What an audio waveform actually tells you

A waveform is not the sound. It is a picture of amplitude over time — and once you know what to look for, it is the fastest way to find the edit point you need.

Peaks versus loudness

The height of the wave is the peak sample level, not perceived loudness. A sparse drum hit can peak higher than a wall of guitars that sounds twice as loud. That is why mastering engineers talk about RMS or LUFS: they average over time, closer to how hearing works. A waveform view is a peak view — use it for structure, not for judging level.

What the shapes mean

• **Flat line:** digital silence. Trim it — leading silence is the most common reason a clip sounds late.
• **Thin, low wiggle:** room tone or breath. Keep a little; it sounds unnatural when cut to absolute zero.
• **Sudden tall spike:** a plosive, a door, a click. Easy to find and fix once you can see it.
• **Squared-off tops:** clipping. The signal hit the ceiling and the tips were cut off; distortion is baked in and cannot be removed by lowering the volume.
• **Regular blocks:** speech syllables or musical bars — handy for cutting on the beat.

How the picture is built

A three-minute stereo file at 44.1 kHz holds about eight million samples per channel, far more than the pixels available. Renderers bucket the samples: for each pixel column, scan the samples that fall inside it and keep the maximum absolute value. That is the peak envelope. Draw it mirrored around a centre line and you have the familiar shape.

const step = Math.floor(data.length / buckets);
for (let i = 0; i < buckets; i++) {
  let max = 0;
  for (let j = i * step; j < (i + 1) * step; j++) {
    max = Math.max(max, Math.abs(data[j]));
  }
  peaks[i] = max;
}

More buckets means more detail and a slower draw; a couple of thousand is plenty for a full-width image.

Using it in practice

Generate a wide waveform image of a recording before you edit. Mark the silences, spot the clipped sections, note where each section starts, then trim to those points. It turns "scrub around and guess" into a two-minute job — and the image itself is useful as a podcast thumbnail or a video overlay.