transcription

name	transcription
description	Use when a video/audio task needs caption editing, subtitle styling, transcript search, or enabling captions via Local Video Editor.

Transcription (Local Adaptation)

✅ ASR (speech-to-text) transcription is available via local Whisper. After importing media with audio, transcription runs automatically in the background. Use track_progress with target: "transcription" and assetIds to check status.

Local Limitations

• Whisper CLI required: The whisper CLI from openai-whisper must be installed. The first run downloads the model (≈1.5 GB for medium).
• Async transcription: After import_media, transcription runs asynchronously. The import returns immediately; use track_progress to wait for completion.
• manage_transcript fix persists corrections — word-level fixes are saved and reflected in subsequent reads.
• edit_captions supports template action for listing caption style presets by language.

Caption Editing Workflow

Caption editing starts from an existing transcript. When the user asks for captions, subtitles, or bilingual captions:

1. local-video-editor_read_project with view: "assets" to get the asset ID
2. local-video-editor_get_transcript to read transcript words for the asset
3. Use local-video-editor_edit_captions with the appropriate action

Available edit_captions actions

Action	Purpose
enable	Turn on captions for an asset
disable	Turn off captions for an asset
style	Apply custom style overrides (fontSize, color, etc.)
template	List available caption style presets by language (new in local)
language_mode	Set translation/bilingual mode with a language code

Caption style presets

For a general style request, first list presets:

{
  "action": "template"
}

This returns language-aware presets. Choose one or offer relevant returned presets for the user to choose from.

Use custom style / layout only when the user clearly requests a custom look or a specific adjustment. For adjustments, start from the closest preset and change only the requested properties.

Manage transcript

{
  "action": "fix",
  "assetId": "<asset-id>",
  "corrections": [
    { "wordIndex": 42, "correctedWord": "their" }
  ]
}

The fix action persists corrections — fixed words are saved and will appear corrected in subsequent transcript reads. This is useful for correcting ASR mistakes or speaker attribution on pre-existing transcripts.

For retry_transcription, the system will re-run local Whisper on the asset's source file.

Tool Reference

• local-video-editor_track_progress with target: "transcription" and assetIds — check transcription status per asset
• local-video-editor_find_transcript — for query-based transcript lookup
• local-video-editor_read_captions — read caption pages grouped from transcript words
• local-video-editor_edit_captions — caption display edits (enable, disable, style, template, language_mode)
• local-video-editor_manage_transcript — transcript repair (fix) or retry (retry_transcription)
• local-video-editor_clean_script — mechanical timeline playback cleanup

Talking-Head Editorial

When a transcript-ready request becomes an editorial talking-head edit, follow talking-head-guide. Use clean_script for mechanical cleanup, then use Script (read_script → edit timeline.md → apply_script) for semantic edits.