Local subtitle recognition
- Category: Generate
What it is
Speech recognition on your computer, outputting sentence-level subtitles and word-level timestamps. Connect "Get video" or "Get audio" first.
This node grew out of the related Lab tools: wires in a blueprint replace the step-by-step clicks in the Lab, which makes it easy to reuse and to hand to an agent.
Related Lab tools
Inputs
| Name | Description |
|---|---|
| Audio input | Audio from upstream |
| Video input | Video from upstream |
| Timeline summary | Timeline position and related info |
Outputs
| Name | Description |
|---|---|
| Word-level subtitles | Word-level subtitles with timestamps |
| Sentence-level subtitles | Subtitles split by sentence |
Settings
| Name | Description |
|---|---|
| On-device recognition model | Set in the node panel |
| Output word-level timestamps | Set in the node panel |
| Maximum sentence length | Set in the node panel |
| Punctuation prediction | Set in the node panel |
| Disfluency handling | Set in the node panel |
| Hot words | Set in the node panel |
Typical downstream
Detect audio sensitive words, Smart cut rules, Filler clean, Import subtitles
Notes
- Before writing back to the editing app, make sure LmBox is connected to the matching app (DaVinci Resolve / Premiere / After Effects and so on).
Related recipes
- Sensitive-word beep write-back: see the recipes in the Blueprint lab overview
- Recognize speech, rough cut and place: see the recipes in the Blueprint lab overview
Agents / MCP
For agents or developers (optional)
- Once LmBox's open MCP is connected, the full ports and settings can be read from the catalog. This page is written for people, not around machine fields.
- Node type:
subtitle-recognition-local
← Back to the Blueprint lab overview · Lab features overview · Blueprint lab guide