Skip to main content

New features

Scripts can switch the call into another language part way through. A Code Step can call set_language("es"), optionally with a voice ID so the switch uses a native voice. From the next step on, the assistant listens and speaks in that language while your script, conditions, knowledge base and tools keep working in English. set_language("en") switches back. The function returns true or false so the script can branch on it, and Arabic is among the supported languages. See Code Step functions. Speak to the Voxworks team for further custom configuration, including the list of language codes and native voices. Transcripts show both languages on non-English calls. On a call that ran in another language, each transcript bubble shows what was actually said first, with the English translation underneath in muted text. This applies to both the caller’s turns and the assistant’s. English calls look exactly as before. See Call details. Classifiers can be shared across a script or a flow. The step menu’s Classifiers item now has Script, Flow and Step groups. A script or flow classifier is written once and ticked on for each step that should run it, so editing it changes it everywhere it’s enabled. The three-per-step limit and the fixed Classifier 1/2/3 slots are gone. A new Shared Classifiers section in Settings lists every shared classifier and the steps it runs on. See Steps and Script settings. Make a step uninterruptible. A new Interruptions item in the step menu lets you set a step to Uninterruptible, so the caller can’t cut the line off. Anything they say during it is answered once the line has finished. Use it for anything that must be heard in full, such as an AI or recording disclosure. Normal stays the default. See Interruptions. Key Terms gain an Overfit option for keyword-shaped answers. The step menu’s Key Terms item is now a submenu with Edit… and a new Overfit switch. Overfit runs a second recogniser biased towards your key terms on that step and uses whichever transcript matches more of them. It’s useful for steps expecting menu choices, product codes or suburb names, and it needs key terms on the step or the script before it can be turned on. See Keywords.

Improvements

Knowledge Base entries no longer have a Mode setting. The Mode dropdown has been removed, so the Knowledge Base editor now has just Question and Response columns. New entries are answered in the assistant’s own words, based on your response. See Knowledge base. Voices are listed Australian-first, and the library has grown. Every voice list now shows Australian HQ voices first, then other Australian voices, then everything else by accent and name. Accent flags now show as proper country flags in every browser. The library also gained native-language voices, such as Arabic, Mandarin, Vietnamese and many European and Indian languages, to use with language switching. See Voice library. Your transcript flags always get a review, even on an unchanged script. Regenerate and the nightly review used to skip a script that hadn’t changed since its last review. Now, if the script has open flags, the review runs anyway and turns your flags into suggestions. If your flag note spells out the exact wording you want, the suggestion uses it word for word. A flag about something a recommendation can’t change is closed without a suggestion, and editing the flag’s note reopens it. See AI recommendations. Researched FAQs are saved to the Knowledge Base. When you create a script with AI from your website, the FAQs it finds are saved as question and answer pairs in the Knowledge Base, with duplicates merged. See Knowledge base. The assistant asks the caller to repeat instead of going quiet. When the caller clearly spoke but nothing could be transcribed, the assistant now says something like “Sorry, I didn’t catch that. Could you say it again for me?” rather than waiting for the silence timer. This happens automatically, so there is nothing to change in the app. Speak to the Voxworks team for further custom configuration. Speech recognition is tuned for Australian English. Live and batch transcription of English calls now use an Australian English recognition model. This happens automatically, so there is nothing to change in the app. Speak to the Voxworks team for further custom configuration. Hold music starts after the assistant’s last word. On a warm handoff, the hold music now starts about half a second after the assistant finishes speaking. See Live Handoff. Large contact books open faster. The Contacts table now loads a page at a time and only draws the rows on screen. See Contacts.