game.staging.lines

The Keeper's script — every word she says in First Light: Lucina, and the recording of each beside it.

One speaker. A woman past sixty, dry and unhurried, heard over the speaking tube from the watch room below. The apprentice never answers, which is why there is one voice and no cast: the game is one side of a conversation.

local Lines = require("game.staging.lines")

local l = Lines.get("opening_a")
l.text                      -- what goes on screen
l.speaker                   -- "The Keeper"
Lines.seconds("opening_a")  -- how long it is heard for, read off the clip
Lines.clip("opening_a")     -- the .soundClip, or nil if it never landed

The text is the asset; the audio is a performance of it

Every line is written here first. A line whose clip never generated, or whose player has the sound off, or who is in a noisy room, is a quiet line rather than a lost one — the words go on screen either way, and Lines.seconds falls back to a reading estimate (2.6 words a second plus a breath) so a beat built without the recordings is still the right length and still ends.

The script

The recordings

game/staging/voice/keeper_<id>.soundClip, one per line, generated once with audio_gen.speech and cached as an asset so a hot reload never re-buys the script. They are resolved one literal at a time rather than by a computed identity: a computed reference cannot be pinned, and a world that pulled this module in would arrive with the words and none of the voice.

Lines.info(id) reads a clip's sampleRate / channels / frames / durationMs off the ZAUD header without decoding it, and Lines.report() is the whole script with whether each line is voiced and how long it runs — which is how "is the voice actually there" is answered without listening:

for _, row in Lines.report() do
    print(row.id, row.voiced, string.format("%.2fs", row.seconds))
end