About narration placement

Audio description has one hard constraint: the description of a scene has to be heard during that scene, and the scene is only as long as it is. Everything our tool does with placement follows from that.

The gaps decide almost everything

Before placing anything, our tool finds the quiet stretches in the original audio — the moments where nobody is speaking. That’s where description belongs. Narration that fits inside those gaps is heard without covering a word of dialogue, and the Narration tab shows you how much room the source actually offers.

Most lines fit. The interesting question is what happens to the ones that don’t.

Three ways to make a line fit, in order

Speed it up. A line that’s only slightly too long can be delivered faster. With Speed lines up to fit the gaps turned on, our tool goes up to 2× before falling back to one of the options below — a marginally quicker sentence is usually less disruptive than either alternative. Past 2×, speed stops buying comprehension, which is why it isn’t tried further. The setting is optional because faster delivery is itself a cost for some listeners.

Let it overlap. The video keeps running and the description plays over the original audio, with that audio lowered so the description stays intelligible. The exported file is exactly as long as the source, which matters if it has to drop into an existing broadcast slot, a playlist, or a synchronised transcript. The cost is that something in the original gets talked over.

Pause the video. The frame freezes, the line plays in full, and the video continues. Nothing is covered and nothing is cut — but the exported file is now longer than the source, and the viewer experiences a stutter that wasn’t in the original work.

Those are the only three things an export can do, and no setting avoids the choice between them. That’s why the control is phrased as what to do when a line doesn’t fit rather than as a quality setting.

A shared player has a fourth response available — cutting a long description short — precisely because playback there is live rather than rendered into a file. Nothing has to be decided in advance for everyone.

Why lowering the original audio is separate from volume

Lowering the source isn’t “narration volume”. It’s how far the original audio drops while narration is speaking — a relationship between two sounds, not a level. A shallow reduction leaves the description competing with dialogue. A full one, at 100%, silences the music, the room tone, and the door slam the description was about. The default sits between the two and leaves the original audible underneath.

Why viewers of a shared link decide for themselves

An exported file bakes in one answer for everyone: one detail level, one delivery method, one policy for long descriptions. That’s correct for a file you hand to a broadcaster, and wrong for a link you hand to an audience.

So a shared player ships the descriptions as data instead. A viewer who uses a screen reader has their own software read them. A viewer who wants less detail moves a slider rather than waiting for a different export. A viewer who can’t stand a description being cut off chooses “pause the video” for themselves, while the person beside them chooses otherwise.

Some of what you set does carry over as a starting point — the detail level and how far the original audio drops. Others don’t: the player decides for itself what to do with a long description, starting at “cut it short”, because that’s the viewer’s call to make and not yours.

See How to share a video and Shared player controls.

Why lines you adjust are left alone

Once you adjust a line by hand, our tool stops re-planning it. Project-level placement changes flow around your pinned lines instead of overwriting them — otherwise every settings change would silently undo your work, and you’d have no way to hold one line in place while tuning the rest.