Skip to main content

​
What You Can Influence

​
Tone and Personality

The wording in your flows and guidelines directly shapes how the assistant communicates. Be explicit about the tone you want:
  • Formal vs casual language
  • Level of empathy in responses
  • How much detail to provide

​
Response Content

The knowledge base and guidelines determine what information the assistant can share. Structure your content clearly to get consistent responses.

​
Behaviours with Limitations

​
Repetition

Repetition is an inherently tricky behaviour that is variable (probabilistic). There is a natural tension in the system between allowing the assistant flexibility to respond naturally and ensuring it follows the script structure. You can reduce repetition by:
  • Varying the wording in your flow steps
  • Avoiding overly rigid scripts that force similar responses
  • Using guidelines to discourage repeating the same phrases
However, repetition will still occur due to the probabilistic nature of language models trying to follow the Script.

​
Interrupt Handling

The assistant can detect when its message has been cut off, but it doesn’t inherently “know” it has been interrupted in the conversational sense. Adding guidelines to manage interrupt behaviour has limited effect because the assistant processes each turn based on the transcript, not the real-time audio event. Instead, control interruptions per step with the Interruptions setting on the step menu. Make a line Uninterruptible so it is always heard in full, use Uninterruptible turn so nothing the caller says derails a step that must finish, or Allow skip to drop a line the caller has already talked past. See Interruptions.

​
Checking the Line

If the caller says just “hello?” into a pause while the assistant is still working out its reply, the assistant immediately reassures them with a short line such as “I’m here, I just take a little to think sometimes.”, and later ones as brief as “Still here.”. Then it gives its real answer. The “hello” isn’t treated as a new question, so it doesn’t restart the reply, and background noise during the pause doesn’t restart it either. This doesn’t happen in the first ten seconds of a call, or on steps set to Uninterruptible turn. There is no setting for this behaviour. Speak to the Voxworks team for further custom configuration.

​
Silence

When the user is silent, the assistant will attempt to continue the conversation. See Silence Tolerance to configure how long the assistant waits before prompting the user.

​
User Accents

The assistant only knows what is directly transcribed — it cannot be prompted to “understand” or respond based on user accents. If the transcription is inaccurate due to accent, the assistant will respond to the transcribed text as-is. English speech is transcribed with a recogniser tuned for Australian English. Key Terms can help improve transcription accuracy for specific terms, set for the whole script or for an individual step.

​
Soft Speech

The assistant only knows what is directly transcribed. It cannot be prompted to handle soft or quiet speech — if the transcription service doesn’t capture the audio, the assistant didn’t “hear” anything. When the caller clearly spoke but nothing could be transcribed, the assistant asks them to repeat themselves (for example “Sorry, I didn’t catch that. Could you say it again for me?”) rather than waiting in silence.

​
Other Languages

Calls start in English. A script can switch a call into another language part-way through with a Code Step. From then on the assistant listens and speaks in that language, while your script, conditions, knowledge base and tools keep working in English behind the scenes. The call transcript shows each translated turn as it was heard or spoken, with the English underneath. For the list of supported languages and native voices, speak to the Voxworks team for further custom configuration.

​
Ongoing Improvements

Voxworks has ongoing programs to improve transcription quality, which will enhance the assistant’s ability to handle accents, soft speech, and other audio challenges over time.