Recognition vocabulary

General speech models often miss proper nouns and specialized terms — people's names, place names, ministry or theological terms, product names, and regional vocabulary. Adding these as custom vocabulary biases recognition toward them in real time, so they are far more likely to be transcribed correctly before translation happens.

1. 1Open Vocabulary from the dashboard (Recognition vocabulary section).
2. 2Add the terms your speakers use most — names, places, and recurring phrases.
3. 3Your vocabulary is saved for your organization and applied automatically to new sessions.

• Add terms exactly as they should appear, including capitalization for names.
• Short, distinctive phrases work better than long sentences.
• Revisit the list after events and add anything that was misheard.

You can pull candidates from sermon notes, an order of service, or talking points in two places: Import from script on the Vocabulary tab (organization or session), or Prep vocabulary on the speaker dashboard (this session only). Only terms you confirm are saved — the full script is not stored as a document.

1. 1On Vocabulary → Import from script, drop a PDF/TXT/Markdown file or paste text. Or open Prep vocabulary when creating a session or on the speaker page.
2. 2Choose Basic or Smart extract, review the chips, and save what matters.
3. 3You can also add terms manually anytime.
4. 4Start the mic when you are ready — confirmed terms bias recognition for that session (and organization terms apply to future sessions).

Basic extract runs on-device heuristics and is free on every plan. Smart extract uses AI to find hard-to-hear names and distinctive phrases more reliably. On Vocabulary → Import from script, Smart can also suggest translation idioms in the same run (one Smart credit).

• Starter includes 20 Smart extracts per month, Pro 100, Event 10. Free and over-allowance fall back to Basic.
• Files are still read in your browser; only the text is sent when you run Smart extract.
• Smart usually takes about 15–20 seconds. Review recognition chips and any idiom meanings before saving.
• Prefer Basic if you do not want script text sent to Azure OpenAI.

When you stop the mic, LABA LIVE can suggest terms from moments that had low recognition confidence (including lines held back from attendees). Add the ones you want to organization Vocabulary so the next session starts stronger. You can also open Improve vocabulary from an ended session on the dashboard.

• Review the suggested chips — skip anything that was noise or a one-off.
• Added terms go to organization Recognition vocabulary and apply to future sessions automatically.
• If a word was heard correctly but translated poorly, use Translation glossary instead.

The Vocabulary tab pays off when you use it as a loop: prep before the event, run a session, accept post-session suggestions (or note what was misheard), then go live again with better recognition. Organization vocabulary applies automatically on the next session — you do not need to rebuild the session.

1. 1Before the event, import or paste notes (Import from script or Prep vocabulary), run Basic or Smart extract, and confirm suggested terms.
2. 2During or right after a session, use Improve next time suggestions — or note words the live speech monitor got wrong.
3. 3Open Vocabulary → Recognition vocabulary for anything you want to keep permanently across sessions.
4. 4If the word was heard correctly but translated badly, add a Translation glossary entry instead — or use Smart extract with idioms enabled to draft both.
5. 5Before the next event, run a 2-minute rehearsal and confirm the terms you care about appear correctly in the monitor.

• Do add: people's names, church or venue names, book or song titles, specialized vocabulary your community uses every week.
• Do add: alternate spellings or spacing if the same term was misheard more than once (e.g. two entries for a hyphenated name).
• Skip: entire sentences, common words the model already knows, or one-off typos you do not expect again.
• Skip: expecting vocabulary to fix wrong source language — if the speaker mixes languages, set the dominant source language and add only the borrowed terms you need.