lays them along the bottom edge, one bar per bin, edge to edge. The tallest a bar gets is 28 percent of the frame height, so the overlay never climbs into the middle where your text and images are. Anything under about 4 percent of full scale is skipped entirely, so quiet bins leave real gaps instead of a permanently lit row of stubs.

Ring arranges the same numbers around a circle in the middle of the frame, each one a spoke pointing outward from a radius of about 17 percent of the frame height. The whole ring turns, one rotation every 57 seconds or so, slow enough to read as drift rather than spin.

Both cycle through the current palette across the bars or spokes, so Hyper repeats red, cyan, yellow, magenta and white around the frame while Ice stays in blues. Mono and B&W carry one color, white, and the shape does all the work. The settings walkthrough covers the palettes.

The waveform ribbon is not a spectrum

Worth knowing before you pick it. The ribbon does not touch the frequency data. It reads the raw waveform, all 512 samples of the current buffer, and draws them straight across the frame. At 44.1 kHz that is about 12 milliseconds of audio on screen at any moment.

Two things follow from that. It is full band, so the kick is back in and the ribbon lurches on every one of them. And it is a shape, not a timeline. Nothing scrolls: you are looking at the last twelve milliseconds, redrawn sixty times a second, so it flexes in place instead of travelling left to right the way a waveform in an editor does. It also draws in the current flash color rather than cycling the palette, so it changes color on every hit.

Vocal glow, for when you want no geometry

The fourth option draws no lines. It averages the band into one number and paints a radial gradient out of the center in the current flash color, brightest in the middle, fading to nothing at the edges, with a radius running from about 15 percent of the frame height up to around 70 percent when the band is loud.

It is drawn additively, so it lights what is already on the frame rather than covering it. Over a single still image that reads as the picture breathing with the track, which is why it belongs with the calm setup where nothing cuts and nothing flashes.

Why the bars twitch

Next to the average spectrum widget this one looks nervous, and that comes down to a single number. A Web Audio analyser has a smoothing constant that averages each frame against the previous one. Browsers default it to 0.8, and that heavy smoothing is what makes most spectrum displays glide. This app runs it at 0.3, which is not a choice made for the overlay: low smoothing keeps drum onsets sharp, and the detection is the more important job. The overlay inherits it. On a busy mix the twitchiness reads as energy, on a sparse one it reads as jitter, and there is no slider to calm it down, so switch to the glow if it bothers you.

Where it sits in the frame

Each frame is built in a fixed order: the scene, then the flash effect, then this overlay, then flyers, then beat invert, then flicker, then the artifact texture.

The practical version of that: glitch slices, RGB split, shake and zoom punch do not touch the overlay. It locks itself to the frame, so while the scene shakes underneath, the bars stay put. That is a stable anchor at moderate settings and an obvious mismatch if the shake is pushed hard. Beat invert does flip it, flicker dims and brightens it, and grain or scanlines land on top.

If you want the spectrum to be the whole video, empty the text box and drop the image chance and the shape chance to zero. Black frames with the overlay on them, flashing on the beat, is the closest thing here to a plain spectrum clip.

The overlay itself never flashes, but the layers around it can. If the persistent flicker or the strobe effect is running underneath, the strobe guide has the thresholds. The short version is that WCAG 2.3.1 asks for no more than three flashes in any one second.

What this is not

The same search turns up tools that do all of the following. This one does not:

• No captions or subtitles. The only text on screen is the lines you type, and they change on the beat rather than on the words.
• Not a podcast audiogram. There is no scrolling waveform of the whole file and no static wave bar to sit under a headshot.
• Not a separate layer. The render is one flat 1080p WebM of the canvas with the overlay baked in, so it is the wrong tool for a spectrum on transparent video.
• Not a measuring instrument. Thirty six bins from an FFT sized for beat detection is a look, not an analysis. Do not mix against it.

What it does do is put a spectrum on top of a video that is already reacting to the drums, free, with no signup and no audio leaving your machine. Pick a look, hit RENDER, keep the tab in front so the frames keep drawing, and a WebM lands in your downloads. The export guide has the one line that converts it if a platform wants an MP4.