showtime
Docs / skills/showtime/references/crew/voice-director.md

Crew brief: voice director

Read this when you were dispatched as the voice director of a showtime job (after references/crew/rules.md).

Your job: the narration as heard: casting, pronunciation, pacing, the rendered voice-over and, for localized versions, the translated read. Reference: references/voice.md (all of it) and, for other languages, references/workflows/localize.md.

Task shapes#

A. Casting (studio sound phase)#

showtime voice list for the language, then three candidates that suit the tone. Each reads the picked hook: showtime voice say "<hook>" -v <id> -o <own dir>/casting/<id>.wav. Write casting.md: per voice the id, why it fits, what it risks (too bright, too slow), and a recommendation. The director masters the samples onto the board.

B. Voice-over (build)#

You own the project's voice/ folder for this phase.

  1. Pronunciation first: resolve the brand kit's pronunciation_candidates and any product, people or code names in the script. showtime voice ipa "<word>" shows what the engine will say; write fixes as inline hints or lexicon entries (voice.md, Pronunciation fixes) and keep them in lexicon.json in your folder.
  2. Render: showtime voice script <script> -o <project>/voice/, with --fit <seconds> when the length is fixed. Speed and pause_after per section, as the script format allows.
  3. Listen by numbers: read timeline.json (line lengths against the storyboard slots) and vo.words.json (no swallowed words, no long gaps). showtime voice master only for recorded voices; TTS output is already at level.

Done when: every line exists, every name is pronounced as the brand says (checked with voice ipa), the total length matches the target within half a second (or the gap is reported), and timeline.json is in voice/.

C. Localization#

Per language: translate the script (keep claim ids, keep on-screen numbers exact), pick a voice for that language, render with --fit to the same slots. Translations run 15 to 30 % longer: shorten the wording before speeding the voice past about 1.1. Write each language to <own dir>/<lang>/ (script, voice/, captions from the voice timings). Mark every machine translation needs native review in your result; never present it as reviewed.

Never#

  • Changing the script's meaning or facts: send wording problems back as notes.
  • Editing audio/, index.html or showtime.json. The director retimes the scenes from your timeline.json.

View or edit this page on GitHub