Skip to main content

​
What are Key Terms?

Key Terms are words and short phrases you give the speech recogniser so it listens for them properly. When a caller says one of these terms, it is far more likely to be transcribed correctly. Common use cases:
  • Brand names — “Voxworks”, “iPhone”, company names
  • Product names — Model numbers, product lines
  • Industry terms — Technical jargon, acronyms
  • Names — Common names that might be transcribed incorrectly
Key Terms do not change how the assistant responds — they improve what the assistant hears.

​
Where to set them

There are two places, and they work together: Both use the same input: type a term and press Enter to add it.

​
Per-step Key Terms

Open a step’s menu and choose Key Terms → Edit… to add terms that only matter on that step. Per-step Key Terms are available on voice steps. The panel explains it on screen: “Words and short phrases the speech recogniser should listen for while [step] is running. Use them for names, products, and jargon it usually mishears.” Add the terms under Terms, then choose Save key terms (or Cancel to discard). This is the right place for the terms that only come up in one part of the call — a list of suburbs on the address step, or product names on the step that asks which product the caller has. The step menu shows a count once a step has terms. The panel warns you when a term is longer than three words, and when one step carries so many terms that the lower-value ones are likely to be trimmed.

​
Overfit

The Key Terms submenu also has an Overfit option. Its tooltip reads: “Runs a second keyword-biased recogniser on this step and uses whichever transcript matches more key terms. For steps expecting keyword-shaped answers; needs key terms to be useful.” With Overfit on, the caller’s answer on that step is transcribed twice: once normally, and once by a recogniser that leans much harder towards your key terms. The transcript that matches more key terms is the one the assistant reads. Once it is on, the menu shows ”· overfit” after the term count. Use it on steps where the answer should be one of a known set of words:
  • choosing from a menu of options (“sales, support or billing”)
  • product codes or model names
  • a list of suburbs or branch names
Leave it off for open questions. A recogniser biased towards your key terms can pull an ordinary answer towards a key term that sounds similar. Overfit needs key terms to work with. It stays greyed out until the step or the script has at least one, with the tooltip “Add key terms to this step, or to the script in Settings, before turning overfit on — without any it has nothing to bias towards.”

​
How the list is built

Every source of terms is merged into a single list for the call:
  • Script-level Key Terms
  • Per-step Key Terms
  • Words drawn from your Intent Pathways and Knowledge Base
  • Terms Voxworks works out from your script automatically
  • Your contact’s first and last name, their company, and your business name
  • Anything set mid-call by the Keywords Set tool
The merged list is ranked by where each term came from — terms set by the Keywords Set tool first, then the step’s own Key Terms, then pathway, handoff and script terms, with automatically found terms last. Duplicates are removed and the list is capped at 70 terms, so the lowest-priority terms are dropped first. The list is rebuilt as each step starts, so a step with many terms of its own crowds out the script-wide ones while it runs — which is why the step panel warns you when one step claims too much of the list. The same list is applied to live listening as each step starts, and to the transcript afterwards, so what the assistant heard and what you read back match.

​
Changing terms mid-call

The Keywords Set tool sets a list of terms while the call is running — useful when the words you need depend on something you have just looked up, such as the product names on the caller’s account. Each use replaces the terms set by the previous Keywords Set (it does not remove your script or step Key Terms), and its terms take top priority in the call’s list. It takes effect from the next speaking step, so place it immediately before the step that needs those words heard correctly. See Keywords | Set for the tool’s own configuration.

​
Examples

A law firm might add:
A tech company might add:
A healthcare provider might add:

​
Best Practices

  1. Use single words where possible — Better performance is achieved with single words. For example, add “Power” and “Attorney” as separate terms rather than “Power of Attorney”
  2. Keep each term to 1-3 words — Longer terms can be dropped
  3. Avoid common words — Don’t include words like “the”, “and”, “is”, or other everyday terms
  4. Keep the list focused — The list is capped at 70 terms across the whole script, so spend that budget on words that are genuinely misheard
  5. Put step-specific words on the step — Terms that only matter in one part of the call belong in that step’s Key Terms
  6. Add real words only — Enter terms as they are correctly spelled. Adding common mishearings as key terms makes the recogniser more likely to hear the wrong word
  7. Match your desired capitalisation — Terms are transcribed as entered, so ensure capitalisation matches how you want them to appear
  8. Test and refine — Review call recordings and transcripts to identify words that need adding

​
Limitations

Key Terms improve transcription but cannot guarantee perfect accuracy. Factors like audio quality, background noise, and speaker clarity still affect results. For guidance on handling of accents and soft speech, see Assistant Behaviour.