Interactive Dialogue and Music
When dialogue speaks, music yields; state changes must land on seams you designed.
Related
Dialogue comes first
While a player listens to a line, the music has one job: stay out of the way. Speech is understood in roughly the one to four kilohertz region, which is also where music is most attention-grabbing, so the two collide and the words blur. The fix is not to turn the whole cue down but to clear the midrange first and then decide how much overall reduction is needed. Dialogue first means the music may sound worse; the line must stay clear.
Ducking is not a volume knob
Sidechain compression can pull music down when dialogue appears, but depth, attack and release all need setting. Too little does nothing, too much sounds like someone turning a knob, and a fast release pumps. A common approach ducks six to nine decibels, attacks within tens of milliseconds and releases over half a second to a second, and touches only a midrange band so the low end and air survive. Narration, shouted lines and radio chatter need separate settings; radio is narrow-band, so the music can stay higher.
States
Levels and timings follow the project; the job and the transition should be fixed first.
| State | Music | Dialogue | Transition |
|---|---|---|---|
| Exploration | Ambience plus sparse motif | None | Fade in over two to four seconds |
| Conversation | Low end and harmonic pad only | Primary | Duck six to nine decibels within three to five hundred milliseconds |
| Combat | Rhythmic layer plus theme | Shouts | Cut on a bar line, keep one layer running |
| Cinematic | Full cue | Primary | Prerendered, follows timecode |
| Menu | Loop without percussion | None | Crossfade |
The seam
Two things break a state change: a mismatched beat and a tail that bleeds into the next state. Every state shares tempo and key, and the switch lands on a bar line; if that is impossible, keep one continuous layer running across the join. Leave half a beat before dialogue starts so the player can react, and do not push the music back to full the instant the line ends; wait a second or two. Use a prerendered cue for cinematics and follow timecode instead of letting a real-time system improvise.
Common failures
Put ducking on the dialogue bus and sound effects will trigger it too, so the music jumps during combat. Generated music dropped straight into an engine often lacks clean stems, so there is no way to keep just one layer. In dialogue-heavy scenes, no music survives the mix, so write a sparse bed instead of a full cue. And never use music to hide artefacts in compressed speech; that only moves the problem into the player's headphones.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- 游戏音频引擎文档中关于对白总线、ducking 与状态切换的说明。
- 影视混音教材中关于对白优先与中频让位的通行做法。
- 交互音乐设计的公开访谈，讨论分层与实时切换的边界。