Idle Lines and Ambient Dialogue in Games

Idle and ambient dialogue is the dialogue nobody plans and everybody hears. The guard muttering on their fifth loop past you. The shopkeeper greeting you again. Two villagers talking about the harvest while you walk between them.

It usually gets written last and does more work than the lines that got written first. A player hears a set-piece line once. They hear the shop greeting forty times in an evening, and that repetition is most of what tells them whether a world is inhabited or staffed.

The part that changes how you plan it: AudioProducer renders the text you give it, and the same text renders the same way every time. There is no randomize button and no variation setting. Whatever variety your ambient layer has, you wrote it. That makes an ambient pass a writing budget rather than a settings pass, and everything below follows from it.

Short repeatable combat and alert lines, the ones usually called barks, are covered in how to voice NPCs in your indie game. This is the other half.

What counts as idle and ambient dialogue

Idle dialogue is what a character says while nothing is happening: a greeting on approach, a mutter to nobody, a work noise with words in it. Ambient dialogue is what the player overhears between other characters, or from around a corner, without being addressed.

Both are separate from ambient sound, the wind and the market din underneath, which is built from sound cues in your text and worked through in adding ambient sound to an audio story. Ambient dialogue is the words inside the din, generated like any other line: as text, spoken by a cast voice.

The practical difference from story dialogue is that the player is not paying attention when it plays, and hears it again later when they are.

The variation has to be written

Give the shopkeeper one greeting line and every visit plays that identical file. There is no per-line randomization on our side and no way to nudge a delivery slightly on a re-render, so three greetings means three written lines generated as three pieces of audio.

Budget the ambient layer in words then, the same way you budget the story dialogue. If you are working out what the total looks like, how many voice lines an indie game needs breaks the count into planning tiers and shows what the repeated lines do to it.

How many variants before a line stops looping

There is no number that works everywhere, and the number people ask for is usually attached to the wrong line. What matters is how often a specific line fires, not how many characters say it. A rough way to spend the budget:

• Count how many times a player plausibly triggers each line in an hour of play.
• Above about twenty an hour, the shop greeting and the door refusal, four or five versions is where most of the benefit lands.
• Under about five an hour, one is fine. A rare line heard twice reads as a callback rather than a loop.
• Everything in between gets two or three.

Most projects get this backwards and write six versions of a line that fires when you finish a quest. Rank by fire rate first, then write.

When you do write variants, change the length as well as the wording. The ear catches rhythm before words, and two lines with the same shape read as the same line even when every noun is different.

Write lines that survive being overheard

Ambient dialogue is heard from behind, at a walk, halfway through, while the player is reading a map. Lines that work on the page often disappear in that condition. What holds up:

• Put the informative word early. A line whose point arrives in the last two syllables is a line the player walks out of.
• Do not lean on what came before. Pronouns pointing at a subject the player never heard turn into noise.
• Keep them shorter than they look at your desk, because you are standing still and the player is not.
• Assume most players miss any given ambient line. If a piece of plot has to land, it does not go here.

The audio-only case is the extreme version, with no screen to carry the meaning a line drops. Voicing an audio-only game works through what changes when the ear is the whole interface.

Faction and role live in the vocabulary first

Casting a different voice per character is the obvious lever, and it is the second one. Before you get there, ask whether a guard, a merchant and a beggar are written differently enough that you could tell them apart from a transcript.

They differ in what they notice, in what they call money, in whether they address the player at all. A guard reports. A merchant sells or declines to. Someone with nothing to trade talks about the weather and the price of things.

A quick test: strip the speaker names off your ambient list, hand it to someone, and see whether they can sort the lines back into roles. If they cannot, a distinct voice will not fix it, it will just make the same line sound different.

Once the writing separates, casting makes the separation audible, and choosing AI voices for characters covers picking voices that stay distinct across a cast.

Where idle lines sit in the export

In AudioProducer the export unit is the chapter: one click renders a chapter to a single audio file, and each chapter downloads separately. For game audio that means one voice line, or one tight group that always plays together, becomes one chapter. Forty idle lines is forty chapters, so settle that structure before you paste anything in.

Check two project settings before you generate rather than after. Chapter Intro reads the chapter name before every chapter, which on a one-line-per-chapter project means every file opens by announcing its own name. Book Intro does the same once at the top, landing on whichever line happened to be first. Both are checkboxes under Edit Project. Name chapters after line identifiers rather than after what the line says, since chapter names become download names, and naming and organizing game voice line files has the scheme.

After the download the pipeline is yours. We export MP3 files and stop there. There is no engine plugin, no API for voice lines, and nothing hosted or distributed on our side, so triggering and randomizing between variants happen in your engine. So does anything about relative loudness: there is no per-line volume control, fader or ducking in the editor, and an ambient line cannot be placed under a scene at a lower level. What you control here is which lines exist and where they sit.

A reasonable first pass is one location: the shop greeting four ways, two overheard exchanges, one mutter, each its own chapter. Listen to the set back to back before scaling it to the rest of the map. To rough the layer in before it is written properly, placeholder voice lines for a game prototype covers the scratch pass and swapping it out later. The full guide index collects the rest of the game-dialogue and export guides.