In live captions, ordinary sentences are rarely the problem. The mistakes usually come from names of people, organizations and products, and from technical terms. These words can be new, rare, or sound like common words. HearGlobe glossaries let you tell the system before the event how these terms are written and how to translate them.
What a glossary contains
A glossary is a list of terms. Each term has:
- Term: the correct original spelling, such as a person's name or a product name.
- Aliases: separated with
|; add common mishearings here. - Translations: the spelling you want in each language. Languages left empty are translated automatically.
- Note: for your own reference; not used in translation.
Terms are also added automatically to the reference used for recognition. So a glossary helps twice: the term is heard correctly, and its translation stays consistent.
Create a glossary
- In the sidebar, open "Glossaries" and click "Create glossary". Give it a clear name, such as "Medical terms" or "Day 1 morning".
- Click "New term" to add terms one by one, or use "Import CSV" to add many at once.
- CSV columns are
term,aliases,note, and other columns are language codes (for exampleen,ja). Export a file first to use as a template; a UTF-8 CSV saved from Excel imports directly. - When importing, "Merge" updates terms with the same name and keeps the rest; "Replace" clears the current terms first.
Add the glossary to an event
A new glossary has no effect until you add it in an event's "Glossaries" tab.
- One glossary can be added to many events. Chat links can use glossaries too.
- One event can use several glossaries. Higher up wins: for the same term, the glossary above is used.
- After you add a glossary, edits to its terms apply immediately. The speaker does not need to restart recording.
Switch glossaries mid-event with slots
When morning and afternoon have different speakers and topics, use "slots":
- In the event's "Glossaries" tab, click "New slot".
- Tick the glossaries each slot uses.
- When the session changes, click "Switch". If the speaker is recording, the change applies at the next pause.
Without slots, every added glossary is active.
Recognition context: give more background
Each glossary can also hold "Recognition context":
- Context groups: describe the field, the speakers and other background with a label and text. The event topic is always group 1.
- Reference text: paste the agenda, speaker bios or product notes. It weighs less than context groups and terms, and it is cut first when capacity runs out, so put important content first.
In the event's "Glossaries" tab, click "Preview" to see what will be used and what will be cut. If it is over capacity, move less important glossaries lower, or shorten the reference text.
Advanced rules: when the same mistake keeps happening
Usually not needed. Add rules in the glossary's "Advanced rules" only when recognition or translation keeps making the same mistake:
- Transcript corrections: before translation, fix misheard words. Source captions change too.
- Translation replacements: after translation, change words in the translation to the spelling you want. Traditional and Simplified Chinese are set separately.
Rules apply from top to bottom and are case-sensitive. "Whole word" only matters for languages with spaces, such as English.
Practical tips
- Start with the most important terms: speaker names, the host organization, the main products.
- If a term sounds like a common word, the common word may be heard as the term. Add the common word as its own term, with a translation.
- Write the field and the speakers in the event's "Topic and background". It works best together with glossaries.
- After the event, read the transcript and add any mistranslated terms to the glossary for next time.
See the glossary guide for step-by-step instructions.