7 min read
Sound design in film: what the audience notices and what it never hears
Most of a film's soundtrack is invisible work. The visible part only lands because the invisible part holds it up.
A film soundtrack is four layers: dialogue, effects, ambience and music. Sound design is responsible for the two in the middle, and for how much room the other two are given. Every scene is an argument about that balance.
Spotting: deciding where sound has a job
Watch the cut once without stopping, then again with notes. Mark the moments where sound must carry information, the moments where it should create tension, and the moments where it should get out of the way. Most beginners design the first two and skip the third, which is why their scenes feel exhausting.
Ambience as geography
Ambience tells the audience where they are and holds scenes together across cuts. One continuous bed under a dialogue scene hides edits, covers differences in room tone between takes, and makes ADR sit. Change the bed and you change the location, even if the picture stays the same.
Contrast beats loudness
An explosion is loud because of the two seconds of near silence before it. Cinema mixes work on dynamic range, not level. If everything is dense, nothing is big. Design the quiet as deliberately as the loud.
Perspective and point of view
Sound tells the audience whose scene it is. Close, dry and detailed puts them inside a character's head. Distant, reverberant and thin puts them across the street watching. Switching perspective on a cut is one of the cheapest and strongest tools in the box.
The technical spine
- Work at 48 kHz, 24 bit throughout, and check every imported file.
- Keep dialogue, effects, ambience and music on separate tracks from day one.
- Deliver stems so the mix can be revised without rebuilding anything.
- Check the whole soundtrack in mono before you sign off.
Files arriving from set as multichannel poly recordings should be split into named mono stems before editing, and extracted audio from reference video should be pulled at its native sample rate rather than resampled blindly. Both are one drag and drop on this site, processed locally.