Waveform

A waveform is a recording drawn as a picture: the audio is chopped into slices, each slice is reduced to one amplitude, and each amplitude becomes a bar, usually mirrored above and below a centre line so the shape reads as sound rather than as a chart. What it buys over a plain progress bar is that the track stops being uniform. Silence is visibly empty, a loud passage is visibly tall, and the reader can aim: the pause before the answer, the applause, the three seconds where somebody says the number. On a two minute voice message that is the difference between listening again and going straight to the part that matters.

It is a control as much as a picture, which is why it usually replaces the scrubber rather than sitting beside it. The played portion is filled and the rest left quiet, a playhead marks the current position, and clicking or dragging anywhere along it seeks. That gives the seeking gesture something a plain track cannot: the reader aims at a feature they can see instead of at a proportion they have to guess. Everything else stays where it was, so the transport controls beside it are still media controls, and in a video player the waveform is one more layer on the same timeline as the chapter marks and the buffered band.

The drawing has to be honest, and often is not. A bar is a summary of a slice, and whether that summary is the peak or the average changes the picture: peaks look punchy and hide how quiet the average is, averages look flat and lose the transients. Worse, plenty of messaging apps draw a decorative pattern that has nothing to do with the audio at all, which looks fine until someone tries to use it the way a waveform invites them to and finds the tall bar is not where the loud part was. If the bars are not measured from the sample data, the control is lying, and it is lying about the one thing it exists to say. Measure once when the file is uploaded, store the summary, and draw from it, so the same recording is drawn the same way every time.

Being a picture of sound, it needs the usual second channel. The elapsed and total times belong in text beside it, the seek needs to work from the keyboard in fixed steps, and the control needs a name and a spoken position, because a waveform tells a sighted reader where they are and tells everyone else nothing. None of that replaces a transcript, which is what actually makes speech searchable, and the two work well together: the transcript to find the sentence, the waveform to land on it.