Documentation
Generation, wording and takes
Review the speech plan, regenerate the right blocks and keep useful recordings.
- Operators
- Developers
Generation brings together the text to speak, recordings, takes and history. The speech plan is a versioned division of the selected source into blocks and phrases. Preparing it does not generate audio.
Prepare and review
- Select the source or corrected/translated revision you intend to use.
- Prepare a speech plan. Let Pandrator apply its native segmentation and the selected model's length defaults.
- Check dialogue boundaries, names, voices and instructions. Use Speech plans to inspect the revision and mark it reviewed.
- Generate a short selection. Listen to beginnings, endings, pauses and dialogue before processing the rest.
Changing the source or segmentation requires a newly prepared plan. An old review applies to the old content signature, not to every later edit.
Written text and spoken wording
Double-click a block, press F2 in its text or use the pencil to edit its wording. The effective speech text uses the optimized version when one exists. A pronunciation spelling can therefore differ from the written script.
Keep captions readable: write Odra 1305 in the display text and, if needed, a pronunciation such as Odra thirteen oh five in the speech-only version. Editing the ordinary text without supplying an optimized replacement clears the old optimized wording. Review the result before regenerating.
Instructions, emotions and vocal cues depend on the model. A sigh generated by the speech model is different from a saved sound effect added to the mix. Multiple voices and emotional directions can be used independently.
Regenerate, recast or re-export?
| Change | Next action |
|---|---|
| Spoken words, voice, speech language or synthesis model | Generate replacement audio for the affected blocks. |
| Saved character voice | Preview Regenerate changed voices…, check the selection and apply. |
| Selected take | Choose the intended take/audio version, then assemble again. |
| Cover, title, bitrate, output format or soundtrack mix | Reassemble or export; those changes do not require new speech by themselves. |
| Caption spelling only | Check the written revision selected for export. |
Recasting keeps earlier audio selected until a replacement succeeds. A recast records Before and After choices. Its guarded undo restores the earlier take choices when still compatible; it keeps the cast settings and historical recordings.
Current settings or a historical run
A new MCP generation request uses current project settings by default. Set settings_source to matching_run to use a compatible run's frozen settings, or to run together with settings_source_run_id to select a particular run. A missing matching run is an error, not permission to fall back silently.
The run selected for playback/export is a separate choice from the settings source for new synthesis. Check both before comparing models. See settings and defaults and the full generation reference.