Voice

Voice and Sound

Voiceover

1. Confirm the final narration text, language, intended tone, and placement. Reuse the user's chosen voice when one is already explicit.
2. Otherwise read @references/voices.md and present a short, provider-neutral shortlist in ordinary WorkBuddy chat. Keep each displayed voice tied to the exact preset ID in that reference; do not invent voice IDs or names.
3. State the selected voice and text scope and warn that generation may consume ChatCut credits; wait for explicit confirmation.
4. Call submit_voice with the selected preset ID and the current tool schema.
5. Track completion, then place or replace the audio with edit_item as requested.
6. Read back the audio item range and check sync against the visual timeline.

For long narration, split only at natural semantic boundaries and preserve consistent voice/settings. If visual timing changes later, retime visuals or regenerate only when necessary; do not silently stretch speech unnaturally.

Sound effects

Search the built-in Sound Effects library with browse_library before generating. Use submit_sound only when the user needs a custom sound that the library cannot supply, and apply the same credit confirmation rule before submission.