explainer-audio-voicevox

Files (2)

• SKILL.md
• voicevox-kana.md · references/

SKILL.md (first lines)

# Prepare one VOICEVOX dialogue

Convert an accepted human-readable dialogue into its synthesis-specific
VOICEVOX representation. Own contextual readings, accent notation, pairing, and
kana validation. Do not change content, revise teaching structure, synthesize
audio, or coordinate human interaction.

## Required inputs

In a source workflow, apply [[explainer-content-workflow-base]] and read the
selected profile. Require exact paths and SHA-256 digests for the run request,
Run Manifest, human-readable dialogue, passing
[[explainer-audio-dialogue-review]] Artifact, immediate source report, planned
`voicevox-dialogue` output, and any selected `dialogue-pilot` checkpoint.

Validate that:

- the Manifest records the supplied readable dialogue and review as inputs to
  this exact VOICEVOX output;
- the review identifies the exact readable-dialogue digest and has
  `result: pass`;
- a selected human checkpoint identifies the same readable digest and is
  `approved`; and
- the output action is `create` or `replace`.

Return a structured missing/incompatible-input result rather than selecting a
latest Artifact or accepting a review for another dialogue revision.

For unrelated standalone input, require exact readable-dialogue and matching
`dialogue/<slug>.review.yaml` paths plus an explicit output choice. Do not
silently overwrite an existing synthesis script.

## Procedure

Read `references/voicevox-kana.md` completely. Resolve ambiguous readings from
the accepted readable dialogue and its exact source, never by blind character
conversion. Write `dialogue/<slug>.voicevox.txt` with the same style marker,
speaker sequence, turn count, line grouping, and propositional content as the
readable authority.

Run the local validation command documented in the reference. A missing engine,
invalid kana turn, pairing mismatch, or propositional change is a phase blocker;
do not produce or accept audio from that output.

Return the readable, review, and VOICEVOX paths and digests, pairing checks,
validation result, and any blocker. [[explainer-audio-narrate]] consumes the
validated `.voicevox.txt` file.

## Success criteria

- [ ] The exact readable dialogue has a matching passing review and any selected
  human approval.
- [ ] The VOICEVOX file has the same style marker, speakers, turns, line grouping,
  and propositions as the readable dialogue.
- [ ] Spoken fields contain no unresolved kanji, hiragana, Latin letters, Arabic
  numerals, URLs, or Markdown.
- [ ] Contextual readings and accent phrases follow
  `references/voicevox-kana.md`.
- [ ] Every complete turn passes the local engine's `/validate_kana` check.
- [ ] This phase writes only the planned `.voicevox.txt` Artifact.

From choplin/explainer-studio, MIT.

More from choplin/explainer-studio